National Medical Care Utilization and Expenditure Survey High-Volume and Low-Volume Users aL of Health Services United States, 1980 Series C, Analytical Report No. 2 li Health Care Financing Public Health Service K) SERVICE 7 | Td sy, | . Administration s « National Center for Office of Research | Z Health Statistics and Demonstrations z %, National Medical Care Utilization and Expenditure Survey The National Medical Care Utilization and Expenditure Survey (NMCUES) is a unique source of detailed national estimates on the utilization of and expenditures for various types of medical care. NMCUES is designed to be directly responsive to the continuing need for statistical information on health care expenditures associated with health services utilization for the entire U.S. population. NMCUES will produce comparable estimates over time for evaluation of the impact of legislation and programs on health status, costs, utilization, and illness-related behavior in the medical care delivery system. In addition to national estimates for the civilian noninstitutionalized population, it will also provide separate estimates for the Medicaid-eligible populations in four States. The first cycle of NMCUES, which covers calendar year 1980, was designed and conducted as a collaborative effort between the National Center for Health Statistics, Public Health Service, and the Office of Research and Demonstra- tions, Health Care Financing Administration. Data were ob- tained from three survey components. The first was a national household survey and the second was a survey of Medicaid enrollees in four States (California, Michigan, Texas, and New York). Both of these components involved five interviews over a period of 15 months to obtain information on medical care utilization and expenditures and other health-related infor- mation. The third component was an administrative records survey that verified the eligibility status of respondents for the Medicare and Medicaid programs and supplemented the household data with claims data for the Medicare and Medicaid populations. Data collection was accomplished by Research Triangle Institute, Research Triangle Park, N.C., and its subcontrac- tors, the National Opinion Research Center of the University of Chicago, Ill., and SysteMetrics, Inc., Berkeley, Calif., under Contract No. 233-79-2032. Co-Project Officers for the Survey were Robert R. Fuchsberg of the National Center for Health Statistics (NCHS) and Allen Dobson of the Health Care Financing Administration (HCFA). Robert A. Wright of NCHS and Larry Corder of HCFA also had major responsibilities. Daniel G. Horvitz of Research Triangle Institute was the Project Director primar- ily responsible for data collection, along with Associate Project Directors Esther Fleishman of the National Opinion Research Center, Robert H. Thornton of Research Triangle Institute, and James S. Lubalin of Systemetrics, Inc. Barbara Moser of Research Triangle Institute was primarily responsible for data processing. For sale by the Superintendent of Documents, U.S. Government Printing Office, Washington, D.C. 20402 High-Volume and Low-Volume Users of Health Services United States, 1980 Series C, Analytical Report No. 2 U.S. DEPARTMENT OF HEALTH AND HUMAN SERVICES Published by Public Health Service National Center for Health Statistics November 1985 Copyright Information All material appearing in this report is in the public domain and may be reproduced or copied without permission; citation as to source, however, is appreciated. Suggested Citation Berki, S. E., Lepkowski, J. N., Wyszewianski, L., et al.: High-vol- ume and low-volume users of health services, United States, 1980. National Medical Care Utilization and Expenditure Survey. Series C, Analytical Report No. 2. DHHS Pub. No. 86—20402. National Center for Health Statistics, Public Health Service. Washington. U.S. Government Printing Office, Nov. 1985. Library of Congress Cataloging in Publication Data High-volume and low-volume users of health services. (Series C, Analytical report; no. 2) (DHHS publication no. 86—-20402) Written by: Sylvester E. Berki and others. Bibliography: p. 1. Medical Care— United States— Utilization— Statistics. 2. Public health—United States—Statistics. 3. Health surveys— United States. I. Berki, Sylvester E. Il. National Center for Health Statistics (U.S.) lll. Series: National medical care utilization and expenditure survey. Series C, Analytical report; no. 2. IV. Series: DHHS publication; no. 86—-20402. [DNLM: 1. Health Services— utilization—United States. 2. Health Surveys—United States. W 84 AA1 H6) RA410.7.H54 1985 362.1'0973'021 85-18844 ISBN 0-8406—-0324-X Foreword This report is a contribution to the literature on health services research and health economics. Specifically, it was designed to investigate differences in the use of health care. There is a great deal of variation in the use of health care among members of the U.S. population with consequent wide variation in the expenditures for that care. This report should lead to improved understanding of the variation and provide data for those making public policy. The research, analysis, and publication was conducted under a contract (No. 282-83-2119) between the Division of Health Interview Statistics, the National Center for Health Statistics division responsible for the National Medical Care Utilization and Expenditure Survey, and the School of Public Health at the University of Michigan. Mary Grace Kovar, Dr.P.H., Special Assistant for Data Policy and Analysis, Office of Interview and Examination Statistics Program iii Contents ToT (=. og JP iii EXEOCULIVE SUMIMANY . Loot t iti t ttt ttt t ete e teeta teeta ciate en anaeaaneenennns 1 IRTOOUICHION «oo 5:50 000 www 6 050050 85: 8000 09 4 508 040 38 9000 0 6 HL BIE 88 RB 0 8 FAR BGR MA RR MEME PR Ere 3 OVBIVIBW «vo ov at min wmv hi 0 350 Bed 505 9 13 AE 18 0 08 BB FS 09 T6948 08 06 BUR 0B BRE 6 0 B30 00 2 TR 8 0 #10 6 0 9 00 BW WR BE WA 3 BOBKOIOUN . vce: «im on iss a 0 0 anna 0 0 8 10 8 1% 0m mo Rk #8 ow of es 4 0 ch 0 0 ek O00 BE OE ck 0 Hl 1 000 i HR 3 Objectives Of the REPO... . titi tiie iia iia aaa aaa 4 SOUICES Of DAA . .. oii i tite t tite tees 5 The National Medical Care Utilization and Expenditure Survey (NMCUES) ............ iii 5 LimRations of the Data . .....cuinisiasi5s0i 9s mai ni Bais i Rains SFiks RNIN RIfRia Mess BIE EN AC IE SOMONE ERNE 6 Reporting of Data On Charges .............iuniiuniin iii ie ie iit iit eria anaes asnsaanasnensnnnns 6 Exclusion of the Institutionalized Population . . . ....... iii i i i i i tite iii ie aaa 6 Reporting of Diagnoses and Medical Conditions ................. iii iii ia 6 Operationalizalion Of VBHADIBS: . x us sss sms mene mrs orm rma tm spi aem was ne sd wens Ss Ns Benes aanrauss sms wens 7 Definition of Low- and High-VOIMB USE «cu: vem rensmaimamue ss mat me/o asm sm Fsm emus mia sise im oes sess ssw omen 8 HOSPHBE DAYS oo ocu nemo 8 nn a am sf ms 20 mi i ik 6% im 8 ot wn i a 9B 6 ck 0 08 0 0 0 R60 B00 0 We E01 0 8 Ambulatory Care Visits . ..............iuiiui ii i eae aaa 9 Prascyribad Medicine ACUISIIONS «x: csm rr m sen sm mess sme ns @ ons em aime 8 Foes £m y sme wee sump sl yw wen own we ww vie 9 RINIVATIBE ANBIYSIS , «csv sivmsim viens @siv siv i 805050 8180 0 © B00 8a 18H 8 518 0 a ow ale Bs 8 81K R18 0 0 iF Wik 810 EW HIER ERR Hi 10 Distribution of Use in Each ServiCo CalBgory ...c:c sss vs rims isis asm isms as ese ves SMe Me @ Ima 200000 e 90 n we 10 Demographic Characteristics... .............iiuiiiiit iii iia iii iia sii e tiie eiaaaanns 1 Health Care COVEIAgE . .... ott tt tt tt ite ee ee ett e ett tienen eaasaenaananaennns 14 Health Status and Functional Limitations . . .......... titi i i i iit ities 15 Diagnostic CONMIIONS «ass vcuiiinmin mui snus insti Emer amin urvseui inn uiainuiniF ass muss nore 17 Decoders ... vo vviaranisinsias sib siss sion dniRs SiN nun sMuineR Ei Sua iuer eri ve ps R sare EME 18 Hospital Use and AMDUIBIONY VIBIS ..: cca vmuinmmemimmin sm stmin rns huss uvems vas vsnsmut nsw ssmuessnsins 18 Summary of Univariate Analyses . .............iuiiiin titi iii iii iii a ii i iain 18 Multivariable Analysis of High Use of Hospital Care . ......... ci iii tiie ii ii ii einen 20 Model for Predicting Use of Hospital Care ............ o.oo iii iie an eiaaannns 20 Model for Predicting High USE Of HOSPHAI Cale... . «occ rurerirrsertmsimsmsrversrnensmssnsmasnsmasnssenne 26 DISCUSSION! 54 isis wins venbii hE sins BR as SHES? RAE HIE TRE REMAP ERE MSO EEN W810 HF 0H A 4 08 0 4 8 we 8 32 ey TE TE tT Til IT 34 LISTE OF DRtBHED TADS: . « «civ ven ve vemos in iw we wim hm wom ms 005 88 8 Bf 30 9 90 od 309 30 hk 5% 00.00 7 8 0B CHE 003 000 8% 00 W600 9 35 APPENAIXES . . . . . eee ete ee eee eee ee eee eee eee 64 I. Sample Design, Data Collection, and Processing ................iuiniiiiniit initia aaaenaannennn 65 II. Data Modifications to Public Use Files . . .......... i iit iti it tana ral Mi. Analytical SIBIBGIBE «cou nemime mr wsms sus m mime abe mrss MEET PEE SHEEHY BIEN R SRSA PRE APA MAS rs ay 73 IV. Sampling BrirOrS . .....c osinimss iiss svermins ns smtu saan iss enim mins gs sivas nsnt ruin esissmmimsnsanens 81 V. Dolinifion Of TRITNS ...usvvrmransmsvs smsmaini sais imps iNsmaisi@uia Rue RBs sims R HIM IRME WER GONE HYE WY 86 List of Text Tables A. Thresholds of low and high use for each type of service: United States, 198V ................ ........ 0. 8 B. Estimated logistic regression coefficients and odds ratios for use of hospital care for all persons, by selected indepen- dent variables: United States, 1980 ......... cotta 22 ~ C. Probability of correct prediction by deciles of predicted probabilities for all persons, users, and nonusers: United SUIAION, VOBO ... ove domvons sR a 85% saw 504 56% 5005 F PIE TH A080 38 EH EEA RHE EW a we a 24 D. Predictive accuracy of alternative thresholds for predicting use and nonuse of hospital care: United States, 1980 ...... 26 E. Estimated logistic regression coefficients and odds ratios for probability of high use of hospital care among users, by selected independent variables: United States, 1980 .................iirnit ime 27 F. Estimated logistic regression coefficients and odds ratios for probability of high use of hospital care among users, by selected diagnostic categories: United States, 1980 .................. iii 28 G. Probability of correct prediction by deciles of predicted probabilities for high users, low and intermediate users, and all users of hospital care: United States, 1980 ................. oui iinitiet te 30 H. Predictive accuracy of alternative thresholds for predicting high use and low and intermediate use of hospital care: United States, 1980 ............... iii 30 List of Figures 1. Lorenz Curve of hospital days: United States, 1980 ................... iin ieee 10 2. Lorenz Curve of ambulatory care visits: United States, 1980 . ................uuuuureeeeee aan, 11 3. Lorenz Curve of prescribed medicine acquisitions: United States, 1980 .................ouuneeeeennnn. 11 4. Percent of persons under 45 years of age in 0-, low-, and high-use categories, by type of service: UNHOO SEAS, TBO .... i. ci vvuinnunaninsesosusmssmimnmmnn reves sss ss sss sss ess tn esm sss ss sss sn sms 12 5. Percent of persons with family incomes less than $15,000 in 0-, low-, and high-use categories, by type of service: United States, 1980 ............ 13 6. Percent of persons who rated their health as fair or poor in 0-, low-, and high-use categories, by type of service: United States, 1980 ............. iii ee 15 7. Percent of eligible persons reporting no activity limitation in 0-, low-, and high-use categories, by type of service: UNMBO SIABS, TOBO' . ooxismiio imam sis osm si a5 0k 58 ook 40m wom mn ho 40m #8 co mcs 006 we 50 0 90 3m 850 90 15 1 10 1 8 670 31 8 0. 0 0 mB 16 8. Percent of persons reporting at least three conditions during the year in 0-, low-, and high-use categories, by type of service: United States, 1980 . ...............uiiiiiiiiiie tee 17 9. Relative frequency distribution of predicted probabilities for users and nonusers of hospital care: United States, 1980 ........ 25 10. Relative frequency distribution of predicted probabilities for high and other users of hospital care: United SUBRES, TIBO ou vinvnime mma ss vines sam si 58% 60 54% £45 84 & 8 mmm 0k 08 8 Sik mak 48 0 0 916 5 90 2m 9 Bm B10 $0 LR ER 6 BR 29 Symbols vi - - - Data not available Category not applicable - Quantity zero 0.0 Quantity more than zero but less than 0.05 Test statistic is significant at 0.05 level ** Test statistic is significant at 0.01 level High-Volume and Low-Volume Users of Health Services: United States, 1980 by Sylvester E. Berki, James N. Lepkowski, Leon Wyszewianski, J. Richard Landis, M. Lou Magilavy, Catherine G. McLaughlin, and Hillary A. Murt, University of Michigan Executive Summary Data from the National Medical Care Utilization and Expenditure Survey of 1980 are used to examine the character- istics of high-volume users of health care services, contrasting them with low-volume users and those who used no services at all. The three major types of medical care services examined are hospital inpatient care, ambulatory visits, and prescribed medications. Low users were defined, respectively, as those who during the year had either one or two hospital days, one nondental visit to a physician or nonphysician, and one prescribed medicine acquisition. High users were those with, respectively, 17 or more hospital days, 20 or more visits, and 25 or more prescribed medicine acquisitions. A very small percent of the U.S. civilian nonin- stitutionalized population and of those who used services at all during the year consume a large percent of services in each of the three service types. High users of inpatient hospital care constitute 1.7 percent of the civilian nonin- stitutionalized population and 15 percent of persons hos- pitalized during the year, yet they used 54.4 percent of all hospital days used by the reference population. High users of ambulatory services constitute 4.5 percent of the reference population and only 5.7 percent of all users of ambulatory services, yet they accounted for 32.3 percent of all ambulatory visits. For prescribed medications, only 3.7 percent of the civilian noninstitutionalized population are high users, comprising 5.9 percent of all users, but they account for 32.9 percent of all prescription acquisitions. At the other extreme, low users of ambulatory care visits represent 17 percent of the reference population, and 21 percent of all users of such care, but only 3.3 percent of all visits. High users share certain characteristics. They are more likely than low users to be older and poorer, to have poorer health status and more medical conditions, and are more likely to have functional limitations. Both univariate and multivariable analyses show that the most important distinguishing characteristics of high users of any of the three medical services are poor health status, severe functional limitations, and the presence of multiple medical conditions—most importantly cancer, cardiac disor- ders, musculoskeletal diseases, respiratory diseases, and in- juries and poisonings. Almost all high-volume users of every category of service (88 percent for hospital days, 89 percent for ambulatory visits, and 94 percent for prescribed medications) had at least three different diagnostic conditions reported during the year. The likelihood of being a high user of hospital inpatient services, the most expensive component of health care, is increased not by the comprehensiveness of health care cover- age but by the need for health care, measured both by reported health status and by the presence of disabling or life threatening conditions. The evidence is persuasive that high-volume hospi- tal use is associated with severe illness, functional limitation, and death. Of persons with no functional limitation only 0.5 percent were high hospital inpatient care users, a percent that increased to 23 percent for those most severely limited in their activities, and to 54 percent of those who died during the year. Among the high-volume users of hospital inpatient ser- vices, 58 percent had 6 or more separate diagnostic conditions, and 53 percent had their health status reported to be fair or poor. This is in sharp contrast to persons who did not experience any hospital episodes during the year, among whom only 11 percent had their health status reported to be fair or poor. The results strongly suggest that demographic variables such as race, sex, and even age are related to hospital use only to the extent that they are associated with other factors such as reported health status, functional limitations, and health care coverage. High users of ambulatory services tend to differ from high users in the two other categories. Although they also tend to have the characteristics of poor reported health, severe illness, and the greater likelihood of at least some functional limitation found among high users of hospital inpatient care and prescription medications, they tend to do so to a lesser extent. Therefore, efforts to reduce high use in one category of service are not likely to yield comparable results in other categories. Across all three service categories, low users were found to be similar to nonusers in terms of age, perceived health status, and presence of activity limitation. However, for hospi- tal days low users were found to be different from nonusers with respect to health care coverage, suggesting that health care coverage continues to be important in determining access to hospitalizations. Although specific patterns vary for each service category, in general low-volume users were found to have significantly different characteristics from high-volume users, especially on need-related variables: They had higher perceived health status, less functional limitation, and fewer medical condi- tions. The findings on the characteristics of high users of medical care services are sobering: 54 percent of all hospital days, 32 percent of all ambulatory visits, and 33 percent of all prescribed medications reported for the U.S. civilian nonin- stitutionalized population in 1980 were consumed by a very small percent of that population. These high users of medical care services were predominantly sick, functionally limited in their activities, or dying. In an effort to contain costs, approaches to reduce the incidence of conditions that lead to high use of medical care resources need to be considered. Alternative treatment modalities and institutional structures to reduce to costs of management of conditions that cannot be prevented should also be explored. NOTE: We are grateful for the support we received during all stages of the prep- aration of this document both from our colleagues at The University of Michi- gan and from the staff of the National Center of Health Statistics. At The Univer- sity of Michigan, Sharon Stehouwer contributed greatly to the initial analyses of NMCUES data and to identification and correction of several problems en- countered in the data base. Kenneth E. Guire was responsible for the overall preparation and correction of data files on which this report is based. P. Ellen Parsons provided much appreciated assistance on many aspects of the prepara- tion of this report. Valuable consultation on matters related to use of medica- tions was given by Dr. Duane Kirking. Quality secretarial support in the prepa- ration of the many tables included in the report was provided by Katherine Met- calf. At the University’s Institute for Social Research, Nan Collier developed software for calculating sampling errors, and Judy Connors performed many of the analyses for generating sampling errors for national estimates. - We also received continual support and guidance from the National Center for Health Statistics and in particular from our project officer, Dr. Mary Grace Kovar, Special Assistant for Data Policy and Analysis. We are indebted to Dr. Robert J. Casady, Chief of the Statistical Methods Staff, for writing the major section in Appendix I, describing the survey design and estimation methodol- ogy for NMCUES. Robert Wright and Michelle Chyba quickly responded to all our queries about problems we encountered with the data. Editors in the Pub- lications Branch provided valuable assistance during all stages of the report, especially preparation of the detailed tables. Introduction Overview This report examines the characteristics of high-volume users of health care services, contrasting them with low-vol- ume users and those who used no services at all. The data for this study are from the public use files for the National Househould Survey component of the National Medical Care Utilization and Expenditure Survey (NMCUES). Interviews for the survey were conducted between early 1980 and mid- 1981. The survey and public use files are described further in the section on “Sources of Data.” The next subsection briefly summarizes relevant past work, followed by a statement of the objectives and the scope of this report. Background The small percent of the population that in any given year makes extensive use of health services has been of continuing interest to policymakers and analysts, because this group accounts for a disproportionate share of total health expenditures and includes individuals who are at high risk for financially catastrophic health care expenses. Yet to date only modest efforts have been made to estimate the number and other characteristics of high-volume users in the United States. The most ambitious of these have focused on estimat- ing, for high-cost illnesses—the incidence, cost, and contribu- tions of different payers. Although based on a variety of sources of data, such efforts rely most heavily on the two major national surveys of health care services use, the National Health Interview Survey (NHIS) and the periodic surveys conducted by the Center for Health Administration Studies and National Opinion Research Center at the University of Chicago (Trapnell, 1977; Congressional Budget Office, 1977; Birnbaum, 1978a and 1978b). Other studies, more limited in scope, focus on subgroups defined by the source of data on which the study is based, such as third-party payers (Forthofer et al., 1982; Congressional Budget Office, 1982; Anderson and Knickman, 1984) or hospital records (Schroeder et al., 1979; Zook and Moore, 1980; and Kobrinski and Matteson, 1981). Additional information on high-volume users is available from studies not directly concerned with this population but which nevertheless include patient groups who tend to be high-volume users, including those with cancer (Cancer Care, 1973; Eldred et al., 1977), stroke (Weinfield, 1981), spinal cord injuries (Webb et al., 1977; Anderson et al., 1980), those in intensive care units (Cullen et al., 1976; Budetti et al., 1981; Detsky et al., 1981), and the terminally ill (Bloom and Kissick, 1980; Gibbs and Newman, 1982; Lubitz and Prihoda, 1984; McCall, 1984; Scitovsky, 1984). This report is based on recent data from a national survey and provides national estimates of the number and characteris- tics of high-volume users. It also contrasts high-volume users with low-volume users and those who used no services at all. Objectives of the Report Relying on the database generated by NMCUES, this report investigates the characteristics of high users of health services and compares such high users with other categories of users, in particular low users. The report relies on a comprehensive set of univariate analyses of the major categories of use: Hospital inpatient care, ambulatory care, and prescribed medications. Inpatient hospital use is further analyzed by the application of multivariable regression methods, designed as a major step in modeling and as an indication of the potential fruitfulness of applying these ap- proaches to the analysis of other major categories of service. This report presents national estimates of the number and characteristics of high users of health services within the civilian noninstitutionalized population of the United States during 1980 for each of three categories of service: Ambulatory care, inpatient hospital care, and prescription medications. The report also illuminates the characteristics of high-volume users by contrasting high-users with low-users and nonusers of services. Detailed national estimates of major policy-relevant attri- butes of high-volume and low-volume users are presented: * Number and percent of the total population each group represents. * Proportions of total volume of services each group ac- counted for. * Average charges per person and use levels per person. o Demographic characteristics. * Sources of payment for the care. e Medical characteristics. The multivariable analysis of inpatient hospital use presents the independent effects of variables hypothesized to be related to high-volume use, holding statistically constant the effects of other variables. As indicated in the section on “Limitations of the Data,” characteristics of the information available on charges made it necessary to limit both types of analysis to use of services rather than incorporating both use of service and charges. Sources of Data The National Medical Care Utilization and Expenditure Survey (NMCUES) Between February 1980 and April 1981, data on 17,123 persons in 6,798 families were collected at approximately 3-month intervals for a total of five interviews: Two personal interviews followed by two telephone interviews, and a final personal interview. At the conclusion of the first interview, survey participants were provided with a specially designed calendar-diary for recording data about medical events and costs in preparation for subsequent rounds of interviewing. Prior to each interview after the first, interviewees were sent a summary sheet with all medical events and costs reports in previous interviews. Specific details concerning the sample design and data collection are outlined in Appendix I. This report is based on data in the public use tapes, which consists of six files: The person, medical visit, dental visit, hospital stay, prescribed medicines and other medical expenses, and condition files. The person file has one record for each of the 17,123 responding eligible persons with data describing the person’s demographic characteristics, health care coverage, employment, income, and usual source of care; numbers of visits, hospitalizations, and other medical events reported for 1980; total charges for each category of care; and limitations and disabilities, including identifica- tion of conditions. Data from the other five files, which have more detailed information about events summarized in the person file, can be linked to records in the person file through a unique identification number assigned to each person. Limitations of the Data This section describes some of the limitations of NMCUES data that determined the scope of this report and that need to be taken into account in interpreting its results. Reporting of Data on Charges Analyses here reported focus on the use of services rather than on charges or costs. The emphasis is on compari- sons of different levels of use, particularly low versus high. The principal reason for emphasizing use of services is that NMCUES data contain a number of improbably low values of total annual charges for ambulatory visits, prescribed medicines, and hospital stays. When total annual charges for all services were examined for those with charges greater than 0, the lowest 5 percent were between $1.00 and $21.00. It is likely that for many of these cases the reported data do not reflect the actual charges for the services received. These reported charges are more likely to be out-of-pocket expenses incurred, in particular the deductibles and copay- ments paid by those who have insurance, or expenses not covered by Medicaid. Inclusion of such cases in the low-charge category distorts the characteristics of that class of individuals, especially since the number of such misreported cases appears to be quite large. Similarly, certain high charge hospitaliza- tions may also be excluded. . The problems posed by out-of-pocket expenses being misreported as total charges are compounded by the high rate of nonresponse on questions about charges, resulting in imputed charges for about 36 percent of all hospital admis- sions and 26 percent of ambulatory visits. This complicates the task of characterizing certain categories of cases: Because imputations of charges were based on such variables as age and type of provider but not on diagnosis, incongruous pairings of charges with diagnoses may result. The combination of these biases from incorrect responses and nonresponses on charges indicated the desirability of focusing these analyses on high and low users defined in terms of volume of services and not in terms of charges. Exclusion of the Institutionalized Population Because the institutionalized population was not included in the NMCUES sample frame, all data presented here on high users of care apply to the U.S. civilian nonin- stitutionalized population only. This is consistent with many previous studies that had a similar focus. It must be recog- nized, however, that most, if not all, of those in health care institutions are high users. For this reason the picture presented here is not complete. Reporting of Diagnoses and Medical Conditions In NMCUES a person was noted as having a particular medical condition only when the condition caused some type of restriction of activity or resulted in an ambulatory visit, hospitalization, purchase of a prescribed medicine, or other encounter with the medical care system. Thus conditions that did not cause any restriction in activity or days in bed, or did not lead to a visit to a doctor, for example, were not recorded. For conditions that were reported, the diagnostic accuracy depends both upon the information provided by the health care provider and on the respondent’s ability and willingness to accurately convey this information to the inter- viewer. For each medical encounter recorded in the survey, re- spondents were allowed to report up to four medical condi- tions. The public use files show that (a) approximately 10 percent of medical visits had two or more conditions recorded, (b) multiple conditions are listed for about 12 percent of all hospital stays, and (c) 4 percent of the prescribed medica- tion records have two or more conditions recorded. The NMCUES survey instrument does not designate which of the underlying conditions is the “principal diagnosis” or the primary reason for each medical encounter. Because a princi- pal diagnosis is not identified for each medical event, it is impossible to determine to which condition health services use should be attributed when multiple conditions are reported. This poses a problem in developing accurate estimates of the extent to which a given illness is associated with given levels of utilization of health care resources. Operationalization of Variables Each of the variables included in this study is briefly e Prescribed medicine use is measured by the number defined in Appendix III. The three key service categories of acquisitions, that is, the total number of times a on which this analysis focuses are defined as follows: person had a prescription filled, regardless of whether . : : it was an initial filling or a refill of a prescription. e Hospital care use is measured in terms of number of g p pe days spent in the hospital. It should be noted that persons The other variables used in the analysis refer to demo- who were admitted to and discharged from a hospital graphic and other characteristics, such as age, sex, income, on the same date, i.e., they did not stay overnight in education, health care coverage, medical conditions, perceived the hospital, are counted as having zero hospital days. health status, and functional disability. Their definitions, given e Ambulatory care use is defined as the number of all in Appendix III, are consistent with those in other reports nondental visits made to physicians as well as to nonphysi- that are based on the same NMCUES public use files. The cians, whether in a private office, a hospital outpatient procedures for applying weights to all these variables to department, or emergency room. derive national estimates are described in Appendix V. Definition of Low- and High-Volume Use No commonly agreed upon definitions of low and high use of health services exist. Several categories of definitions can be identified in past studies, and within categories the thresholds specified vary as well. In many instances the definitions are in terms of costs. When defined in terms of an absolute dollar amount, the threshold for annual inpatient expenses has ranged, in studies done in the late 70’s and early 80’s, from $3,000 to $10,000 (see Birnbaum, 1978; Schroeder et al., 1979; Congressional Budget Office, 1982). Other definitions are based on the distribution of cases, focus- ing on the top 5 percent (e.g. Lubitz and Prihoda, 1982), or on a more elaborate statistical definition such as the specifi- cation of “outliers” under Medicare’s Prospective Payment System. In this study, four levels were defined for each category of use analyzed: Low-volume and high-volume as well as zero and intermediate use, to provide a fuller contrast for the low and high categories on which the study focuses. The fol- lowing two criteria were specified a priori for the determination of cut points: 1. The number of cases included in the high and low categories must be sufficiently large to permit meaningful comparisons across characteristics of interest. Based on the analyses anticipated, 300 was the estimate of the minimum number needed in the high and the low categories for each service. 2. To the extent possible, the cut points should isolate the extremes of the distribution of use of services that account for disproportionate shares of the total, and are therefore often the subject of study. That includes, in particular, the top 5 percent, which tends to account for one-third to one-half of total volume, and the bottom 20 to 25 percent, which accounts for 3 to 5 percent of total volume. The requirement that there be at least 300 cases in each category was found to be the binding constraint in most instances, except for the low-use categories of service for ambulatory care and prescribed medicine acquisitions, where even taking the lowest possible nonzero number, one visit and one acquisition, yielded several thousand cases in each instance. Each threshold is discussed next, and the values and characteristics of thresholds are shown in Table A. (Percents shown in Table A are based on unweighted counts and therefore differ from similar percents given in Table 1-29, which are all based on weighted data.) Table A Thresholds of low and high use for each type of service: United States, 1980 Number in Percent Percent Type of service the category of all of total and level Threshold (unweighted service service of use (7) count) users volume Hospital days LOW oc aiainninisivins 0**** = 0.0794, and the odds converted to the predicted probability by computing 0.0794 / (1 + 0.0794) = 0.074. That is, the hypothetical white male with the specified characteristics has a predicted probability of being hospitalized of 7.4 percent. The relative importance of the different factors in the model can be examined by changing the characteristics of the hypothetical individual and estimating the probability for an individual with the new characteristics. For example, sup- pose that the hypothetical white male had reported poor health status instead of good health status. The predicted probability of hospitalization for an individual with all the same characteristics as the original hypothetical white male except for poor health status can be computed as previously shown. The sum of appropriate coefficients (i.e., the estimated logarithm of the odds) is —2.8445 + 0.0984 + 1.5169 — 0.0728 + 0.3468 — 0.0608 = — 1.0160. Exponentiating, the estimated odds are e'%'%® = 0.3620, which is converted to the predicted probability as 0.3620/(1 + 0.3620) = 0.2658. That is, poor health status relative to good health status has increased the risk of hospitalization for the hypothetical person by 3.35 times (i.e., 0.2658 / 0.0794). This increase is reflected in the large odds ratio for the poor health status category relative to the good health status category in Table B. Alternatively, consider the risk of hospitalization for a white female with the same characteristics as the original hypothetical white male (i.e., 35-54 years of age, middle income, private insurance, good health status, a usual source of care, residence in the Northeast Region). The predicted probability of hospitalization for the female is computed in the same three steps as outlined for the hypothetical male: (1) summation of coefficients to obtain the logarithm of the odds as —2.8445 + 0.0984 + 0.0766 — 0.0728 + 0.3468 — 0.0608 = — 2.4563; (2) exponentiation to obtain the odds as e 24563 = 0.0858; and computation of the predicted probability as 0.0858 /(1 + 0.0858) = 0.0790. Thus, the predicted probability of hospitalization for this female is 7.9 percent, somewhat higher than for the hypotheti- cal male because the partial regression coefficient for the female group is positive. In addition, if the female were actually 35-44 years of age (and hence the indicator for females 17-44 years of age applies), the predicted probability 23 of hospitalization is 11.7 percent. The interested reader can readily construct predicted probabilities for other individuals using the coefficients in a similar manner. A more detailed description of the estimation procedure for predicted prob- abilities is given in Appendix III. These methods for estimating predicted probabilities are useful for interpreting the relative importance of the factors examined in the model, but there are several limitations to the general use of these logistic models. For one, the predicted probabilities for a hypothetical individual do not necessarily define patterns of hospitalization because the hypothetical person defined may represent a small portion of the population. In addition, variables are included in the model which are not statistically significant because findings in the medical care literature indicate that these variables represent important predictors of hospitalization. Nonetheless, including these variables in the model most likely leads to less efficient estimates than could be obtained from a model without them. Finally, the model in Table B explains a small, but useful, amount of variation in the probability that a person was hospitalized in 1980. The model accounts for 5.5 percent of the error remaining if the mean probability of 12.2 percent was used as a predicted value for every person in the sample. Thus, there is still a large amount of error that may be explained by other factors that have not been included in the model. Those other factors may reduce the importance of the factors considered in this model and substantially change the predicted probabilities for hypothetical individuals considered here. To assess the goodness of fit of the model further, the predicted probability of being hospitalized was computed for every person in the dataset. The distribution of the pre- dicted probabilities was then examined for users and nonusers of hospital care to determine whether the model was able to discriminate between these two groups. Figure 9 shows the relative frequency distribution of the predicted probabilities by deciles for users and nonusers of hospital care. A good fit of the model to the data would be observed if the predicted probabilities under the model were small (e.g., less than 50 percent) for nonusers and large for users. The results in Figure 9 suggest that the model does provide some discrimination between users and nonusers, although perhaps not as much as would be desired. In particular, 53 percent of the nonusers had predicted probabilities less than 10 percent, although only 29 percent of the users had predicted probabilities in that range. Similarly, 91 percent of the nonusers had predicted probabilities less than 20 percent compared with 71 percent of the users. Conversely, only 3 percent of the nonusers had predicted probabilities greater than 0.30, with 13 percent of the users had predicted prob- abilities greater than that value. The ability of the model to discriminate users and nonusers through the assignment of predicted probabilities close to one and zero, respectively, is evident from the distribution in Figure 9. At the same time, though, the low proportion of the unexplained error accounted for in the model is also demonstrated by the lack of clear discrimination of users and nonusers by the predicted probabilities. 24 The goodness of fit of the model may be examined further by considering the extent to which the model improves the prediction of use or nonuse. Without the model, one could classify all persons in the sample as users or as nonusers. If all are classified as nonusers, then because 11.5 percent of the population are users, 88.5 percent of the persons would be predicted correctly as nonusers. With the model, a predicted probability that the person is a user can be obtained. By designating selected values of the predicted probabilities as thresholds (such that persons with a predicted value above the threshold are predicted users and those below are predicted nonusers), the ability of the model to predict use or nonuse can be examined. Table C presents the probability of correctly classifying users, nonusers, and all persons when thresholds correspond- ing to deciles of predicted probabilities are used to define predicted use and nonuse groups. For example, if the threshold were chosen to be 0.1, then persons with predicted prob- abilities greater than 0.1 would be predicted users and the remainder are predicted nonusers. Using this threshold value, 70.6 percent of the users are correctly predicted to be users, while only 52.7 percent of the nonusers are correctly predicted. Overall, 54.7 percent of the persons in the sample would be predicted correctly using the 0.1 threshold for predicted probabilities. The predicted probabilities for users and nonusers in Table C correspond to the sensitivity and specificity of a test procedure based on thresholds at different levels. The probability of a correct prediction for users is the sensitivity of the test at a given threshold, and the probability of a correct prediction for nonusers is the specificity of the test. As the threshold value increases in Table C, the probabil- ity of a correct prediction drops for users and increases for nonusers. For all persons, the probability increases up to the 0.5 threshold and then decreases slightly to the preva- lence of nonuse (i.e., 88.5 percent). In other words, the threshold of 0.5 would provide the largest number of correct predictions over all the tests given in Table C, but this finding is dominated by the fact that 88.5 percent of all persons were nonusers. Table C Probability of correct prediction by deciles of predicted probabilities for all persons, users, and nonusers: United States, 1980 Probability of correct prediction All persons Users Nonusers 0.115 1.000 0.000 0.547 0.706 0.527 0.836 0.282 0.908 0.873 0.124 0.970 0.882 0.047 0.990 0.886 0.015 0.998 0.885 0.002 1.000 0.885 0.001 1.000 0.885 0.000 1.000 0.885 0.000 1.000 0.885 0.000 1.000 Figure 9 Relative frequency distribution of predicted probabilities for users and nonusers of hospital care: United States, 1980 Proportion 0.50 0.40 0.30 0.20 0.10 0.53 0.29 0.38 0.42 20 Predicted probability of hospital use ] Nonusers of hospital care a Users of of hospital care 100 25 Table D Predictive accuracy of alternative thresholds for predicting use and nonuse of hospital care: United States, 1980 Predictive accuracy Threshold of predicted probability Use Nonuse OO .cvvnunnnnnnnsissns vgs Buss 0.115 wt & 01 0.162 0.933 D2... ssn —————— 0.284 0.907 OB wis tmmmnmnnnsssmmmmn om mmmemmmm 0.345 0.895 D0 oiiuinssasnanmnnnnnnssenmmms 0.376 0.889 05 ovevvsusssnnsssniuniasnnpnss 0.522 0.887 OB oouunncnnnsznssnnns mss mmmmm 0.465 0.885 07 .eviinissironssonrsnnnmmrnas — 0.885 OB .uvunusnsonsnssnansnnsbosning dn 0.885 09 aw 0.885 10 seoprronnsnsossabinibssoisons ose 0.885 It is also useful to examine the ability of the model to pre- dict use among all persons classified as users when a given threshold is selected. Table D presents the predictive accuracy of classification schemes corresponding to different thresholds. For example, if a threshold of 0.1 is used, 16.2 percent of those classified as users are actually users; 93.3 percent of those clas- sified as nonusers are actually nonusers. The predictive accu- racy for nonuse does not fall below 88.5 percent (when a threshold of 0.6 or greater is used), while the predictive accu- racy for use does not fall below 11.5 percent (when a threshold of 0.0 is used, or when everyone is classified as a user). As in Table C, the best predictive accuracy for use occurs at the 0.5 threshold when 52.2 percent of those classified as users are actually users. The general impression from this examination of the probability of correct prediction and predictive accuracy is consistent with the goodness of fit measure given in Table A (i.e., proportional reduction in error of 5.1 percent). The sensitivities of the tests based on each threshold level examined are low, as are the predictive accuracies. The model offers some improvement in the ability to predict use and nonuse compared with simple ‘all or none’ classification schemes, but there is still room for considerable improvement in the fit of the model to the data. The separate model constructed by adding a functional limitation score (treated as a continuous variable) to the model shown in Table B is not presented here. Functional limitation score in that model had the largest partial correlation with hospitalization of any variable in the model. Yet per- ceived health status indicator variables continue to have signifi- cant, though smaller effects than in the model using all the data, as does the indicator variable for low income persons with poor health status. This suggests that functional limitation is associated with hospitalization in some way that is different from reported health status. Model for Predicting High Use of Hospital Care Given that factors have been identified which are predic- tive of hospitalization, it is of interest to examine whether these 26 same factors or a different set are important in determining high use of hospital care. As described previously, a second logistic regression analysis was conducted to investigate factors related to high use among those reporting at least one overnight hospital stay during the year. Although this second model takes its point of departure at hospitalization, the focus of the first model, it is misleading to consider it a second-stage model since it does not incorporate any of the estimates from the first model. The dependent variable in this second model is again a dichotomous measure, with high-use persons (i.e., those with 17 or more hospital days reported in 1980) given a value equal to one and all other persons, those with more than one but fewer than 17 days, given a value of zero. Those with no hospitalization are excluded from this analysis. The independent variables are listed in Table E and include the same demographic, income, region-of-residence, health- care-coverage, and perceived-health-status variables as were “used in the first model. The rationale for including these same variables in both models is that these factors may not only affect whether any hospital care was obtained—the focus of the first model—but also affect how much care was used, and more specifically whether someone was a high user of hospital care or not. In addition to these basic independent variables, six diag- nostic categories were included in the model and are shown in Table F. They include the disease categories that in other studies have been found to be associated with high-cost hos- pitalizations and which were also found to account for the largest percent of high-use hospitalizations in the descriptive analyses presented earlier. The six categories are neoplasms, diseases of the circulatory system, diseases of the respiratory system, diseases of the digestive system, diseases of the musculoskeletal system and connective tissue, and injury and poisoning. It would have been preferable to identify persons who not only had these diagnoses, but who also had surgery; these diagnoses combined with surgery have been found to be highly resource-intensive. As a preliminary investigation of the joint effects of surgery and each diagnosis, a set of indicator variables was added to the model that was equal to one if the individual had surgery during 1980 and a selected diagnosis during at least one hospital stay and was equal to zero otherwise. These indicators when added to the model in Table F accounted for only another 0.5 percent of the unexplained error. In addition, approxi- mately 80 percent of the persons hospitalized during the year for neoplasms or a cardiac condition also had surgery during the year. Thus, the addition of the combined surgery and diagnosis indicators to the model contributed little to the explanatory power of the model, and the indicators were not included in the final model. The estimated logistic regression coefficients and odds ratios from the model are presented in Tables E and F. The high-use logistic regression analysis was based on 2,087 persons with one or more hospital days in 1980, of whom 315 or 15.1 percent were hospitalized 17 or more days in 1980. As a result of the smaller sample size, standard errors of coefficients are higher for this model than for the previous Table E Estimated logistic regression coefficients and odds ratios for probability of high use of hospital care among users, by selected independent variables: United States, 1980 Odds ratio Regression coefficient 95-percent re eeee———— Ratio of confi eval Adjusted coefficient aan ee Independent standard to standard Lower Upper variables Estimate error error Estimate limit limit Constant ......... iii -2.1128 Sex (Male) .....ccovuvusrssansannnsuonssnsssnsns «an ee oe oe. oe. ce. Female .............coiiiiiiiiiiiiiiinennnn —-0.5087 0.1579 -3.2216 0.6013 0.4412 0.8194 Race White andother ........................on. —0.2447 0.2516 -0.9727 0.7829 0.4782 1.2820 Black) ........ iii Perceived health status Excellent... —-0.5871 0.1592 —3.6889 0.5559 0.4070 0.7595 (GOODY .:sssvrvrressnsvsnssspsoesissssssgians : ye _— —y _— ce ce PIF oo covssnssesrireivavivronsnsnnusnnanas 0.0730 0.2073 0.3521 1.0757 0.7165 1.6151 POOF .oovvunvvnrrnssssnnnnnshisaaviissieis 0.8624 0.2792 3.0889 2.3688 1.3705 4.0944 Age Under35Years .........coooessssnnnssssnsss —-0.7582 0.2377 —-3.1899 0.4685 0.2940 0.7465 (35-BAYORISY : .:viuuvvssvsnssssnssnsnsnnnan sown ce cee cen ce en B5-TAYOArS .........ccoviiiiuuninnnenneenns 0.0986 0.2678 0.3682 1.1036 0.6530 1.8653 75yearsandover.................ciinniannn 0.6290 0.3702 1.6991 1.8757 0.9079 3.8752 Poverty status’ Below 200 percent of poverty level ............. 0.1967 0.1225 1.6059 1.2174 0.9575 1.5477 200-499 percent of poverty level ............... -0.0277 0.1405 -0.1971 0.9727 0.7385 1.2811 500-699 percent of poverty level ............... -0.1159 0.1929 —0.6009 0.8906 0.6102 1.2997 (700 percent of poverty level or more) ........... a. Health care coverage MUIDIG PUBRE + + 225 cannaanssinsssssnosvemmne 0.5040 0.3285 1.56342 1.6553 0.8695 3.1515 SINQIOPUDNG & 4555553000 000s05smvmmrrmmwene 0.3899 0.3253 1.1985 1.4768 0.7805 2.7942 Private and public ............ cc... 0.4550 0.2732 1.6656 1.5762 0.9227 2.6923 PHVB ONY . oo onus ssanasssnsssrsmmmmmmpmvoe 0.1308 0.2147 0.6093 1.1397 0.7483 1.7359 (Noneorother) ...............coovviininnnnn. vot wha wo _ vom oo. Region of residence (Northeast) ..............covuunnninnnnnnnnn —— we ee awe North Comtral . ... 5 +55 5 93 3 8 svmsEemesmosns —-0.1840 0.2241 -0.8210 0.8319 0.5362 1.2908 SOUth ooo a -0.4174 0.2162 —-1.9308 0.6588 0.4312 1.0063 WBE .......onninniis ss ves Shamma wmme. -0.3941 0.2564 -1.6372 0.6743 0.4080 1.1145 Analysis of variance parameterization. NOTE: Proportional reduction of error = 18.8 percent. model, which was based on data for the entire sample popula- tion. There are several interesting contrasts with the results from the first model. The income and health care coverage variables which were so prominent in the first model are not significant predictors of high use. This suggests that although those factors may be important determinants of whether someone becomes a user of hospital services (that is, they are indicators of access), they are not determinants of whether someone who had a hospitalization will be a high user. Other factors, more closely related to need, are prominent in the high-use model. Among the diagnostic categories that were hypothesized to entail the need for large amounts of care, five of the six are strongly and significantly related to high use of hospital days. Moreover, if the high-use logistic regression analysis is carried out without the diagnostic indicators, the coefficients for the remaining variables in the model are quite similar to those shown in Table F. In other words, the diagnostic categories are providing differ- ent information about risk of high use of hospital care than the other variables in the model. In addition, this model explains 18.8 percent of the error in predicting the probability of high use, and the diagnostic categories alone account for 6.6 percent of the unexplained error. Thus, including diagnostic-category variables directly improves the explana- tory power of the model. On the other hand, the surgery indicator variable was not significant. This finding is not very surprising, because 27 Table F Estimated logistic regression coefficients and odds ratios for probability of high use of hospital care among users, by selected diagnostic categories: United States, 1980 Odds ratio Regression coefficient percent R Ratio of Wo interval Adjusted coefficient —_— Independent standard to standard Lower Upper variables Estimate error error Estimate limit limit Neoplasm PIES .....ovvvvivvivvnmsmmms wmv dE RRR 1.7690 0.3278 5.3966 5.8650 3.0849 11.1505 (Absent) ...............iiiii i, oe & ah 4 hi _— -— a Circulatory system Present ........... ii. 1.1163 0.2015 5.5337 3.0505 2.0550 4.5282 CABBOIMY .u5nininnis nists mex sommes eg _— ve ces ces ces Respiratory system PYOBBIN «5 aaa 333550255%4mn0s 00mm wmmmmminns 0.7006 0.2198 3.1879 2.0150 1.3098 3.0998 (ADSBNY) .........c000unsusss58vannssnsnnsss oy Ce was en wi phd Digestive system Present ................. i. 0.0263 0.2032 0.1294 1.0266 0.6894 1.5288 (ADSBNMY ...::vuvisssioinnmnnrmnnnnnnnnnnns cen wun op cen _ es Musculoskeletal system PrOSOML . ...... ives iacivismrannnnnsssnsn 1.0892 0.3191 3.4128 2.9719 1.5899 5.5552 (Absent) ................ ii sd cee —_— wes ahs “4 Injury or poisoning Present ..................... a. 1.3743 0.2326 5.9077 3.9523 2.5051 6.2355 (ADB): cuiiiihiviiis mmm wwe www vv nx wu 68 _ "eh cen cee wow yo Surgery ANDY, om 000 rT EB Ea bi rma es vo wes 2 i ces ces wwe YES 0.2624 0.1646 1.5945 1.3000 0.9416 1.7949 NOTE: Proportional reduction of error = 18.8 percent. it was expected that surgery would be an important factor not through its direct effects, but through interaction with the diagnostic categories. Sex is significantly related to high use of hospital care in the second model. Being female is strongly and negatively associated with high use when the other variables in the model are taken into account, something that is not readily apparent from the univariate analyses presented earlier. Age, on the other hand, is significant only to the extent that those under 35 years of age are significantly less likely to be high users than the reference category of those 35-54 years of age. Perceived health status is significantly related to high use, even though the five diagnostic categories described earlier are also significantly related to high use. This suggests that the diagnostic categories capture a different dimension than perceived health status. As in the hospitalization model, the relative importance of the factors in the logistic regression model may be examined by estimating the probability that a hypothetical individual may be a high user of hospital care once hospitalized. Consid- er, for example, a black female 35-54 years of age in good health with low income (i.e., below 200 percent of poverty level) and no health care coverage living in the Northeast 28 Region, who is hospitalized with a diagnosis other than the six considered in Table F—and who does not have surgery. The predicted probability that this hypothetical female will have 17 or more hospital days in 1980 is computed in three steps as for the previous model: (1) summation of appropriate coefficients as —2.1128 — 0.5087 + 0.1967 = —2.4248; (2) exponentiation of this logarithm of the odds as e 24248 = 0.0885; and (3) computation of the predicted probabilities as 0.0885/(1 + 0.0885) = 0.0813. If her health care coverage had been through Medicaid only rather than none, the predicted probability increases to 11.6 percent, a 42-percent increased risk of high use. On the other hand, if she still had no health care coverage, but reported poor instead of good health status, the predicted probability increases to 17.3 percent, a 114-percent increase in risk over the base set of characteristics. Finally, the importance of the diagnosis is illustrated by considering the same hypothetical female with the original set of characteristics, but with a diagnosis of respiratory disease and surgery during the year while in a hospital. The estimated risk of high use of this individual is 18.8 percent, 2.32 times higher than for the original person. Again, caution should be exercised in interpreting the predicted probabilities from the model in Tables E and F. The model does not explain all the variation in high use of hospital care, and hence the predicted probabilities are subject to error. The error is also increased by the use of less stable coefficients that were not statistically significant. In addition, these predicted probabilities must be applied to population estimates to assess the importance of the factors considered in the model to high use of inpatient hospital care. The goodness of fit of the model for predicting high use of hospital care was assessed by examining the distribution of predicted probabilities for high users and all other users (i.e., low and intermediate users). Figure 10 presents the relative frequency distribution of the predicted probabilities from the model for high users and all other users. As before, Figure 10 Relative frequency distribution of predicted probabilities for high and other users of hospital care: United States, 1980 0.60 p— 0.50 p— 0.40 — Proportion 0.30 — 0.20 p— 0.13 0.10 |— 0.08 10 20 30 40 Predicted probability of high use of hospital care 1 Low and intermediate users of hospital care SR High users of hospital care 50 60 70 80 90 100 29 the model fit is better to the extent that high users are assigned predicted probabilities that are closer to one and other users are assigned values closer to zero. Although the ability of the model to make such an assign- ment is clearly less than perfect, there is evidence in Figure 10 that the model is discriminating between these two groups. For example, 97 percent of the low and intermediate users have predicted probabilities less than 0.50; only 74 percent of the high users have such values. Conversely, 26 percent of the high users have predicted probabilities greater than 0.50, compared with only 3 percent of the low and inter- mediate users. The model is showing some discrimination between high and all other users. The ability of the model to improve prediction of high use of hospital care can be further understood by examining the probability of correct prediction and predictive accuracy at selected thresholds for the predicted probabilities. Without the model, a simple classification scheme with good predictive power would assign all users to the low or intermediate use group, and since only 15.0 percent of the population were high users, the simple classification scheme would cor- rectly predict low or intermediate use for 85.0 percent of the users. It is of interest, therefore, to examine whether the model can improve the ability to predict correctly the status of users beyond that achieved by the simple classifica- tion scheme. Table G presents the probability of correct classification for high, low or intermediate, and all users at deciles of the predicted probabilities for each user. For example, suppose the scheme is to classify all persons with predicted probabilities greater than 0.1 as high users and all others as low or intermediate users. Among high users, the 0.1 threshold would predict correctly 84.9 percent of the time, whereas for low or intermediate users the probability of a correct prediction is 63.5 percent. The probability of a correct prediction with the 0.1 threshold is 0.667 for all users. As noted previously, the probability of a correct prediction for high users is also the sensitivity of the 0.1 threshold test procedure, and the Table G Probability of correct prediction by deciles of predicted probabilities for high users, low and intermediate users, and all users of hospital care: United States, 1980 Probability of correct prediction Low and Threshold of intermediate predicted probability High users users All users [1 7 + J 1.000 0.000 0.150 {+5 TI TTI 0.849 0.635 0.667 02 ... 0.652 0.814 0.789 03 cosceverisirssiinenng 0.495 0.896 0.836 04 ... 0.364 0.946 0.859 08 itt 5H E8845 mmm nas 0.265 0.975 0.868 0B | uncucain 3 5 ow & HE EBRRRA 8 0.169 0.987 0.864 07 0.089 0.996 0.860 OB ion £3 4 0 5.4 Bn sahndba 8 8 0.037 0.999 0.855 09 cosine vv sae suns 4 0.020 1.000 0.853 10 0.000 1.000 0.850 probability of a correct prediction for the low or intermediate users is the specificity of the test. Table H Predictive accuracy of alternative thresholds for predicting high use and low and intermediate use of hospital care: United States, 1980 Predictive accuracy Low and Threshold of intermediate predicted probability High use © use 00 oii 0.150 vig OI ne EE 0.291 0.960 02 cena mR 0.382 0.930 03 0.457 0.909 08. mn R GE 0.546 0.894 O05 ....ccnvnnnsmmmamonmmensniiomemmngs 0.650 0.882 O08 nimvininimmsmmsmsmmmma————"— 0.703 0.871 07 wramvmnmmam ness Ses 0.807 0.861 08 iii 0.886 0.854 3 1.000 0.852 10 _— 0.850 As the threshold increases, the probability of a correct prediction for high users decreases and that for low or inter- mediate users increases. For all users, the probability increases to 86.8 percent at a 0.5 threshold and declines to 85.0 percent at higher thresholds. Thus, a threshold of approxi- mately 0.5 will yield the highest probability of a correct prediction for all users, but this probability is dominated by the large number of low or intermediate users relative to the high users. The model for high use predicts high use somewhat better than the model for any use discussed previously. It is also important to examine the accuracy of the model for predicting high use or low or intermediate use. Table H presents the probability at a given threshold that a user classi- fied by the model as a high or low or intermediate user is actually a high or low or intermediate user. For example, for the threshold of 0.5, 65 percent of those classified as high users under the model are actually high users. Similarly, 88.2 percent of those classified as low or intermediate users at the 0.5 threshold are actually low or intermediate users. As before, the lowest predictive accuracy for high users occurs when every user is classified as a high user (i.e., a threshold of 0.0). For low or intermediate users, the lowest predictive accuracy occurs when all users are classified as low or inter- mediate users (i.e., a threshold of 1.0). The results in Tables G and H are improved over those in Tables C and D since the sensitivity of the test procedures at the various thresholds is improved, and the specificity of the test procedures at the various thresholds is improved, and the specificity remains high for most thresholds. But the sensitivity of the various tests remains low and indicates that improved fit of the model is desirable. As with the first model, a separate model was also constructed that includes the functional limitations variables, and therefore excludes anyone who died or was under 17 years of age. The logistic regression model with functional limitation added to the model (not shown here) again shows that functional limitation is significantly and positively related to high use. But the health status and age variables are no longer significant in the model with functional limitation present as a predictor. One possible source of this changed relationship may be attributable simply to the dropping of those who are under 17 years of age and those who died. But this finding also suggests that in this population functional limita- tion is correlated with age and with perceived health status. 31 Discussion The analyses in this report once again emphasize the importance of high users of medical care services. A small percent of the users who are high users of a medical service tend to account for a large percent of all such services consumed, for all three medical care services examined. This is most striking for hospital inpatient care. High users of hospital days constituted only 15 percent of those hospitalized and only 1.7 percent of the U.S. civilian noninstitutionalized population, yet they consumed 54 percent of all hospital days and generated 45 percent of all hospital inpatient charges in this population. High users of each of the three types of medical services share certain characteristics. The univariate analyses show that high users are more likely than low users to be older and poorer, to have lower health status, to have more reported conditions, and to have functional limitations. They are also more likely to have multiple public coverage. This is exemplified by finding that individuals under 65 years of age who are covered by two or more public programs are found to be among high users of hospital care 4.4 times as frequently as they are in the civilian noninstitutionalized population. When all of the variables that appear to be related to high hospital use in univariate analyses are entered simul- taneously in a multivariable regression model, significant association is found with respect to only three of the factors: Age, sex, and need as measured by reported health status and the presence of any one of five diagnoses. Although regression analysis does not address causality “and hence does not, in and of itself, permit discussion of the directionality of observed effects, when considered in the context of univariate analyses and the previous literature, the results are not only meaningful and clear but warrant some strong conclusions. These results show that the likeli- hood of being a high user of hospital inpatient services is increased not by the comprehensiveness of health care coverage but by need for health care, measured both by perceived health status and by the presence of disabling or life-threatening conditions. When individuals covered by Medicaid, two or more public programs, and by a combination of private and public sources are considered, they are found to have an increased probability of being hospitalized at all. However, when the probability of being a high hospital user is investigated, the coverage variables become less impor- tant, and in fact fail to attain statistical significance, and need-related factors dominate. Thus it is reasonable to suggest that those who are sicker are more likely to be eligible 32 for Medicaid and are more likely to be covered by two or more public programs. Hence, it is because they are sicker and not because of their coverage that they are more likely to be hospitalized and, once hospitalized, to be high users of hospital inpatient services. The findings with respect to health status and each of five diagnostic conditions (neoplasms, cardiovascular dis- eases, respiratory diseases, musculoskeletal diseases and in- juries and poisonings) are unambiguous. Poor health status and the presence of any of these five conditions at least doubles the probability that a person will be a high user of hospital inpatient services. The age effect found might well be an artifact resulting from the way in which the categorical age variables were created. The lower probability of being a higher user of hospital days, once hospitalized, for women may be attributa- ble to the fact that women, particularly in the age range of 17-44 years, are more likely to be hospitalized at all, principally for deliveries. Hence, they constitute a larger percent of the hospitalized population. The univariate analysis reinforces the dominant role of health status. Although the health status of only 12.9 percent of the civilian noninstitutionalized population as a whole, and only 11.0 percent of those who did not have a hospital inpatient episode in 1980, was reported to be fair or poor, 52.6 percent of those who were high users of hospital days had their health status rated as either fair or poor at the beginning of the survey year. Further, 57.7 percent of high users of hospital inpatient services reported having 6 or more separate medical conditions, compared with 15.2 percent for the entire civilian noninstitutionalized population, and 21.9 percent of those who were hospitalized for only 1 or 2 days during 1980. The findings regarding the relationship between level of disability and high use of hospital inpatient services reinforces this line of reasoning. In the civilian noninstitutionalized population eligible for the survey for the full year, 82.4 percent had no or minimal functional limitation, and, as one might expect, only 0.6 percent of these persons were high users of hospital inpatient services. At the other end of the disability scale, only 2.1 percent of the reference population had very severe functional limitation (Levels 7 and 8), but 20.2 percent of these persons were high users of hospital days, constituting 21.2 percent of all high users of hospital days. That is, they were found among high hospital users almost 10 times more frequently than in the civilian noninstitutionalized population. Further, the 0.4 percent of the reference population who died during the year comprised 12 percent of all high users of hospital days. It might also be noted that of those with no limitation, only 0.5 percent were high hospital users—a percent that increased to 22.5 for those most severely limited (Level 8) and to 54.2 percent of those who died during the year. The evidence is persuasive that high hospital inpatient care use is associated with severe illness, functional limitation, and death. The patterns of use by use level, as well as the differences between the characteristics of persons who are low versus high users of prescribed medications replicates those found for hospital inpatient services. With respect to ambulatory care, the patterns are some- what different. High users of ambulatory services are more similar to low users and to nonusers than is the case for hospital days and prescribed medicines with respect to age, family income, education of head of family, perceived health status, and activity limitations. These differences suggest that related, but different, dynamics result in high use for different categories of service. This is confirmed by the finding that only one-third of high users of hospital days were also high users of ambulatory care, while two-thirds of high users of ambulatory visits used no inpatient hospital care whatever, and only 12 percent were high users of hospital days. An important implication of these distinctions is that any gains in reducing high use in one category of service will not necessarily result in comparable gains in another category, and therefore achieving such reductions may require adopting, for each service category, approaches that are tai- lored to the distinctive characteristics of the category’s high users. On the other hand, nonusers and low users were found to have many similar characteristics, including the key ones of perceived health status and functional limitation. For exam- ple, with respect to ambulatory care, 7.4 percent of nonusers and 6.5 percent of low users had their health status reported as fair or poor. Among persons for whom data on functional limitation are available, the civilian noninstitutionalized popu- lation 17 years of age or over who survived the year, 88.5 of the nonusers and 88.1 percent of the low users had no reported functional limitation. When nonuse and low use of hospital inpatient services by insurance status is considered a difference emerges among persons under 65 years of age who had no health care coverage of any kind during the entire year. These persons are found much more frequently in the nonuser group (9.2 percent) than in the low hospital use category (3.8 percent). In fact, the proportion of persons who were not hospitalized during the year is the highest in this group of persons without any health care coverage, 95 percent, as is the proportion who were hospitalized for only one or two days, 1.0 percent. This might well indicate that the lack of health care coverage poses a barrier to the use of inpatient hospital services. It is worthwhile noting, however, that family income does not differentiate between persons who are nonusers and low users of hospital inpatient services. When arrayed by relative poverty status, between 2.1 percent (200-499 percent of poverty level) and 2.9 percent (below 200 percent poverty level) of persons under 65 years of age are low users of hospital days. At the same relative poverty levels, 89.5 percent and 86.1 percent of persons did not have a hospitalization episode during the year. In fact, the percent of persons with no hospitalization is effectively the same ‘(between 89.4 and 90.6 percent) at all adjusted income levels above 200 percent of the poverty level. The findings on the characteristics of high users of medical care services are sobering. They indicate that 54.3 percent of hospital days, 32.3 percent of ambulatory visits, and 32.9 percent of prescribed medications were consumed by a very small percent of the U.S. civilian noninstitutionalized popula- tion. They also indicate that these high users of medical care services are predominantly sick, functionally limited in their activities, or dying. High use of services, and espe- cially of hospital days, which are the most expensive compo- nent of medical care, does not appear to be associated with health care coverage. Although the data used for these analyses did not permit the exploration of the potential effects on level of use of medical resource availabilities and institutional structures, such as the effects of Health Maintenance Organiza- tions, the results indicate that standard cost containment ap- proaches, such as increased patient cost sharing, are not likely to have significant impacts on that large percent of expenditures generated by the small percent of high users. These findings indicate that high levels of use result from serious illness, severe disability, and death. If this is true, long-range efforts to contain costs will have to attempt to reduce the incidence cf the conditions that lead to high use of medical care resources. Alternative means of treatment and new organizational structures will have to be considered in order to reduce the costs of management of conditions that cannot be prevented. 33 References Anderson, D. W., and McLaurin, R. L., eds.: Report on the national head and spinal cord injury survey. J. Neurosurg. 53: Suppl., Nov. 1980. Anderson, G. and Knickman, J.: Patterns of expenditures among high utilizers of medical care services: The experience of Medicare beneficiaries from 1974 to 1977. Med. Care 22(2): 143-149, Feb. 1984. Birnbaum, H.: A National Profile of Catastrophic Illness. DHEW Pub. No. (PHS) 78-3201. National Center for Health Services Research, Public Health Service. Hyattsville, May 1978a. Bimbaum, H.: The Cost of Catastrophic lliness. Lexington, Mass. D.C. Health and Company, 1978b. Bloom, B. and Kissick, P.: Home and hospital cost of terminal illness. Med. Care, 318(5): 560-564, May 1980. Budetti, P., McManus, P., Barrand, M., et al.: The costs and effectiveness of neonatal intensive care. The Implications of Cost-Ef- Jfectiveness Analysis of Medical Technology. Background Paper 2, Case Study 10. Office of Technology Assessment, Congress of the United States. Washington, Aug. 1981. Cancer Care, Inc.: The Impact, Costs, and Consequences of Catas- trophic Illness on Patients and Families. New York. Cancercare, Inc. and the National Cancer Foundation, 1973. Casady, R. J.: The survey design and estimation methodology for the NMCUES. Proceedings of the 19th National Meeting of the Public Health Conference on Records and Statistics. DHHS Pub. No. (PHS) 81-1214. National Center for Health Statistics, Public Health Service. Hyattsville, Md., 1983, pp. 181-185. Congressional Budget Office: Catastrophic Health Insurance. Washington. U.S. Government Printing Office, 1977. Congressional Budget Office: Catastrophic Medical Expenses: Pat- terns in the Non-elderly, Non-poor Population. Washington. U.S. Government Printing Office, Dec. 1982. Cullen, D. J., Ferrara, L. C., Briggs, B. A., et al.: Survival, hospitalization charges and follow-up results in critically ill patients. N. Engl. J. Med. 294(18): 982-287, April 29, 1976. Detsky, A. S., Stricker, S. C., Mulley, A. G., et al.: Prognosis, survival, and the expenditure of hospital resources for patients in an intensive-care unit. N. Engl. J. Med. 305(12): 667-672, Sept. 17, 1981, Eldred, C. A., Woolsey, T. D., Simmons, W. R., et al.: A Report of the Survey of Intracranial Neoplasms. Contract No. No—1-NS—4— 2336. Rockville, Md. Westat, Inc., 1977. Forthofer, R. N., Lairson, D. R., and Glasser, J. H.: Catastrophic health insurance and HMO's. Soc. Sci. Med. 16(20): 1775-1779, 1982, 34 Gibbs, J., and Newman, J.: Study of Health Services Used and Costs Incurred During the Last Six Months of a Terminal Illness. Chicago. Blue Cross and Blue Shield Association. Nov. 1982. Kish, L.: Survey Sampling. New York. J. W. Wiley and Sons, Inc., 1965. Kobrinski, E. J., and Matteson, A. L.: Characteristics of high cost treatment in acute care facilities. Inquiry 18(2): 179-184, Summer 1981. Lepkowski, J. M., Landis, J. R., Parsons, P. E., et al.: A statistical methodology for analyzing data from a complex sample survey, the National Medical Care Utilization and Expenditure Survey. National Medical Care Utilization and Expenditure Survey. Series A. National Center for Health Statistics, Public Health Service. Hyattsville, Md. To be published. Lepkowski, J. M., Stehouwer, S., and Landis, J. R.: Strategies for the analysis of imputed data in a sample survey. Proceedings of the Section on Survey Research Methods, American Statistical Association, 1984. Lubitz, J., and Prihoda, R.: The use and costs of Medicare services in the last two years of life. Health Care Fin. Rev. 5(3): 117-131, Spring 1984. McCall, N.; Utilization and costs of Medicare services by benefi- ciaries in their last year of life. Med. Care 22(4): 329-342, Apr. 1984. Parsons, P. E., Lichtenstein, R., Berki, S. E., et al.: The costs of illness, United States, 1980. National Medical Care Utilization and Expenditure Survey, Series C. National Center for Health Statis- tics, Public Health Service. Hyattsville, Md. To be published. Schroeder, S. A., Showstack, J. A., and Roberts, H. E.: Frequency and clinical description of high cost patients in 17 acute care hospitals. N. Engl. J. Med. 300(23): 1306--1309, June 7, 1979. Scitovsky, A. A.: The high cost of dying: what do the data show? Milbank Mem. Fund Q. 62(4): 591-608, Fall 1984. Trapnell, G. R.: The rising cost of catastrophic illness. National Center for Health Services Research, DHEW, Dec. 1977. Webb, S., Berzins, E., Wingardner, T. S., et al.: First year hospitali- zation costs for the spinal cord injured patient. Paraplegia, 15: 311-318, 1977. Weinfield, F. D., ed.: The national survey of stroke. Stroke 12(2): Suppl. No. 1, Mar.—Apr. 1981. Zook, C. J., and Moore, F. D.: High-cost users of medical care. N.Engl.J. Med. 302(18) 996-1002, May 1, 1980. List of Detailed Tables 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. Numbers and percent distributions, and means for service units used and charges for service, by type of service and level of use: United States, 1980 . . . . . . . «oc tt iii Numbers and percent distributions of persons by race, sex, and age, according to level of hospital use: United Silos, 1980 : cc vs uss ramps ur rsa waani LRAUBREDHL IIIA B DEFLATE Ts sw E Numbers and percent distributions of persons by race, sex, and age, according to level of ambulatory care use: United States, 1980 . . . . «ot vc vi ee ee eee eee eee eee eee Numbers and percent distributions of persons by race, sex, and age, according to level of prescribed medicine use: United States, 1980 . . . . «oo tit i eee ee eee eee eee ee Numbers and percent distributions of persons by family income, poverty status, and education of head of family, according to level of hospital use: United States, 1980. . . . . . . ..... o.oo Numbers and percent distributions of persons by family income, poverty status, and education of head of family, according to level of ambulatory care use: United States, 1980 . . . . . ............vee een Numbers and percent distributions of persons by family income, poverty status, and education of head of family, according to level of prescribed medicine use: United States, 1980. . . . . ..................... Number and percent distribution of persons by region, according to level of hospital use: United States, 1980 . . . . «+ isis hs sa Ere PE rea ma msm SA aE SRE A EEE a Ha an Number and percent distribution of persons by region, according to level of ambulatory care use: United SALES, 1980 . . . . «i i tree eee aaa eae eee esas Number and percent distribution of persons by region, according to level of prescribed medicine use: United SUAS, 1980 . . «vv nt ss rE EE Ef At Ae EAM mer ras AE EAS EAD SME AEE ew wwe aw Number and percent distribution of persons under 65 years of age by type of health care coverage, according to level of hospital use: United States, 1980 . . . . . . . . . . o.oo Number and percent distribution of persons under 65 years of age by type of health care coverage, according to level of ambulatory care use: United States, 1980 . . . . . . . . ... oc. Number and percent distribution of persons under 65 years of age by health care coverage, according to level of prescribed medicine use: United States, 1980 . . . . . .. ...... oe Number and percent distribution of persons 65 years of age and over by type of Medicare coverage, according to level of hospital use: United States, 1980 . . . . . . . . ... «oe Number and percent distribution of persons 65 years of age and over by type of Medicare coverage, according to level of ambulatory care use: United States, 1980 . . . . . . . . .... co c vii e Number and percent distribution of persons 65 years of age and over by type of Medicare coverage, according to level of prescribed medicine use: United States, 1980 . . . . ........... nnn Number and percent distribution of persons by perceived health status, according to level of hospital use: United SAEs, 1980 . . . oo i i ee ee eee eee eee eee eee eee eae eee eee Number and percent distribution of persons by perceived health status, according to level of ambulatory care use: United States, 1980 . . . . . . . «cc tt thie eee eee ae eae eae eee Number and percent distribution of persons by perceived health status, according to level of prescribed medicine use: United States, 1980 . . . . «oo i ti eee eee eee eee eee eee eee eee Number and percent distribution of persons 17 years of age and over and living at the end of the survey period by functional limitation score, according to level of hospital use: United States, 1980 . . ............. 21. 28. 24. 25, 26. 27. 28. 30. 31. 32. 36 Number and percent distribution of persons 17 years of age and over and living at the end of the survey period by functional limitation score, according to level of ambulatory care use: United States, 1980 . ......... Number and percent distribution of persons 17 years of age and over and living at the end of the survey period by functional limitation score, according to level of prescribed medicine use: United States, 1980 . ....... Number and percent distribution of persons by number of unique conditions occurring during the year, according to level of hospital use: United States, 1980 . . . . .. . . .. ... vi Number and percent distribution of persons by number of unique conditions occurring during the year, according to level of ambulatory care use: United States, 1980 . . . . . . . . . . . . 0 vv vii Number and percent distribution of persons by number of unique conditions occurring during the year, according to level of prescribed medicine use: United States, 1980 . . . . .. .. ......... iii... Number and percent distribution of persons with specified diagnoses associated with hospital admissions by diagnostic category, according to level of hospital use: United States, 1980 . . . . . . . . .. .. .. vv vv... Number and percent distribution of persons by whether surgery was performed during the stay, according to level of hospital use: United States, 1980 . . . . . . ......... vier Number and percent distribution of persons by level of ambulatory care use, according to level of hospital use: United States, 1980 . . . . . . . . iit Number and percent distribution of persons with hospital use by number of hospital admissions, according to level of hospital use: United States, 1980 . . . . . . . . ... vr em Number and percent distributions of persons by survival status, according to level of hospital use: United States, 1980 . . . LL Number and percent distributions of persons by survival status, according to level of ambulatory care use: United + Number and percent distributions of persons by survival status, according to level of prescribed medicine use: United States, 1980 . . . . . LL. LE Table 1 Numbers and percent distributions and means for service units used and charges for service, by type of service and level of use: United States, 1980 Percent distribution Service unit used Charges for service Population Number Amount Mean per in in Percent, Per in Percent. Per unit, of Type of service and level of use thousands All persons Users thousands distribulion capita millions distribution capita service Hospital days Total... .cinsmuimranimenms me ame wnn wns 222,824 100.0 100.0 271,417 100.0 1.2 $87,138 100.0 $391 $321 Level of use 1 1 J TY Ep Ep Epp gE 197,283 88.5 . ae -~ — 1,057 1.2 -~ we | 5X ST 5,242 2.4 20.5 9,120 3.4 1.7 5,251 6.0 1,002 576 Intermediate ..... covers srmrcnernrnsnns 16,462 74 64.5 114,770 42.3 7.0 41,473 47.6 2,519 361 High conimmimsrmrmmrmmmmrmonr ne CLIHERTEHE 3,837 1.7 15.0 147,557 54.3 38.5 39,357 45.2 10,257 267 Ambulatory care visits gi 71 + 1 JPN ppg pup ppg pg SE A 222,821 100.0 100.0 1,150,642 100.0 5.2 37,504 100.0 168 33 Level of use 0 ce comm oon owe ck BEE HEE ER 0 BE 8 8 Oa eww 46,716 21.0 -~ we FF a 20s sv as LOW i siursmpinsmpimonmemmsmemmuemssbi®sms 37,921 17.0 21.5 38,665 3.3 1.0 1,344 3.6 35 35 Intermediate . ...... connie 128,163 57.5 72.8 740,724 64.4 5.8 25,394 67.7 198 34 HIZH ....c imine sssmsnmenscmemammuememnes 10,024 4.5 Hh.7 371,253 32.3 37.0 10,766 28.7 1,074 29 Prescribed medicine acquisitions Total . oot eee 222,824 100.0 100.0 1,026,767 100.0 41.6 7,831 100.0 35 8 Level of use SS 83,814 37.6 i» .. . . . . i» . BOW wv env mss ms maa as Mme sie mss sassse wenn 30,492 13.7 21.9 30,954 3.0 1.0 204 2.6 7 7 Intermediate . . ........ iii 100,345 45.0 72.2 658,061 64.1 6.6 4,917 63.2 19 8 High ..cooveseririinrssnssnsatrcrveenvens 8,173 3.7 59 337,752 32.9 41.3 2,680 34.2 328 8 Represents charges for persons who were counted as having at least. 1 hospital stay but never were in the hospital overnight. Table 2 Numbers and percent distributions of persons by race, sex, and age, according to level of hospital use: United States, 1980 Level of hospital use Race, sex, and age Total J Low Intermediate High Race Number in thousands Allraces ................. 222,824 197,283 5,242 16,462 3,837 Black oi nnintinsmmensmmens cose 26,046 22,952 599 1,978 516 White and other ................ 196.779 174,330 4,644 14,484 3,321 Sex and age Both sexes, all ages ........ 222,824 197,283 5,242 16,462 3.837 Under 17 years ................. 61.575 57,050 1.681 2,713 131 17-24 years .................... 32,886 29,031 993 2,647 215 25-34'years .................... 35,827 31,493 970 2,991 373 35-44 years ................ .... 25,489 23.093 476 1,631 389 45-54 years .................... 22,443 19,797 408 1,737 500 55-64 years .................... 21.135 18,404 286 1,797 648 65-T4years .................... 15,165 12,190 240 1,927 808 75 years and over ............... 8,305 6,224 189 1,119 77 Male, allages ............... 107,481 96,929 2,185 6,404 1,963 Under 17 years ................. 31,585 29,091 913 1,521 60 IT=28 YRALE ovis cma nme mnmans 15,7582 14,647 295 G88 123 25-34 years .................... 17,506 16.288 245 761 211 35-44 years .................... 12,318 11,319 249 579 170 45-54 years .................... 10,859 9.697 154 711 297 55-64 years .................... 9,970 8,618 133 834 386 65-74 years .................... 6,486 5.012 103 987 383 75 years and over ............... 3,006 2,257 93 323 333 Female, all ages ............. 115,344 100,354 3,058 10,058 1,874 Under 17 years ................. 29,990 27.959 768 1,192 71 17-24 years .................... 17,134 14,384 G98 1,959 92 25-34years .................... 18,321 15,205 724 2,230 162 35-44 years .................... 13,171 11,773 227 952 219 45-54 years .................... 11,684 10,101 255 1,026 203 55-64 years ................. ... 11,165 9.786 153 964 262 65-74 vears .................... 8.679 7.178 137 940 424 75 years and over ............... 5,299 3.968 96 796 440 38 Table 2 — continued Numbers and percent distributions of persons by race, sex, and age, according to level of hospital use: United States, 1980 Level of hospital use Race, sex, and age Total 0 Low Intermediate High Race Percent distribution AU YASS ou wmems ems s mamma 100.0 100.0 100.0 100.0 100.0 Black ...iwsms smsamenmewmemmens 11.7 11.6 11.4 12.0 13.5 White and other ................ 88.3 88.4 88.6 88.0 86.5 Sex and age Both sexes, all ages ........ 100.0 100.0 100.0 100.0 100.0 Under 17 Vears ..........c..ovouen 27.6 28.9 32.1 16.5 3.4 17-24 YOATS sou enmemmompoms sos vor 14.8 14.7 18.9 16.1 5.6 25-34 VEATS . viii 16.1 16.0 18.5 18.2 9.7 35-04 YEATES vv vs isis isismusomns®s sn 11.4 * 117 9.1 9.3 10.1 45-D YEATS .....eovvvun aren 10.1 10.0 7.8 10.6 13.0 55-64 VEArS ... viene 9.5 9.3 5.5 10.9 16.9 B5-TA YEATS ....covrvrrnvrnnrnns 6.8 6.2 4.6 11.7 21.1 75 vears and over ............... 3.7 3.2 3.6 6.8 20.1 Male, allages ............... 100.0 100.0 100.0 100.0 100.0 Under 17 vears ................. 29.4 30.0 41.8 23.8 3.0 17-24 Vears .......coenvnnrtrunnns 14.7 15.1 13.5 10.7 6.3 25-34 VATS .. iii 16.3 16.8 11.2 11.9 10.8 35-44 Years .........0 naan 11.5 11.7 11.4 9.0 8.7 45-B4 Years ....c vv nvm v ney 10.1 10.0 7.0 11.1 15.1 55-64 YEArS ... cannes 9.3 8.9 6.1 13.0 19.6 B5-T4 Years ........covaseenervunn 6.0 5.2 4.7 15.4 19.5 75 vears and OVer ............... 2.8 2.3 4.3 5.1 17.0 Female, all ages ............. 100.0 100.0 100.0 100.0 100.0 Under 17years ........ vues eees 26.0 279 25.1 11.9 3.8 17-24 years ......ooonnrriunn 14.9 14.3 22.8 19.5 4.9 25-34 VEArS .... cities 15.9 15.1 23.7 22.2 8.6 35-44 years ........ ann 11.4 17 7.4 9.5 11.9 45-54 VATS .. iia 10.0 10.1 8.3 10.2 10.8 55-84 YEArS ... evra 9.7 9.8 5.0 9.6 14.0 B5=T4 YRATS +. uvvr rvs rans nns 7.5 7.2 4.5 9.3 22.6 75 vears and over ............... 4.6 4.0 3.2 7.9 23.5 Table 3 Numbers and percent distributions of persons by race, sex, and age, according to level of ambulatory care use: United States, 1980 Level of ambulatory care use Race, sex, and age Total 0 Low Intermediate High Race Number in thousands AUTAES ivvviinnsnsmmsnne 222,824 46,716 37,921 128,163 10.024 Black .............. ci 26,046 7,120 4,852 13,196 878 White and other ................ 196,779 39,596 33,069 114,968 9,146 Sex and age Both sexes, all ages ........ 222,824 46,716 37,921 128,163 10,024 Under 17 years ................. 61,575 12,741 12,504 34,911 1,420 17-24 YBAYE vn nswminsantitimnmemn 32,886 7,220 6,418 18,281 967 25-84 years ..........0. i... 35,827 7,954 6,373 20.038 1,463 35-44 years .................... 25,489 6,329 4,377 13,325 1,458 45-54 years ........... 00... 22,443 5,342 3,326 12,621 1,253 55-64 years .................... 21,135 3,673 2,781 13,543 1,237 65-74years ...............0..... 15,165 2,514 1,466 9.819 1,365 75 years andover ............... 8,305 1,042 676 5,725 862 Male, allages ............... 107,481 27,573 19,710 56,431 3,766 Under 17 years ................. 31,585 6,435 6,459 17,921 769 17-24 years .................... 15,752 4,777 3,625 7,009 342 25-34 years ............uiiun... 17,506 5,520 3,555 7,888 543 35-44 years .............. Bik 12,318 3,971 2,048 5,809 490 45-b4 years ............. a... 10,859 3,127 1.730 5,513 489 55-64 years .................... 9,970 1,967 1,354 6,268 381 65-74 years .................... 6,486 1,277 647 4,009 552 75 years and over ............... 3.006 499 292 2,014 201 Female, allages ............. 115,344 19,143 18,211 71,732 6,258 Under 17 years ................. 29,990 6,305 6,045 16,989 651 17-24 years ........ocnnnnvnnnnn 17,134 2,443 2,794 11,272 625 25-34years ..........uuun. 18,321 2,434 2,818 12,149 920 35-44 years .................... 13,171 2,359 2,329 7,516 967 45-54 years .............u...... 11,684 2,215 1,596 7,009 765 55-64 years .................... 11,165 1,607 1,427 7.275 857 65-T4 years .................... 8,679 1,238 819 5,809 813 75 years and over ............... 5,299 543 383 3.712 661 Table 3 — continued Numbers and percent distributions of persons by race, sex, and age, according to level of ambulatory care use: United States, 1980 Level of ambulatory care use Race, sex, and age Total 0 Low Intermediate High Race Percent distribution Allraces ....... oven 100.0 100.0 100.0 100.0 100.0 Black .,. viv inremurvrsansvrrses 11.3 15.2 12.8 10.3 8.8 White and other ................ 88.3 84.8 87.2 89.7 91.2 Sex and age Both sexes, all ages ........ 100.0 100.0 100.0 100.0 100.0 Under 17 years ........ooeeeevos 27.6 27.3 33.0 27.2 14.2 17-24 YEATES «vv vos 14.8 15.5 16.9 14.3 9.6 25-34 years .......cc eee 16.1 17.0 16.8 15.6 14.6 35-44 VeArs uae 11.4 13.5 115 10.4 14.5 45-54 VEArS iia 10.1 11.4 8.8 9.8 12.5 55-64 YEATS viva 9.5 7.6 7.8 10.6 12.3 B5-T4 VEATS ovr nas 6.8 5.4 3.9 20? 13.6 75 vears and OVer ..........eeee 3.7 2.2 1.8 4.5 8.6 Male, all ages ............... 100.0 100.0 100.0 100.0 100.0 Under 17 years ......... coven 29.4 23.3 32.8 31.8 20.4 17-24 VEArS vv vv nnn 14.7 17.3 18.4 12.4 9.1 25-34 Years ... iia 16.3 20.0 18.0 14.0 14.4 35-44 vears ...... coe 11.5 14.4 10.4 10.3 13.0 45-54 vears ...... enna 10.1 11.4 8.8 9.8 13.0 55-64 VEATS . vivir 9.3 7.1 6.9 11.3 10.1 B5-T4 YEArS . vv v tries G.0 4.6 3.3 73 14.7 75 years and OVer ..........ooene 2.8 1.8 1.5 3.5 5.3 Female, all ages . ............ 100.0 100.0 100.0 100.0 100.0 Under 17 years .........oovveeos 26.0 32.9 33.2 23.7 10.4 17-24 Years .......coona nes 14.9 12.8 15.3 15.7 10.0 25-34 years ........ iia 15.9 12.7 15.5 16.9 14.7 35-44 YEArS . o.oo 11.4 12.3 12.8 10.5 15.5 45-B4 vears ...... ieee 10.0 11.6 8.8 9.8 12,2 55-64 VEArS . . vivo 9.7 8.4 7.8 10.1 13.7 B5-T4 VEArS .. ov v verte 7.5 6.5 4.5 8.1 13.0 75 years and OVer .............o 4.6 2.8 2.1 5.2 10.6 41 Table 4 Numbers and percent distributions of persons by race, sex, and age, according to level of prescribed medicine use: United States, 1980 Level of prescribed medicine use Race, sex, and age Total 0 Low Intermediate High Race Number in thousands Allraces ................. 222,824 83,814 30,492 100,345 8,173 Black «oisumsmsmmsnaiscsomennonen 26,046 12,172 3.726 9,373 774 White and other ................ 196,779 71,642 26,766 90,972 7.399 Sex and age Both sexes, all ages ........ 222,824 83,814 30,492 100,345 8,173 Under 17 years ................. 61,575 27,241 10,635 23,518 182 17-24 years .................... 32,886 13,568 5,469 13.822 26 25-34 years ............0 i... 35,827 13,747 5,458 16,246 375 35-44 years .................... 25,489 10,158 3,526 11,235 569 45-64 years .................... 22,443 8,336 2,168 10,607 1,332 55-64 years .................... 21,135 5,871 1,864 11,325 2,075 65-74 years .................... 15,165 3,596 983 8,584 2,001 75 vears and over ............... 8.305 1,296 388 5,008 1,612 Male, all ages ............... 107,481 47,729 15,648 41,131 2,973 Under 17 years ................. 31.585 14.056 5,309 12,105 115 17-24 vears .................... 15,752 8,469 2,712 4.545 26 25-84 vears .................... 17,506 8,841 2,751 5,745 169 35-44 years .................... 12,318 5,780 1.937 4,438 162 45-64 years .................... 10,859 4,832 1,204 4,289 534 55-64 years .................... 9,970 3,276 1,070 4918 706 65-74 vears .................... 6,486 1,888 532 3,373 692 75 vears and over ............... 3,006 586 133 1.718 569 Female, all ages ............. 115,344 36,086 14,844 59,214 5,200 Under 17 years ................. 29,990 13,185 5,326 11,413 66 17-24 years .................... 17,134 5,100 2,756 9,278 rw 25-34 years .................... 18.321 4,906 2,707 10,502 206 35-44 years .................... 13,171 4,378 1,689 6,797 407 45-54 years .................... 11,584 3,504 964 6,318 798 55-64 years .................... 11,165 2,595 795 6.406 1,369 65-74 vears .................... 8,679 1,708 451 5,211 1,309 75 vears and over ............... 5,299 710 256 3.290 1.043 42 Table 4 — continued Numbers and percent distributions of persons by race, sex, and age, according to level of prescribed medicine use: United States, 1980 Level of prescribed medicine use Race, sex, and age Total 0 Low Intermediate High Race Percent distribution All races ...oovrrnevavnnas 100.0 100.0 100.0 100.0 100.0 Black +... viii iii 11.7 14.5 12.2 9.3 9.5 White and other ........... 000 88.3 85.5 87.8 90.7 90.5 Sex and age Both sexes, all ages ........ 100.0 100.0 100.0 100.0 100.0 Under 17 years ..........cooouee. 27.6 32.5 34.9 23.4 2.2 17-24 y€8r8 ......covv iii 14.8 16.2 17.9 13.8 0.3 25-84 Y@ArS . . iia 16.1 16.4 17.9 16.2 4.6 85-44 years ........o0nninn 11.4 12.1 11.6 11.2 7.0 45-54 years ......... coun 10.1 9.9 7.1 10.6 16.3 55-64 years ..........c00einnnan 9.5 7.0 6.1 11.3 25.4 B5-T4 Years .......covvvvsennnns 6.8 4.3 3.2 8.6 24.5 75 years and OVer ............00- 3.7 1.5 1.2 5.0 19.8 Male, all ages ............... 100.0 00.0 100.0 100.0 100.0 Under 17 years ........c.cooeeens 29.4 29.4 33.9 29.4 3.9 17-24 years .......ccontiunenaan 14.7 17.7 17.3 11.0 0.9 25-34 YEArS .. viii 16.3 18.5 17.6 14.0 5.7 85-44 years ........ cnn 11.5 12.1 12.4 10.8 5.4 45-54 years ..........cenenn 10.1 10.1 7.7 10.4 18.0 55-64 years ..........c00 cern 9.3 6.9 6.8 12.0 23.7 BD-T4 years ....... evn r eens 6.0 4.0 3.4 8.2 23.3 75 years and over ............... 2.8 1.2 0.8 4.2 19.1 Female, allages ............. 100.0 100.0 100.0 100.0 100.0 Under 17years ..........cocuun- 26.0 36.5 35.9 19.3 1.3 17-24 Years ....... coven ennonnns 14.9 14.1 18.6 15.7 - 25-34 years .........na cae 15.9 13.6 18.2 17.7 4.0 35-44 Years .........oaneenn 11.4 12.1 10.7 11.5 7.8 45-54 years .........ee een 10.0 9.7 6.5 10.7 15.3 55-64 years ..........0a enn 9.7 7.2 5.4 10.8 26.3 B5-T4 Years .....onvvvassecnssss 7.5 4.7 3.0 8.8 25.2 75 years and Over ............... 4.6 2.0 1.8 5.6 20.0 Table 5 Numbers and percent distributions of persons by family income, poverty status, and education of head of family, according to level of hospital use: United States, 1980 Level of hospital use Family income, poverty status, and education of head of family Total 0 Low Intermediate High Family income Number in thousands Less than $5,000 .................. 16,225 13,165 664 1,704 693 $5,000-14,999 ................ 58,157 50,116 1,471 5,101 1,470 $15,000-34,999 ................... 103,400 92,908 2,123 7,219 1,149 $35,000 ormore ........... 0... 45,043 41,094 985 2,438 526 Poverty status Below 200 percent of poverty level .... 70,023 60,257 2,021 5,947 1,798 200-499 percent of poverty level ...... 115,466 103,352 2,371 8,179 1,564 500-699 percent of poverty level ...... 23,872 21,634 549 1,445 243 700 percent of poverty level or more ... 13,464 12,040 302 891 232 Education of head of family! None or some elementary ........... 36,837 31,568 747 3,443 1,078 Some high school .................. 35,152 30,745 813 2,787 808 High school graduate ............... 78,063 69,472 1,944 5,581 1,087 Some college ...................... 34,849 31,303 726 2,307 513 College graduate .................. 37,784 34,056 1,012 2,345 371 Family income Percent distribution Total .................. 0... 100.0 100.0 100.0 100.0 100.0 Less than 85,000 .................. 7.8 6.7 12.7 10.4 18.1 $5,000-14,999 .................... 26.1 25.4 28.1 31.0 38.3 §15,000-34999 ................... 46.4 47.1 40.5 43.9 29.9 $35,000 ormore ................... 20.2 20.8 18.8 14.8 18.7 Poverty status Total ....................... 100.0 100.0 100.0 100.0 100.0 Below 200 percent of poverty level .... 31.4 30.5 38.5 36.1 46.9 200-499 percent of poverty level . ..... 51.8 52.4 45.2 49.7 40.8 500-699 percent of poverty level ...... 10.7 11.0 10.6 8.8 6.3 700 percent of poverty level or more . . . 6.0 6.1 5.8 5.4 6.0 Education of head of family! TOURL 54 ims wmammuins cms snsmmsn 100.0 100.0 100.0 100.0 100.0 None or some elementary ........... 16.5 16.0 14.3 20.9 28.1 Some high school .................. 15.8 15.6 15.5 16.9 21.1 High school graduate ............... 35.0 35.2 37.1 33.9 27.8 Somecollege . ..................... 15.6 15.9 13.9 14.0 13.4 College graduate .................. 17.0 17.3 19.3 14.2 9.7 ncludes only families with heads 17 years of age and over. Table 6 Numbers and percent distributions of persons by family income, poverty status, and education of head of family, according to level of ambulatory care use: United States, 1980 Family income, poverty status, Level of ambulatory care use and education of head of family Total 0 Low Intermediate High Family income Number in thousands Less than $5,000 .................. 16,225 2,858 1,984 10,254 1,130 $5,000-14,999 ..........00iiinnn 58,157 12,786 9,640 32,795 2,936 $15,000-84,999 .......... cee 103,400 21,719 18,472 59,319 3,890 $35,000 00 MOTE ..... hein 45,043 9,353 7,826 25,795 2,069 Poverty status Below 200 percent of poverty level .... 70,023 16,293 11,448 38,947 3,336 200-499 percent of poverty level ...... 115,466 23,526 20,517 66,902 4,520 500-699 percent of poverty level ...... 23,872 4,654 3,824 14,161 1,233 700 percent of poverty level or more . .. 13,464 2,243 2,133 8,153 935 Education of head of family! None or some elementary ........... 36,837 9,510 5,442 20,086 1,799 Some highschool .................. 35,152 8,820 5,990 19,181 1,160 High school graduate ............... 78,063 15,452 14,335 44,882 3,394 Somecollege ...........ciiiiiinn 34,849 6,650 5,536 20,953 1,710 College graduate .................. 37,784 6,257 6,577 22,989 1,961 Family income Percent distribution Total ...icusnsnsrmmsamswrmys 100.0 100.0 100.0 100.0 100.0 Less than $5,000 .................. 7.3 6.1 5.2 8.0 11.3 $5,000-14,999 .........i iii 26.1 27.4 25.4 25.6 29.3 $15,000-84,999 ............00enn 46.4 46.5 48.7 46.3 38.8 $35,000 0rmore .......uueeen ATED 20.2 20.0 20.6 20.1 20.6 Poverty status Total woivssvimpmmenmemmwenons 100.0 100.0 100.0 100.0 100.0 Below 200 percent of poverty level .... 31.4 34.9 30.2 30.4 33.3 200-499 percent of poverty level ...... 51.8 50.4 54.1 52.2 45.1 500-699 percent of poverty level ...... 10.7 10.0 10.1 11.0 12.3 700 percent of poverty level or more ... 6.0 4.8 5.6 6.4 9.3 Education of head of family! TOA ...ovsvnimpsmmsmsmmonsy 100.0 100.0 100.0 100.0 100.0 None or some elementary ........... 16.5 20.4 14.4 15.7 17.9 Some highschool .................. 15.8 18.9 15.8 15.0 11.6 High school graduate ............... 35.0 33.1 37.8 35.0 33.9 Somecollege . ......... cco 15.6 14.2 14.6 16.3 17.1 College graduate ............ooun. 17.0 13.4 17.3 17.9 19.6 lincludes families with heads 17 years of age and over. 45 Table 7 Numbers and percent distributions of persons by family income, poverty status, and education of head of family according to level of prescribed medicine use: United States, 1980 Level of prescribed medicine use Family income, poverty status, and education of head of family Total 0 Low Intermediate High Family income Number in thousands Less than $5,000 .................. 16,225 4,805 1,766 8,116 1,639 $5,000-14,999 .................... 58,157 21,331 6,762 26,839 3,225 $15,000-34,999 ............ 103,400 40,108 15,598 45,328 2,366 $35,000 ormore ................... 45,043 17,671 6,366 20,063 1,043 Poverty status Below 200 percent of poverty level . ... 70,023 7,534 8,790 30,031 3.669 200-499 percent of poverty level ...... 115,466 43,448 16,564 51,939 3,514 500-699 percent of poverty level .. .... 23,872 8,344 3,392 11,610 525 700 percent of poverty level or more . . . 13,464 4,488 1,746 6.765 465 Education of head of family! None or some elementary ........... 36,837 14,126 4,044 15,778 2,889 Some high school ............... ... 35,152 14,174 4,532 14,957 1,488 High school graduate ............... 78,063 29,602 11,002 35,345 2,114 Some college . ..................... 34,849 12,265 5,272 16,467 846 College graduate .................. 37,784 13,576 5,621 17,751 836 Family income Percent distribution TOBA 4 vis 5 i wow cmon 500 0 en 9 050 0 100.0 100.0 100.0 100.0 100.0 Less than $5,000 .................. 7.3 5.7 5.8 8.1 18.8 $5,000-14,999 .................... 26.1 25.5 22.2 26.7 39.5 $15,000-34,999 ................... 46.4 47.9 51.2 45.2 29.0 $35,000 ormore ................... 20.2 21.0 20.9 20.0 12.8 Poverty status Total ....................... 100.0 100.0 100.0 100.0 100.0 Below 200 percent of poverty level . ... 31.4 32.9 28.8 29.9 44.9 200-499 percent of poverty level . ..... 51.8 51.8 54.3 51.8 43.0 500-699 percent of poverty level . ..... 10.7 10.0 11.1 11.6 6.4 700 percent of poverty level or more . .. 6.0 5.4 5.7 6.7 5.7 Education of head of family! Total «ius iwrmunmeswonn ones 100.0 100.0 1100.0 100.0 100.0 None or some elementary ........... 16.5 16.9 13.3 15.7 35.3 Some high school .................. 15.8 16.9 14.9 14.9 18.2 High school graduate ............... 35.0 35.3 36.1 35.2 25.9 Some college . ..................... 15.6 14.6 17.3 16.4 10.3 College graduate .................. 17.0 16.2 18.4 17.3 10.2 Uncludes families with heads 17 years of age and over. Table 8 Number and percent distribution of persons by region, according to level of hospital use: United States, 1980 Level of hospital use Region Total 0 Low Intermediate High Number in thousands Northeast ...........ovvuvuenn 46,899 41,788 930 3,259 921 North Central ... ceoncvvvms ens 59,257 51,950 1,447 4,782 1,079 South .......cciiiiinrnnennns 69,475 61,059 1,756 5,635 1,126 WESL vine msimpinimmsmusams es 47,194 42,486 1,110 2,886 711 Percent distribution Total +o vvimevmenmmishiinss 100.0 100.0 100.0 100.0 100.0 Northeast ......c.usemsrsasrren 21.0 21.2 17:7 19.8 24.0 NorthCentral ................. 26.6 26.3 27.6 29.1 28.1 South .......c.ctiiivnrnnannn 31.2 31.0 33.5 33.6 29.3 West ...:vsimsamsmEnamuweme sms 21.2 21.5 21.2 17.5 18.5 Table 9 Number and percent distribution of persons by region, according to level of ambulatory care use: United States, 1980 Level of ambulatory care use Region Total 0 Low Intermediate High Number in thousands Northeast .......oceevvsvvsaen 46,899 9,350 7,793 27,315 2,441 North Central ................. 59,257 11,091 9,969 35,572 2,625 SoUtD «ivr vmiTnrms RBIs EEE 69,475 16,255 12,526 38,607 2,087 WES vv e eit eee 47,194 10,020 7,633 26,669 2,871 Percent distribution Total ccvsipsmprmnrvnrnsms 100.0 100.0 100.0 100.0 100.0 Northeast .........covvvveeenn 210 20.0 20.5 21.3 24.4 North Central ........c0vvuvenn 26.6 23.7 26.3 27.8 26.2 South ..... civ nnnrnns 31.2 34.8 33.0 30.1 20.8 WEEE ov ssi Famers sms esn 21.2 21.4 20.1 20.8 28.6 47 Table 10 Number and percent distribution of persons by region, according to level of prescribed medicine use: United States, 1980 Level of prescribed medicine use Region Total 0 Low Intermediate High Number in thousands Northeast .................... 46,899 18,410 6,022 20,918 1,648 North Central ................. 59,257 21,713 8,591 26,836 2,118 South ....................... 69,475 25,206 9,090 31,892 3,287 West ........................ 47,194 18,485 6,789 20,699 1,221 Percent distribution Total .................... 100.0 100.0 100.0 100.0 100.0 Northeast .................... 21.0 22.0 19.8 20.8 18.9 North Central ................. 26.6 25.9 28.2 26.7 25.9 South ....................... 31.2 30.1 29.8 31.8 40.2 WEBEL .oiuivuiiiomrnnnsusnmenns 21.2 22.1 22.3 20.6 14.9 Table 11 Number and percent distribution of persons under 65 years of age by type of health care coverage, according to level of hospital use: United States, 1980 Level of hospital use Health care coverage All persons 0 Low Intermediate High Coverage all year Number in thousands Single source: Private insurance ................................. 129,515 117,224 2,970 8,158 1,164 Medicaid ...................... 9,029 7,618 314 946 151 Other public programs . ............................. 3,348 2,918 132 199 98 More than 1 source: Private and public ................................. 15,912 13,607 482 1,431 392 2 or more public programs ....................0.0... 4,341 3,162 215 745 219 Other Covered for part of yearonly ........................... 19,869 17,856 519 1,324 170 No coverage allyear .............................. 17,342 16,482 182 614 63 Percent distribution TORRY ivi mina smn mein mee sa a5 8 08 5s msn asm are 100.0 100.0 100.0 100.0 100.0 Coverage all year ‘Single source: Private insurance ........................... .... .. 65.0 65.5 61.7 60.8 51.6 Medicaid ...................................."— 4.5 4.3 6.5 7.1 6.7 Other public programs . ........................... 1.7 1.6 2.7 1.5 4.4 More than 1 source: Private and public ................................. 8.0 7.6 10.0 10.7 17.4 2 or more public programs . .......................... 2.2 1.8 4.5 5.5 9.7 Other Covered for part of yearonly ........................... 10.0 10.0 10.8 9.9 7.5 No coverage allyear .............................. 8.7 9.2 3.8 4.6 2.8 48 Table 12 Number and percent distribution of persons under 65 years of age by type of health care coverage, according to level of ambulatory care use: United States, 1980 Level of ambulatory care use Health care coverage All persons 0 Low Intermediate High Coverage all year Number in thousands Single source: Private inSUranCe . ......ccovver rasan orerarsesons 129,515 26,027 23,455 75,083 4,951 Medicaid + ovo vv 9,029 1,863 1,600 5,118 448 Other public Programs . ........covevenonrerasons 3,348 835 773 1,681 159 More than 1 source: - Private and public ......ovvviiiiiiiiiiiiiie 15,912 2,368 2,136 10,280 1,128 2 or more public programs ........ coon 4,341 329 504 3,146 362 Other Covered for part of Yar only ...........eoeeeoeoniens 19,869 4,944 3,584 10,806 535 No coverage all Year .......ovvrutinrrrrronsnrennnn: 17,342 6,794 3,728 6,805 215 Percent distribution Total ....... conven errr 100.0 100.0 100.0 100.0 100.0 Single source: Private iNSUTANCE ... oo. coon ver orn asanooononsssnns 65.0 60.3 65.6 66.7 63.5 Medicaid . . ov vivre 4.5 4.3 4.5 4.5 5.7 Other public programs .......... Epp 1.7 1.9 2.2 1.4 2.0 More than 1 source: Private and public .......viiniiiiiii iii 8.0 5.5 6.0 9.1 14.5 2 or more public programs ...........ciieieen 2.2 0.8 1.4 2.8 4.6 Other Covered for part of yearonly ..........covevvvneernn 10.0 11.5 10.0 9.6 6.9 No coverage all year .........covveuierirernaarnsanees 8.7 15.7 10.4 5.9 2.8 49 Table 13 Number and percent distribution of persons under 65 years of age by health care coverage, according to level of prescribed medicine use: United States, 1980 Level of prescribed medicine use Health care coverage All persons 0 Low Intermediate High Coverage all year Number in thousands Single source: Private insurance .................co0virrnrn i. 129,615 49,056 19,670 58,394 2,395 PRORICAEA. iv vst vis0i iv va 0 50d tee tm ne me n w pe nn 9,029 3,638 1,433 3,669 289 Other public programs ...................0vvun... 3,348 1,488 333 1,309 218 More than 1 source: Private and public ................000vvinini. 15,912 5,077 2,105 7,818 912 2 or more public programs ...............0..r..... 4,341 964 627 2,338 411 Other Covered for part of yearonly ...............o0voo.... 19,869 8,636 2,997 7,999 238 Nocoverage all year ...............00ovueununn.... 17,342 10,063 1,956 5,226 97 Percent distribution TORE pis 000k 5150550 95 kc wm ws ws ow 00 0 0 0 a 8 8 7 2 100.0 100.0 100.0 100.0 100.0 Coverage all year Single source: Private insurance ....................0.0o uo... 65.0 62.2 67.5 67.3 52.5 Medicaid ............. 0.0 4.5 4.6 4.9 4.2 6.3 Other public programs ................c.coooo.... 3.7 1.9 1.1 1.5 4.8 More than 1 source: Private and public .............................. 8.0 6.4 7.2 9.0 20.0 2 or more public programs ........................ 2.2 1.2 2.2 2.7 9.0 Other Covered for part of yearonly ........................ 10.0 10.9 10.3 9.2 5.2 No coverage all year .................ccouvurnunn... 8.7 12.8 6.7 6.0 2.1 50 Table 14 Number and percent distribution of persons 65 years of age and over by type of Medicare coverage, according to level of hospital use: United States, 1980 Level of hospital use Type of Medicare coverage All persons 0 Low Intermediate High Covered by Medicare Number in thousands Medicare only ...........cuvueuuuunrnnnsnnsnsvnnnnns 3,836 3,193 63 403 177 Medicare and private COVErage ..............ooeovunonn 15,681 12,123 310 2,172 1,076 Medicare and other nonprivate coverage ...... ........ 2,647 1,866 41 423 318 Not covered by Medicare OLthEY COVEIAZE . ivi vo sinh in ss usm ns ase mss ros emsmse 1,033 973 15 35 10 Nocoverage at @l] ..qu: vn mrimnvmome smomnsns suis sites 272 260 - 12 0 Percent distribution TOBY .oicns sas nau snas mre wsms soa nsms «mn omen 100.0 100.0 100.0 100.0 100.0 Medicare only ........ootintrrerrannnrnnsrnasnnnns 16.3 17.3 14.6 13.2 11.2 Medicare and private COVerage ................oooooon 66.8 65.8 72.3 71.3 68.1 Medicare and other nonprivate coverage ............... 11.3 10.1 9.5 13.9 20.1 Not covered by Medicare Other COVETBZR «ov viv v vis as mgs ms eovme nme nt sas sasnns 4.4 5.3 3.6 1.2 0.6 Nocoverageatall ........... viii 1.2 1.4 - 0.4 0.0 Table 15 Number and percent distribution of persons 65 years of age and over by type of Medicare coverage, according to level of ambulatory care use: United States, 1980 Level of ambulatory care use Type of Medicare coverage All persons 0 Low Intermediate High Covered by Medicare Number in thousands Medicare only +... .xvrvtvvtnvrarrvcnsrarnnvvrrrnns 3,836 997 361 2,265 212 Medicare and private COVErage .............ooooouvunn 15,681 1,853 1,381 10,802 1,645 Medicare and other nonprivate coverage ............... 2,647 198 231 1,857 361 Not covered by Medicare Other COVErAEe .... vt v vrei nre vrs uriasenssnnsens 1,033 322 156 546 10 Nocoverage at all ..........covviivn viii iinns 272 187 12 74 - Percent distribution TOLAY visi vss ts CsI E RI NIE E IAFL aps she bh 100.0 100.0 100.0 100.0 100.0 Medicare only ...... cov nrr trian 16.3 28.0 16.9 14.6 9.5 Medicare and private COVerage ..........ooveve inno 66.8 52.1 64.5 69.5 73.9 Medicare and other nonprivate coverage ............... 11.8 5.6 10.8 11.9 16.2 Not covered by Medicare Other COVErage ........ eso vrtruranonusronarnsns 4.4 9.0 7.8 3.5 0.4 Nocoverage at all ..........ccvviiv nines 1.2 5.2 0.6 0.5 - 51 Table 16 Number and percent distribution of persons 65 years of age and over by type of Medicare coverage, according to level of prescribed medicine use: United States, 1980 Level of prescribed medicine use Type of Medicare coverage All persons 0 Low Intermediate High Covered by Medicare Number in thousands Medicare only ............. iii 3,836 1,267 261 1,896 412 Medicare and private coverage ....................... 15,681 2,847 897 9,341 2,597 Medicare and other nonprivate coverage ............... 2,647 291 175 1,663 518 Not covered by Medicare OUNBICOVOIABR 4 ic viv ois ins vias 4 58 bed 504 5 500 90004 0 nin im 1,033 316 25 617 75 Nocoverageatall ...............cvvvvviinnunnnn.. 272 172 13 76 11 Percent distribution TPOUBY «ccm pms 000 58 BEE Bl 23 0 8H BIE mn 5 313 oh 0 9 om on 100.0 100.0 100.0 100.0 100.0 Covered by Medicare Medicare only ....uiiinuimerassiisienenesmenmennes 16.3 25.9 19.1 13.9 11.4 Medicare and private coverage ....................... 66.8 58.2 65.4 68.7 71.9 Medicare and other nonprivate coverage ........ Cre nens 11.3 5.9 12.8 12.2 14.3 Not covered by Medicare Other Coverage ..............uiiiiiinnennnnennnnnns 4.4 8.5 1.8 4.5 2.1 NO COVELAZE ABU 4 4 io sims iv 515 9 51 4 955 0 5k 4 908 50 300 0 900 me 9 mn 1.2 3.5 1.0 0.6 0.3 Table 17 Number and percent distribution of persons by perceived health status, according to level of hospital use: United States, 1980 Level of hospital use Perceived health status All persons 0 Low Intermediate High Number in thousands Excellent ...............cc0iivivininnnn, 111,641 102,687 2,503 5,828 623 Good i 82,293 72,953 2,011 6,131 1,197 POlY tii ii amrmme mmm ms megan ine EAs 20,834 16,300 617 3,016 900 POOY vis nvit Ramus mas wm Ema hatin es mne nem 8,057 5,342 111 1,487 1,116 Percent distribution Total... ie 100.0 100.0 100.0 100.0 100.0 BROCUBHE wrmnvmimmsm msm sn i5hsian mime moe me 50.1 52.1 47.7 35.4 16.2 Good Li ee 36.9 37.0 38.4 37.2 31.2 Ei 1 9.3 8.3 11.8 18.3 23.5 Poor 3.6 2.7 21 9.0 29.1 52 Table 18 Number and percent distribution of persons by perceived health status, according to level of ambulatory care use: United States, 1980 Level of ambulatory care use Perceived health status All persons 0 Low Intermediate High Number in thousands EXCEllent eevee 111,641 26,724 22,302 60,174 2,440 Good ET 82,293 16,555 13,176 48,768 3,794 PHEE io 05 50 ni bo i) oi ee, 3 05 0 50 mobo oe ot om wim me Bc 20,834 2,788 1,963 13,796 2,286 POOL somnamssormaanstms@nsns Rain sess sniy 8,057 648 479 5,425 1,505 Percent distribution TORY v5.0 55 50 0 5 ch i vw 5 cht on 30 wl wm chin mows om 100.0 100.0 100.0 100.0 100.0 Excellent «. cc suvivnsmasmssasvsmmsmann san 50.1 87.2 58.8 47.0 24.3 Good 02 1 ch 0 + oe a a2 520 va eh Be 9 bs 202 4 2 43 is 2B 36.9 35.4 34.7 38.1 37.8 FOIY sivismsarmmsduaavismusmsivsns sndmsin 9.3 6.0 5.2 10.8 22.8 Poor ee 3.6 1.4 1.3 4.2 15.0 Table 19 Number and percent distribution of persons by perceived health status, according to level of prescribed medicine use: United States, 1980 Level of prescribed medicine use Perceived health status All persons 0 Low Intermediate High Number in thousands Excellent ........... iii 111,641 49,779 17,335 43,823 703 GOOAQ. isin ima iRs sR IANNIS tHE NI SRS E 82,293 29,116 11,094 39,832 2,250 Fair ee ee 20,834 3,911 1,701 12,560 2,662 POOL .ovvrmemnsms ss asa nifommiis shsas 8,057 1,008 362 4,129 2,558 Percent distribution Total ........iviiiiii iii ii inne 100.0 100.0 100.0 100.0 100.0 Excellent cici-usssinmimatvi@assns snsneses 50.1 59.4. 56.9 43.7 8.6 Good i eee ea 36.9 34.7 36.4 39.7 275 FPalY iain ins snramins RIGO VIR maEms ee 9.3 4.7 5.6 125 32.6 POOF sivcimmins@mimmsms smasps ess seems nism 3.6 1.2 1.2 4.1 31.3 53 Table 20 Number and percent distribution of persons 17 years of age and over and living at the end of the survey period by functional limitation score, according to level of hospital use: United States, 1980 Level of hospital use Functional limitation score All persons 0 Low Intermediate High Number in thousands TOBY os mriemimi ami REN En ROWE SERS Rd Hh 160,437 140,051 3,524 13,615 3.247 Level 1 — nolimitation ................00vvuu... 122,298 111,182 2,626 7,842 648 Level 2 — minimal limitation .................... 9,945 8,392 252 1,094 207 Level 8... ee 6,900 5,650 143 871 337 Level 4... ii 7,100 5,255 279 1,142 425 Level 5... LIER Ar ERIE 6,604 4,995 134 1,069 407 Level B ....o iii ee eee 4,178 2.722 B57 864 535 Eo ST TTT LIL TET TI 1,721 1,039 13 360 309 Level 8 — most severe limitation ................. 1,691 917 20 373 381 Percent distribution i 1 ET IT ITI TITY 100.0 100.0 100.0 100.0 100.0 Level 1 — nolimitation .................cc.oov... 76.2 79.4 74.5 57.6 20.0 Level 2 — minimal limitation .................... 6.2 6.0 7.1 8.0 6.4 Level 3. ee 4.3 4.0 4.1 6.4 10.4 1oVE 4 nimi imimmen inns mid demas scnmisas anne 4.4 3.8 7.9 8.4 13.1 I 4.1 3.6 3.8 7.9 12.5 Level 6 ....coiiiii iii i ea 2.6 1.9 1.6 6.3 16.5 LoVe] 7 viv r smn ma i me RRS RI EI EEE EEE 1.1 0.7 0.4 2.6 9.5 Level 8 — most severe limitation ................. 1.2 0.7 0.6 2 11.7 Table 21 Number and percent distribution of persons 17 years of age and over and living at the end of the survey period by functional limitation score, according to level of ambulatory care use: United States, 1980 Level of ambulatory care use Functional limitation score All persons 0 Low Intermediate High Number in thousands Total civ ivn ins vs Rss TITIAN AIE HIRAM 160,437 33,889 25,388 92,734 8,426 Level 1 = no limitation . ........... cc. iiinnn 122,298 29,992 22,358 66,531 3,417 Level 2 = minimal limitation .................... 9,945 1.065 1,022 7,032 826 evel 8 i nimi iwsi $5 beim RH RING IE H EE Le 6,900 924 460 4,467 1,049 LEVEL. oo 0 rvs ve or 0 ws cot son tw cor won 4 ok bc 6 v8 9 a oe BE 7,100 720 661 4,939 780 LevelD ..uiuniis inns mains iniNs Bene Ehsms ewms® 6.604 572 482 4,704 846 BRVEY 8 iioo vivo mms mw som ws mss gowns ws wny www www bw 4,178 313 170 2,928 767 LBVRY 7 ev vm mivmm sms So ho B80 $00 0 B20 010 0 UB J BS RI 1,721 184 97 1,095 345 Level 8 — most severe limitation ................. . 1,691 118 137 1,038 398 Percent distribution POA] or mmr imei i a mi RISA EGET EHR NY 100.0 100.0 100.0 100.0 100.0 Level l “ no Bmitatlon . cis uiwsuninsnmins coo wmns 76.2 88.5 88.1 717 40.6 Level 2 = minimal limitation .................... 6.2 3.1 4.0 7.6 9.8 LEVEL Bui vin ins s 0s issn rnesiomeppsvs cows 4.3 2.7 1.8 4.8 12.5 Level 4 oot i i ee 4.4 2.1 2.6 5.3 9.9 LReVEID ,inivn ims sss Amman ANE WCW ME CE 4.1 3.7 1.9 5.1 10.0 Level B «ccvivirnsomrmusrnsnnmanasyemenasns ems 2.6 0.9 0.7 3.2 9.1 LEVEY 7 vc vvnmrc mmr mr mmr aban A SE BERING FY 1.1 0.5 0.4 1.2 4.1 Level 8 — most severe limitation ................. 1.1 0.3 0.5 1.1 4.7 Table 22 Number and percent distribution of persons 17 years of age and over and living at the end of the survey period by functional limitation score, according to level of prescribed medicine use: United States, 1980 Level of prescribed medicine use Functional limitation score All persons 0 Low Intermediate High Number in thousands FORAY 5: 4055.5 5 5.6 m0 5 om ow cor wre wk x er a8 we pwede 160,437 56,465 19,801 76,389 7,782 Level 1 — no limitation ......................... 122,298 50,301 17,295 53,389 1,313 Level 2 = minimal limitation .................... 9,945 2,244 885 6,201 615 LBVEL BF irom ow 5010 618 16008 1 5005.50 57 505 905 kw oot me mo 6,900 1,274 468 4,302 856 Level 4 . 7,100 1,102 511 4,372 1,115 LE I 6,604 796 349 3,969 1,490 Level 6 ...oouii 4,178 371 127 2,372 1,308 LBVEL 7 ot v0 vais mosis dn bom aon voir son sen oe oe wie ot win wis is 1,721 165 89 894 574 Level 8 — most severe limitation ................. 1,691 214 76 890 511 Percent distribution TOA) tn vvms a msm matin i Shams cs Harn ms ains mornin 100.0 100.0 100.0 100.0 100.0 Level 1 = no limitation ......................... 76.2 89.1 87.3 69.9 16.9 Level 2 = minimal limitation .................... 6.2 4.0 4.5 8.1 7.9 Level 3 ii 4.3 2.3 2.4 5.6 11.0 JUVE 5c tom 15 0% hh mia 8 om 0m 0 9 ie wm wa x he 4.4 2.0 2.6 5.7 14.8 Level 5... i 4.1 1.4 1.8 5.2 19.1 Level 6 oii 2.6 0.7 0.6 3.1 16.8 Lt 1.1 0.3 0.4 1.2 7.4 Level 8 — most severe limitation ................. 1 | 0.4 0.4 1.2 6.6 56 Number and percent distribution of persons by number of unique conditions occurring Table 23 during the year, according to level of hospital use: United States, 1980 Level of hospital use Number of unique conditions All persons 0 Low Intermediate High Number in thousands Oi ciinssvsn iis Bede sd nid nsuaiNs 0 AImsunInp Lens 32,385 31,834 216 335 - Xr ie 32 wr 20 0 0m 0) 9 8 vu: 0 0 ig ed A dr i cs Bf vw oe 42,428 40,267 580 1,444 137 ni ar HA RIERA BAREIS NERNEY 40,341 37,338 773 1,907 324 Sh a ER EEE EEE EE WE IEE WS RK ye 33,010 29,150 914 2,610 336 Bs vein ee RB 8 Be BE RRR ER STR 23,485 20,022 750 2,289 424 BD rir RARER HES MEER AEN INGER SW 17,409 14,398 533 2,074 405 B i ee ei eee eee 10,952 8,381 550 1,641 480 ED RA A RANI BRINN NEI SHORE OW 10 s ™ 8,003 6,042 248 1,331 382 Sentai nyu rn te we HG mre wy Ft wt a Poh een 5,281 3,885 259 868 269 D eins m rR R A REBAR RAE eH mR aE 3,127 2,140 203 582 202 0 oi snr s aR INI META EE SEACH RANGER DIME EEE 2,250 1,452 7 472 249 BABY wiv ve vorew ora on bom vw oom in i Se Be WE WR dd oe 6 BT 4,153 2,373 140 1,010 G29 Percent distribution TOA] civssmmemsnmsensns smammsnsnws ems wy 100.0 100.0 100.0 100.0 100.0 0 x 1 + ve wt wn ce we 80 br lH eB PWR 14.5 16.1 4.1 2.0 - HEE Inert 19.0 20.4 11.1 8.8 3.6 0 £0 le 0 0 ew 0 en 1 0 0 og en ed 8 ke red vz we 9 Bde Ben in x ez 4 we 18.1 18.9 14.7 11.6 8.4 JR Pa a TY RIL Inn 14.8 14.8 17.4 15.9 8.8 EE Rp 10.5 10.1 14.3 13.9 11.0 B iim ama ESE RARE ER TEARS FE 7.8 7.3 10.2 12.6 10.6 B ius sas HIRE RIE IAN CHIE R EWI E ERATE EE 4.9 4.2 10.5 9.4 12.5 Tet eee ee eas 3.6 3.1 4.7 8.1 10.0 B nN AIR RESHAPE RSME IRIN I WIA FREE 2.4 2.0 4.9 5.3 7.0 Or et oe wie BE pid int nn rie ee pa 1.4 1.1 3.9 3.5 5.3 JO wenn itri mrs nits BISA ECE ESR REHAB EHD 1.0 0.7 1.5 2.9 6.5 T1m28 vi vss ms NIFH IOP ERSTE HT NTE 1.9 1.2 2.7 6.1 16.4 57 Table 24 Number and percent distribution of persons by number of unique conditions occurring during the year, according to level of ambulatory care use: United States, 1980 Level of ambulatory care use Number of unique conditions All persons 0 Low Intermediate High Number in thousands 0 sis cS BSE IEE GEE GES EEE Br ao ed one ee 32,385 23,711 6,094 2,632 48 1 I TE I IT 42,428 12.682 14,042 15,393 311 0 HR er Ba Be wo ow XC ET 40,341 5.907 9,258 24,387 789 Ep 33,010 2.607 4,792 24,698 913 BE oii 4 1040 0 ne or ve en i SHE i 1 ER ET Tm 01 23,485 998 1,981 19,307 1,199 BD i EAE AN ARVIN ME Bh IEEE ERIE EW 17,409 367 837 14,746 1,459 LO 10,952 224 619 9,261 848 FE ASAE WE SME MS ATE HH RR Bok of Becki ro ch ®ve wo er nic 8,003 129 135 6,849 891 Be ee 5,281 28 90 4,336 827 0 08 5 Bw ter a RX 3 3,127 35 73 2,380 638 YO urns omens ini sus HEINE SHADER N BERL sR a 2,250 12 - 1,709 530 11-24 eee 4,153 17 LL 2,564 1,572 Percent distribution Total ....... ee 100.0 100.0 100.0 100.0 100.0 0 cms mis amt m sims SH RT Ris EME mai bi dn ®d Hm bars 14.5 50.8 16.1 2.0 0.5 ET 19.0 27.1 37.0 12.0 3.1 2 BR REE SR hhh Tinh Henman vr oo cw ol wows 32 ae ew ew nw of 18.1 12.6 24.4 19.0 7.9 GS mmr A EEE EN AE EWE EE RE EEE ENE 14.8 5.6 12.6 19.3 9.1 BE 0 Gos 56030 Jor 4 oR cot i wt eT I ET RE ORL 10.5 2.1 5.2 15.1 12.0 BD nse IE IPR EE ER EMER EN EEE RCE 7.8 0.8 2:2 11.5 14.6 8 4 6 st ans 00 cui ow ht 9 sO ER TH TR 4.9 0.5 . 1.6 7.2 8.5 BIR ARI ABER SE AE EOIN EIRENE R IEE 3.6 0.3 0.4 5.3 8.9 Bee 2.4 0.1 0.2 3.4 8.2 0 ih PERE REE EE RAE BEE lo 4108 ew Fh ee 1.4 0.1 0.2 1.9 6.4 0 ovat i ww ee EE eR REE SE EE 1.0 0.0 i 1.3 5.3 X28 40 5 05 choi oh wm ew Ye 1.9 0.0 —- 2.0 15.6 58 Table 25 Number and percent distribution of persons by number of unique conditions occurring during the year, according to level of prescribed medicine use: United States, 1980 Number of unique conditions All persons Level of prescribed medicine use 0 Low Intermediate High 32,385 42,428 40,341 33,010 23,485 17,409 10,952 8,003 5,281 3,127 2,250 4,153 18.1 Number in thousands 30,760 873 751 - 24,445 8,204 9,787 41 14,255 9,199 16,458 429 7.729 5,841 18,856 584 3,552 3,042 16,214 678 1,624 1,640 13,231 915 855 806 8,470 821 289 439 6,268 1,007 95 218 4,034 934 95 92 2,179 761 53 61 1,675 561 64 76 2,672 1,442 Percent distribution 100.0 100.0 100.0 100.0 36.7 2.9 0.7 ~ 29.2 26.9 9.7 0.5 17.0 30.2 16.4 5.3 9.2 19.2 18.8 7.1 4.2 10.0 16.2 8.3 1.9 5.4 13.2 11.2 1.0 2.6 8.4 10.0 0.3 1.4 6.2 12.3 0.1 0.7 4.0 11.4 0.1 0.3 2.2 9.3 0.1 0.2 1.6 6.9 0.1 0.2 2.6 17.6 59 Table 26 Number and percent distribution of persons with specified diagnoses associated with hospital admissions by diagnostic category, according to level of hospital use: United States, 1980 Level of hospital use Number of hospitalized persons in Intermediate Diagnostic category thousands All users Low users users High users Percent distribution TOUT 1c 5 cn 200m pom oe iw si wilt oi she ee res ee 0 4 40 sg 02 0 2 2 he po 100.0 100.0 100.0 100.0 Infectious and parasiticdiseases ............................ 731 2.9 2.3 3.2 2.3 Neoplasms . . . . LL... 1,638 6.4 3.3 5.0 16.7 Endocrine, nutritional and metabolic diseases, and immunity ‘ disorders . ... Lo... 1,026 4.0 1.0 4.5 6.1 Diseases of the blood and blood-formingorgans ................. 306 1.2 0.5 1.0 3.0 Mental QISONGOIS: « cc momo v5 5 9 8 500 5 6 £ 5 5, 0 8 ah 0 0.0 Aen 0c 0 41 iin om 0 0c wise 695 2.7 1.8 2.0 7.1 Diseases of the nervous system and sense organs ............... 1,558 6.7 6.9 5.2 8.7 Diseases of the circulatory system ........................... 3,506 13.7 5.6 11.9 32.8 Diseases of the respiratory system ........................... 3,113 12.2 15.0 10.4 16.1 Diseases of the digestive system ............................ 3,366 13.2 8.3 14.8 12.8 Diseases of the genitourinary system ......................... 2,977 11.7 15.0 11.3 8.5 Complications of pregnancy, childbirth, and the puerperium ....... 1,901 7.4 6.9 8.7 2.6 Diseases of the skin and subcutaneous tissue .................. 406 1.6 1.0 1.5 2.7 Diseases of the musculoskeletal system and connective tissue ..... 1,952 7.6 3.5 7.2 15.1 Congenital anomalies ..................................... 263 1.0 2.3 0.6 0.9 Certain conditions originating in the perinatal period ............ 87 0.3 - 0.4 0.7 Symptoms, signs and ill-defined conditions .................... 1,834 7.2 4.8 6.7 12.6 Injury and poisoning ............. LL... LL... a... 3,123 12.2 13.5 9.6 21.8 No condition or condition unknown .......................... 3,699 14.5 14.7 17.1 2.8 1 Unduplicated count of persons within diagnostic categories; a person who had multiple hospital admission for the same condition is counted only once. However, since each hospital admission may have been associated with up to 3 separate diagnostic categories, the same person may be counted in more than 1 diagnostic category. Diagnoses were examined only for persons having at least 1 overnight hospital stay. Number and percent distribution of persons by whether surgery was performed Table 27 during the stay, according to level of hospital use: United States, 1980 Level of hospital use Presence of surgery All persons Low Intermediate High Number in thousands SUEY vin n i IH IMIR WIE RR ps Ee ew, 11,904 2,566 7,389 1,949 NO BULZEIY ott ieee ei iiin eee eens 13,637 2,676 9,073 1,888 Percent distribution Total c.vusiiinssmsnsrsansvsmnivtansnyrms 100.0 100.0 100.0 100.0 SUrgerY ii aa 46.6 49.0 44.9 50.8 NOBUPZOIY sist sats i RUini@MIBIETIWIRE REBEL NS 53.4 51.0 55.1 49.2 Table 28 Number and percent distribution of persons by level of ambulatory care use, according to level of hospital use: United States, 1980 Level of hospital use Level of ambulatory care use All persons 0 Low Intermediate High Number in thousands Total coins sn nuswainssRiwmmIAs REI mE ny 222,824 197,283 5,242 16,462 3,837 0 Lins vv ism rsd ss BF HR ERA Re 46,716 46,204 70 324 119 LOW orn mmr msm nis FRSA RRR ERI BRE AW 37,921 36.824 293 668 136 Intermediate ...covvsnvinssssnwsnnsmsa mows sue 128,163 107,790 4,541 13,488 2,344 High 10,024 6,465 339 1,982 1,238 Percent distribution Total Lo. eas 100.0 100.0 100.0 100.0 100.0 0 iiurininsanimni mist aai aussi amen emyinsny 21.0 23.4 1.3 2.0 3.1 BOW: sims mm smo sm sw mn ® oma s® somes mes wenn ens 17.0 18.7 5.6 4.1 3.6 Imermediate ....... oo oar vninssnsnnsmemusinn 57.5 54.6 86.6 81.9 61.1 High i a 4.5 3.3 6.5 12.0 32.3 61 IT. Number and percent distribution of persons with hospital use by number of hospital admissions, according to level of hospital use: United States, 1980 Level of hospital use 62 Number of hospital admissions All persons Low Intermediate High Number in thousands ] vr mre nmin in ERI RAE A SRA E REE 19,629 5,182 13,228 1,219 2 et te ei a aa 4,274 GO 2,800 1,414 J TL LL IIIT ITI 986 - 327 659 de a 351 - 93 258 BOPIOTE .uciniirinrnmsvmenmsnemummenmsewenn 303 - 15 288 Percent distribution Total ....... 100.0 100.0 100.0 100.0 I tint it mi ns BEB? FEA mE win ro oo wow mm wim vn 76.8 98.9 80.4 31.8 rE Erm EE EE ERE RS AE DERE 16.7 1.1 17.0 36.8 Bi ERE BERR Ren rm hr hw i ay ewe 3.9 - 2.0 17.2 EE ET I I PET Tr 1.4 - 0.6 6.7 Sormore ..... LL. ee 1.2 i 0.1 7.4 Table 30 Number and percent distributions of persons by survival status, according to level of hospital use: United states, 1980 Level of hospital use Survival status All persons 0 Low Intermediate High Number in thousands TONE) cinssnnins ibn ia Rin BnsMbIns@nins 222,824 197,283 5,242 16,462 3,837 SUIVIVOYS ott it it tt ee ee ee ee eee 221,969 197,079 5,204 16,312 3,373 Deaths ....ovviisntaait sds dni Bidmi Bhidianisd 856 . 204 38 150 464 Percent distributions Total ..... oi 100.0 88.5 2.4 7.4 1.7 BUrvivors uv avianin sia sure RIS REIB NIB IP HEED 100.0 88.8 2.3 7.3 1.5 Deaths ......... iii 100.0 23.8 4.4 17.5 54.2 Total voir imsmprarsmssmswermieessms pms vn 100.0 100.0 100.0 100.0 100.0 SUIVIVOIS ica insinrmbasmmsino smbmamboaibomedins 2m 99.6 99.9 99.3 99.1 87.9 Deaths ,urusvnsininmi aims iw snsEpimusns cpsen 0.4 0.1 0.7 c.9 12.1 Table 31 Number and percent distributions of persons by survival status, according to level of ambulatory care use: United States, 1980 Level of ambulatory care use Survival status All persons 0 Low Intermediate High Number in thousands Total vine emsvmenmprmrensmprnsmvemnrwos 222,824 46,716 37,921 128,163 10,024 BUIVIVOIS «vv vem mm smn rns bmimmbmammsmns ns 221,969 46,601 37,892 127,638 9,838 Deaths « vuiinrvsr ms imesmaeis census snimmegnswes 856 115 29 525 187 Percent distributions TOAL + eevee eee ee 100.0 21.0 17.0 57.5 4.5 SWIVIVOIS oovnsnssoramamsinsamtnsmevsmps mus wes 100.0 21.0 17.1 57.5 4.4 Deaths ........ii iii iii iii iii 100.0 13.4 3.4 61.4 21.8 POUR] 4 vo ur aimazvir gms vie 4 mew ei ol oy @ rls enn 100.0 100.0 100.0 100.0 100.0 BULVIVOYS +o isms ens THI mbm 20 Rb EIB IAD ERR EF E A 99.6. 99.8 99.9 99.6 98.1 Deaths ; co vivins sms s va wusms 4 pia Bt ns maI@ He wns 0.4 0.2 0.1 0.4 1.9 Table 32 Number and percent distributions of persons by survival status, according to level of prescribed medicine use: United States, 1980 Level of prescribed medicine use Survival status All persons 0 Low Intermediate High Number in thousands TOLAL + vv vvsnimome smsmmows smsmmems sms enn e 222,824 83,414 30,492 100,345 8,173 SUrvIVOYS . cvs vosamivs 1ms sins auemsmusny sms veges 221,969 83,670 30,436 99,900 7,964 Deaths .. iii ee a 856 144 56 446 209 Percent distributions Total cuvimssmimrinmsnrer ins vmsnrerren es 100.0 37.6 13.7 45.0 3.7 SUrVIVOLS «ovr nvr sms mamas sss a dE ni wn smns 100.0 37.7 13.7 45.0 3.6 Deaths .. vivir sss i wa impis i sni@svRimdsevs®esy 100.0 16.9 6.6 52.1 24.5 Total onsionins eas RINT PEIRUT VIR IBGE 100.0 100.0 100.0 100.0 100.0 SUIVIVOIS . viii i iii iii t ntti cine enn 99.6 99.8 99.8 99.6 97.4 Deaths svi unsmmens s 5s SH r a BINS W HOHE IIW HELIS 0.4 0.2 0.2 0.4 2.6 Appendixes I. Sample Design, Data Collection, and Processing ................uuunne eee eee 65 IrOAUCHION Le 65 Survey Background . . ..... 65 Sample Design of NMCUES . . eee 65 Research Triangle Institute Sample Design . . ............ iii 66 National Opinion Research Center Sample Design . ............. o.oo eee 66 Collection of Data . . ......... ii 67 WBIghtING . . «oe 67 SUEY NONMMBSPONSE . . . o.oo itt ttt teeta ee eee ee ee eee 68 AHtrtion IMPULAtioN . . «LL... 68 Item Nonresponse and IMpUtatioN . ................uiniiiiniiinii ini tee teeta eta ee saaaersanannns 68 Il. Data Modifications to Public Use FileS . . . ............ oo. eee 71 Overview of Data Changes . . .... oo. eee ee eee 71 ll. Analytical Strategies . ............... ii 73 Notion of an Average Population . ............ 73 Role of Weights and Imputation . ............. 74 EESTIMIBEION PTOCOOUIBE. xix mse im om 55 5550555 0 46 506.589.5515 Bok 928 50 sh 0 ot 08 41 sem 2 1 1 50 50 3 000 it 0 0 ms 77 The Logistic Regression Model . . . ....... o.oo iii eee eee ee 78 IV. Sampling ErrOrS 81 TOS tutti rrr a aaa. ete errr iaaaaaannnn. 81 aN 82 Proportions and Percents ................. i 83 Mutually Exclusive Subgroup Differences . . .............euiteiiiit ee 84 * Subgroup to Total Group DIfferences . . ............... nee 84 V. DOfiNION Of Terms .......ccvvini iii iiiii itis teetnnnnnnnnnnnnnnesnssessssssssssnsinsnsessnnnnnns 86 List of Appendix Figures I. Dynamic population for 12 time period panel SUIVEY . . .............ouni ieee eee 73 Il. Estimated mean charges per hospital outpatient department visit, by 4 family income classes for all and real data: United States, 1980 . ......... ttt 76 ll. Estimated mean charge per hospital outpatient department visit, by 16 imputation classes for all persons and for per- sons in families with income less than $5,000: United States, 1980 . . .. ...... outer ere, 76 List of Appendix Tables I. Percent of data imputed for selected survey items in 4 of the NMCUES public use data files: United States, 1980 ..... 69 Il. Effect of person-year adjustment on counts and sampling weights, by 4 population groups: United States, 1980 . . ..... 74 lll. Sample size, means, and standard errors for 5 disability measures, by all and real data subgroups: United SUBIOS, TBO 4s coin ws whe mw 5 5505 5% 5 905 515 wm m0 0 #0 mmm 850 cm om 03 I 0 AE 75 IV. Sample size, means, standard errors, and element variance for total charge for a hospital outpatient department visit, by data type: United States, 1980 .............. iii 75 V. Coefficients for standard error formula for estimated aggregates or totals . .................oouuureunrennnnnn... 82 VI. Values of roh and §? for standard error formula for estimated MEANS . ...............ouerenen een, 83 VII. Values of roh for standard error formula for estimated proportions . . ................c.uuue een, 83 Appendix | Sample Design, Data Collection, and Processing Introduction The National Medical Care Utilization and Expenditure Survey (NMCUES) was designed to collect data about the U.S. civilian noninstitutionalized population during 1980. The complexity of the survey requires the analyst to be familiar with a range of design features during the analysis, both to determine appropriate analytic methods and to investi- gate the impact that the design may have on a particular analysis. Several topics are addressed in this appendix: The overall design of NMCUES, the survey background, sampling methods, data collection methods, weighting, and compensa- tion procedures for missing data. These descriptions essentially present NMCUES data as they are available to the user of the public use data tape. This appendix draws heavily from a paper in the Proceedings of the 19th National Meeting of the Public Health Conference on Records and Statistics (Casady, 1983). Survey Background During the course of NMCUES, information was obtained on health, access to and use of medical services, associated charges and sources of payment, and health care coverage. The survey was cosponsored by the National Center for Health Statistics (NCHS) and the Health Care Financing Ad- ministration (HCFA). Data collection was provided under contract by the Research Triangle Institute (RTI) and its subcontractors, National Opinion Research Center (NORC) and SysteMetrics, Inc. The basic survey plan for NMCUES drew heavily on two previous national surveys: The National Health Interview Survey (NHIS), which is conducted by NCHS, and the Na- tional Medical Care Expenditure Survey (NMCES), which was cosponsored by the National Center for Health Services Research (NCHSR) and NCHS. NHIS is a continuing, multipurpose health survey first conducted in 1957. The primary purpose of NHIS is to collect information on illness, disability, and the use of medi- cal care. Although some information on medical expenditures and insurance payments has been collected in NHIS, the cross-sectional nature of the NHIS survey design is not well suited for providing annual data on expenditures and payments. NMCES was a panel survey in which sample households were interviewed six times over an 18-month period in 1977 and 1978. NMCES was designed specifically to provide comprehensive data on how health services were used and paid for in the United States in 1977. 'NMCUES is similar to NMCES in survey design and question wording, so that analysis of change during the 3 years between 1977 and 1980 is possible. Both NMCUES and NMCES are similar to NHIS in terms of question wording in areas common to the three surveys. Together they provide extensive information on illness, disability, use of medical care, costs of medical care, sources of payment for medical care, and health care coverage at two points in time. Sample Design of NMCUES General plan—The NMCUES sample of housing units and group quarters, hereafter jointly referred to as dwelling units, is a concatenation of two independently selected national samples, one provided by RTI and the other by NORC. The sample designs used by RTI and NORC are quite similar with respect to principal design features; both can be character- ized as stratified, multistage area probability designs. The principal differences between the two designs are the type of stratification variables and the specific definitions of sampling units at each stage. Target population—All persons living in a sample dwell- ing unit at the time of the first interview became part of the national sample. Unmarried students 17-22 years of age who lived away from home were included in the sample when their parent or guardian was included in the sample. In addition, persons who died or were institutionalized between January 1 and the date of first interview were included in the sample if they were related to persons living in the sample dwelling units and were living in the sample dwelling before their death. All of these persons were considered “key” persons, and data were collected for them for the full 12 months of 1980 or for the portion of time they were part of the U.S. civilian noninstitutionalized population. In addition, children born to key persons during 1980 were considered key persons, and data were collected for them from the time of birth. Relatives from outside the original population (i.e., institutionalized, in the Armed Forces, or outside the United States between January 1 and the first interview) who moved in with key persons after the first interview were also considered key persons, and data were collected for them from the time they joined the key person. 65 Relatives who moved in with key persons after the first interview but were part of the civilian noninstitutionalized population on January 1, 1980, were classified as “nonkey” persons. Data were collected for nonkey persons for the time that they lived with a key person; but because they had a chance of selection in the initial sample, their data are not used for general analysis of persons. However, data for nonkey persons are used in an analysis of families because they contribute to the family’s utilization of and expenditures for health care during the time they are part of the family. Because family analysis is not part of this investigation, it will not be discussed further. Persons included in the sample were grouped into “report- ing units” for data collection purposes. Reporting units were defined as all persons related to each other by blood, marriage, adoption, or foster care status who lived in the same dwelling unit. The combined NMCUES sample consisted of approxi- mately 7,200 reporting units, of which 6,600 agreed to partici- pate in the survey. In total, complete data were obtained on 17,123 key persons. The RTI sample yielded approximately 8,300 respondents and the NORC sample 8,800. Research Triangle Institute Sample Design Primary sampling units (PSU’s)—A PSU was defined as a county, a group of contiguous counties, or parts of counties with a combined minimum 1970 population size of 20,000. A total of 1,686 nonoverlapping RTI PSU’s cover the entire land area of the 50 States and Washington, D.C. The PSU’s were classified as one of two types. The 16 largest Standard Metropolitan Statistical Areas (SMSA’s) were designated as self-representing PSU’s, and the remaining 1,670 PSU’s in the primary sampling frame were designated as nonself-representing PSU’s. Stratification of PSU’s—PSU’s were grouped into strata whose members tend to be relatively alike within strata and relatively unlike between strata. PSU’s derived from the 16 largest SMSA’s were of sufficient 1970 population size to be treated as primary strata. The 1,659 nonself-representing PSU’s from the continental United States were stratified into 42 approximately equal-sized, primary strata. Each of these primary strata had a 1970 population size of about 3.3 million. One supplementary primary stratum of 11 PSU’s, with a 1970 population size of about 1 million, was added to the RTI primary frame to include Alaska and Hawaii. First-stage selection of PSU’ s—The total RTI primary sample consisted of 59 PSU’s of which 16 were self-represent- ing. The nonself-representing PSU’s were obtained by select- ing 1 PSU from each of the 43 nonself-representing primary strata. These PSU’s were selected with probability proportional to 1970 population size. Secondary stratification—In each of the 59 sample PSU’s, the entire PSU was divided into nonoverlapping smaller area units called secondary sampling units (SSU’s). Each SSU consisted of one or more 1970 census-defined enumeration districts (ED’s) or block groups (BG’s). Within each PSU the SSU’s were ordered and then partitioned to form approximately equal-sized secondary strata. Two secondary 66 strata were formed in the nonself-representing PSU drawn from Alaska and Hawaii, and four secondary strata were formed in each of the remaining 42 nonself-representing PSU’s. Thus, the nonself-representing PSU’s were partitioned into a total of 170 secondary strata. In a similar manner the 16 self-representing PSU’s were partitioned into 144 secon- dary strata. Second-stage selection of SSU’s—One SSU was selected from each of the 144 secondary strata covering the self-repre- senting PSU’s, and two SSU’s were selected from each of the remaining secondary strata. All second-stage sampling was with replacement and with probability proportional to the SSUs’ total noninstitutionalized population in 1970. The total number of sample SSU’s was 2 X 170 + 144 = 484. Third-stage selection of areas and segments—Each SSU was divided into smaller nonoverlapping geographic areas, and one area within the SSU was selected with probability proportional to the 1970 total number of housing units. Next, one or more nonoverlapping segments of at least 60 housing units (HU’s) were formed in the selected area. One segment was selected from each SSU with probability proportional to the segment HU count. In response to the sponsoring agencies’ request that the expected household sample size be reduced, a systematic sample of one-sixth of the segments was deleted from the household sample. Thus, the total third- stage sample was reduced to 404 segments. Fourth-stage selection of housing units—All of the dwell- ing units within the segment were listed, and a systematic sample of dwelling units was selected. The procedures used to determine the sampling rate for segments guaranteed that all dwelling units had an approximately equal probability of selection. All of the reporting units within the selected dwelling units were included in the sample. National Opinion Research Center Sample Design Primary sampling units (PSU’s)—The land area of the 50 States and Washington, D.C. was divided into nonoverlap- ping PSU’s. A PSU consisted of SMSA'’s, parts of SMSA’s, counties, parts of counties, or independent cities. Grouping of counties into a single PSU occurred when individual coun- ties had a 1970 population of less than 10,000. Zoning of PSU s—The PSU’s were classified into two groups according to metropolitan status (SMSA or not SMSA). These two groups were individually ordered and then par- titioned into zones with a 1970 census population size of 1 million persons. First-stage zone selection of PSU’ s—A single PSU was selected within each zone with a probability proportional to its 1970 population. It should be noted that this procedure allows a PSU to be selected more than one time. For instance, an SMSA PSU with a population of 3 million may be selected at least twice and possibly as many as four times. The full general-purpose sample contained 204 PSU’s. These 204 PSU’s were systematically allocated to 4 subsamples of 51 PSU’s. The final set of 76 sample PSU’s was chosen by randomly selecting 2 complete subsamples of 51 PSU’s; 1 subsample was included in its entirety, and 25 of the PSU’s in the other subsample were selected systematically for inclu- sion in NMCUES. Second-stage zone selection of SSU’ s—Each of the PSU’s selected in the first stage was partitioned into a nonoverlapping set of SSU’s defined by BG's, ED’s, or a combination of the two types of census units. SSU’s were selected from the ordered list of these SSU’s. The cumulative number of households in the second-stage frame for each PSU was divided into 18 zones of equal width. One SSU had the opportunity to be selected more than once, as was the case in the PSU selection. If a PSU had been hit more than once in the first stage, then the second-stage selection process was repeated as many times as there were the first-stage hits. Some 405 SSU’s were identified by selecting 5 SSU’s from each of the 51 PSU’s in the subsample that were included in their entirety, and 6 SSU’s from each of the 25 PSU’s in the group for which one-half of the PSU’s were included. Third-stage selection of segments—The selected SSU’s were subdivided into area segments with a minimum size of 100 housing units. One segment was then selected with probability proportional to the estimated number of housing units. Fourth-stage selection of housing units—Sample selection at this level was essentially the same as for the RTI design. Collection of Data Field operations for NMCUES were performed by RTI and NORC under specifications established by the cosponsor- ing agencies. Persons in the sample dwelling units were interviewed at approximately 3-month intervals beginning in February 1980 and ending in March 1981. The Core Question- naire was administered during each of the five interview rounds to collect data on health, health care, health care charges, sources of payment, and health care coverage. A summary of responses was used to update information reported in previous rounds. Supplements to the Core Questionnaire were used during the first, third, and fifth interview rounds to collect data that did not change during the year or that were needed only once. Approximately 80 percent of the third- and fourth-round interviews were conducted by tele- phone; all remaining interviews were conducted in person. The respondent for the interview was required to be a house- hold member 17 years of age or over. A nonhousehold proxy respondent was permitted only if all eligible household members were unable to respond because of health, language, or mental condition. Weighting For the analysis of NMCUES data, sample weights are required to compensate for unequal probabilities of selection, to adjust for the potentially biasing effects of failure to obtain data from some persons or RU’s (i.e., nonresponse), and failure to cover some portions of the population because the sampling frame did not include them (i.e., undercoverage). Basic sample design weights—Development of weights reflecting the sample design of NMCUES was the first step in the development of weights for each person in the survey. The basic sample design weight for a dwelling unit is the product of four components which correspond to the four stages of sample selection. Each of the four weight components is the inverse of the probability of selection at that stage, when sampling was without replacement, or the inverse of the expected number of selections, when sampling was with replacement, and when multiple selection of the sample unit was possible. Two-sample adjustment factor—As previously discussed, the NMCUES sample is comprised of two independently selected samples. Each sample, together with its basic sample design weights, yields independent unbiased estimates of population parameters. Because the two NMCUES samples were of approximately equal size, a simple average of the two independent estimators was used for the combined sample estimator. This is equivalent to computing an adjusted basic sample design weight by dividing each basic sample design weight by 2. In the subsequent discussion, only the combined sample design weights are considered. Total nonresponse and undercoverage adjustment—A weight adjustment factor was computed at the RU level to compensate for RU-level nonresponse and undercoverage. Because every RU within a dwelling unit is included in the sample, the adjusted basic sample design weight assigned to an RU is simply the adjusted basic sample design weight for the dwelling unit in which the RU is located. An RU was classified as responding if members of the RU initially agreed to participate in NMCUES and as nonresponding otherwise. Initially, 96 RU weight adjustment cells were formed by cross-classifying the following variables: Race of RU head (2 levels), type of RU head (3 levels), age of RU head (4 levels), and size of RU (4 levels). These cells were then collapsed to 63 cells so that each cell contained at least 20 responding RU’s. Within each cell an adjustment factor was computed so that a sum of adjusted basic sample design weights would equal the March 1980 Current Popula- tion Survey estimate for the same population. The nonresponse and undercoverage adjusted weight was computed for each RU as the product of the adjusted basic sample design weight and the nonresponse and undercoverage adjustment factor for the cell containing the RU. Poststratification adjustment—Once the nonresponse and undercoverage adjusted RU weights were computed, a post- stratification adjusted weight was computed at the person level. Because each person within an RU is included in the sample, the nonresponse and undercoverage adjusted weight for a sample person is the nonresponse and undercoverage adjusted weight for the RU in which the person resides. Each person was classified as responding or nonresponding, as discussed subsequently in the section on attrition imputation. Sixty poststrata were formed by cross-classifying age (15 levels), race (2 levels) and sex (2 levels). One poststratum (black males 75 years of age and over) had less than 20 67 respondents, so it was combined with an adjacent poststratum (black males 65-74 years of age), resulting in 59 poststrata. Estimates based on population projections from the 1980 census were obtained from the Bureau of the Census for the U.S. civilian noninstitutionalized population by age, race, and sex poststrata for February 1, May 1, August 1, and November 1, 1980. The mean of these mid-quarter population estimates for each of the poststrata was computed and used as the 1980 average target population for calculating the poststrata adjustment factors. Survey-based estimates of the average poststrata popula- tion were developed using the nonresponse and undercoverage adjusted weights. First, a survey-based estimate of the target population of each poststratum for each quarter was computed by summing the nonresponse and undercoverage adjusted weights for respondents eligible for the survey on the mid- quarter date. Then the survey-based estimate of the 1980 average population was computed as the mean of the 4 mid-quarter estimates. Finally, the poststratification adjust- ment factor was computed in each poststratum as the ratio of the 1980 average target population (obtained from Bureau of the Census data) to the NMCUES 1980 average population. The poststratified weight for each respondent was then com- puted as the product of the nonresponse and undercoverage adjusted weight and the poststratification adjustment factor for the poststratum containing the respondent. Thus, the weighting procedure is composed of three steps: Development of base sample design weights for each RU, adjustment for RU-level nonresponse and undercoverage, and adjustment for person-level nonresponse and undercover- age. A further adjustment for the number of days a person was an eligible member of the U.S. civilian nonin- stitutionalized population was made, but this adjustment only affects certain types of estimates from NMCUES and is discussed in the appendix on estimation (Appendix III). Survey Nonresponse Nonresponse in panel surveys such as NMCUES occurs when sample individuals refuse to participate in the survey (total nonresponse), when initially participating individuals drop out of the survey (attrition nonresponse), or when data for specific items on the questionnaire are not collected (item nonresponse). Response rates for RU’s and person in NMCUES were high, with approximately 90 percent of the sample RU’s agreeing to participate in the survey and approxi- mately 94 percent of the individuals in the participating RU’s supplying complete information. Even though the overall response rates are high, survey-based estimates of means and proportions may be biased if nonrespondents tend to have different health care experiences than respondents or if there is a substantial response rate differential across sub- groups of the target population. Furthermore, annual totals tend to be underestimated unless allowance is made for the loss of data attributable to nonresponse. Two methods commonly used to compensate for survey nonresponse are data imputation and adjustment of sampling weights. For NMCUES, data imputation was used to compen- 68 sate for attrition and item nonresponse, and weight adjustment was used to compensate for total nonresponse. The calculation of the weight adjustment factors was discussed previously in the section on sampling weights. Attrition Imputation A special form of the sequential hot deck imputation method was used for attrition imputation. First, each sample person with incomplete annual data (referred to as a “recip- ient”) was linked to a sample person with similar demo- graphic and socioeconomic characteristics who had complete annual data (referred to as a “donor”). Second, the time periods for which the recipient had missing data were divided into two categories: Imputed eligible days and imputed ineligi- ble days. The imputed eligible days were those days for which the donor was eligible (i.e., in scope), and the imputed ineligible days were those days for which the donor was ineligible (i.e., out of scope). The donor’s medical care experiences, such as medical provider visits, dental visits, and hospital stays, during the imputed eligible days were imputed into the recipient’s record for those days. Finally, the results of the attrition imputation were used to make the final determination of a person’s respondent status. If more than two-thirds of the person’s total eligible days (both reported and imputed) were imputed eligible days, then the person was considered a total nonrespondent, and the data for the person were removed from the data file. Item Nonresponse and Imputation Among persons who are classified as respondents, there is still the possibility that they may fail to provide information for some or many items in the questionnaire. In NMCUES, item nonresponse was particularly a problem for health care charges, income, and other sensitive topics. The extent of missing data varied by question, and imputation for all items in the data file would have been expensive. Imputations were made for missing data on key demographic, economic, and charge items across five of the six data files in the public use data tape. Table I illustrates the extent of the item nonre- sponse problem for selected survey measures that received imputation in four data files used in this report. Demographic items tend to require the least amount of imputation, some at insignificant levels such as for age, sex, and education. Income items had higher levels of nonre- sponse. For total personal income, which is a cumulation of the earned income and 11 sources of unearned income, imputation was required for at least one component for nearly one-third of the persons. The bed-disability days, work-loss days, and cut-down days have levels of imputation that are intermediate to the demographic and income items. The highest levels of imputation occurred for the impor- tant charge items on the various visit, hospital stay, and medical expenses files. Total charges for medical visits, hospi- tal stays, and prescribed medicines and other medical expenses were imputed for 25.9 percent, 36.3 percent, and 19.4 percent of the events, respectively. Among the source of payment Table Percent of data imputed for selected survey items in 4 of the NMCUES public use data files: United States, 1980 Tape oats Percent location Description imputed Person file (n = 17,123) P54 Age 0.1 P57 Race 20.0 P59 Sex 0.1 P62 Highest grade attended 0.1 P67 Perceived health status 0.8 P592 Functional limitation score 3.2 P125 Number of bed-disability days 7.9 P128 Number of work-loss days 8.9 P135 Number of cut-down days 8.2 P399 Wages, salary, business income 9.7 P434 Pension income 3.5 P445 Interest income 21.6 P462 Total personal income 230.4 Medical visit file (n = 86,594) M117 Total charge 25.9 M123 First source of payment 1.8 M125 First source of payment amount 11.6 Hospital stay file (n = 2,946) H252 Nights hospitalized 3.1 H124 Total charge 36.3 H130 First source of payment 2.2 H132 First source of payment amount 17.6 Medical expenses file (n = 58,544) E117 Total charge 19.4 E123 First source of payment 2.8 E125 First source of payment amount 10.0 "Race for children under 14 years of age imputed from race of head of reporting unit. 2Cumulative across 12 types of income. data, the imputation rates for the source of payment were small, but the rates for the amount paid by the first source of payment were generally subject to high rates of imputation. Nights hospitalized on the hospital stay file were imputed at a rate comparable to the first source of payment. The methods used to impute for missing items were diverse and tailored to the measure requiring imputation. Three types of imputation predominate: Editing or logical imputations, a sequential hot deck, and a weighted sequential hot deck. The edit or logical imputations were used to elimi- nate missing data that could reasonably be determined from other data items that provided overlapping information for the given item. The sequential hot deck was used primarily for small numbers of imputations for the demographic items; the weighted sequential hot deck was used more extensively and for virtually all other items for which imputations were made. The edit or logical imputation is a process in which the value of a missing item is deduced from other available information in the data file. For example, race was not recorded for children under 14 years of age during the survey. Instead, a logical imputation was made during data processing that assigned the race of the head of the reporting unit to the child. Similarly, extensive editing was performed for the charge data before any imputations were made. If first source of payment was available, only one source of payment was given; and if total charge was missing, the value of the first source of payment amount was assigned to the total charge item. In the sequential hot deck procedure, the data are grouped within imputation classes formed by variables thought to be correlated with the item to be imputed. An additional sorting within imputation classes by other variables also thought to be correlated with the imputed item is also typically used. An initial value, such as the mean of the nonmissing cases for the item, is assigned as a “cold deck” value. The first record in the file is then examined. If it is missing, the “cold deck” value replaces the missing data code; if "real, the real value replaces the “cold deck” value and becomes a “hot deck” value. Then the next record is examined. Again, if missing, the “hot deck” value is used to replace missing data, and, if real, the “hot deck” value is replaced by that real value. The process continues sequentially through the sorted file. The weighted hot deck, a modification of the sequential hot deck, uses the weights to determine which real values are used to impute for a particular record needing an imputation. The imputation process will be described for two items to illustrate the nature of imputation for NMCUES. For His- panic origin, two different imputation procedures were used: logical and sequential hot deck. Because Hispanic origin was not recorded during the interview for children under 17 years of age, a logical imputation was made by assigning to the child the Hispanic origin of the wife of the head of the reporting unit, if present, and the origin of the head of the reporting unit otherwise. For the remaining cases that were not assigned a value by this procedure, the data were grouped into classes by observed race of the head of the reporting unit; within classes, the data were sorted by reporting unit identification number, primary sampling unit, and seg- ment. An unweighted sequential hot deck was used to impute values of Hispanic origin for the remaining cases with missing values. The imputations for medical visit total charge were made after extensive editing had been performed to eliminate as many inconsistencies as possible between sources of payment data and total charge. The medical visit records were then separated into three types: emergency room, hospital outpatient department, and doctor visits. Within each type, the records were classed and sorted by several measures, which differed across visit types, prior to a weighted hot deck imputation. For example, the records for doctor visits were classified by reason for visit, type of doctor seen, whether work was. done by a physician, and age of the individual. Within the groups formed by these classing variables, the records were then sorted by type of health care coverage and month of visit. The weighted hot deck procedure was then used to impute for missing total charge, sources of payment, and sources of payment amounts for the classified and sorted data file. Because imputations were made for missing items for a large number of the important items in NMCUES, they 69 can be expected to influence the results of the survey in several ways. In general, the weighted hot deck is expected to preserve the means of the nonmissing observations when those means are for the total sample or classes within which imputations were made. However, means for other subgroups, particularly small subgroups, may be changed substantially by imputation. In addition, sampling variances can be substan- tially underestimated when imputed values are used in the estimation process. For a variable with one-quarter of its 70 values imputed, for instance, sampling variances based on all cases will be based on one-third more values than were actually collected in the survey for the given item. That is, the variance would be too small by a factor of at least one-third. Finally, the strength of relationships between meas- ures that received imputations can be substantially attenuated by the imputation. A more complete discussion of these issues can be found in Lepkowski, Stehouwer, and Landis (1984). Appendix li Data Modifications to Public Use Files Overview of Data Changes During the preparation of this report, a number of prob- lems were discovered in the NMCUES public use files that required modification of the data. Eight sets of problems were identified: (1) Sampling weights for 68 newborns (i.e., persons born in 1980) were in error. (2) Six respondents had extremely high hospital stay charges. (3) Forty-seven respondents had health care coverage categories inconsistent with source of payment for some medical events. (4) For 173 respondents, fewer bed-disability days were re- ported than hospital nights. (Length-of-stay data were recorded in terms of the number of nights—as opposed to days—spent in the hospital.) (5) Four respondents had extremely long lengths of stay in the hospital as a result of incorrect hospital admission dates. (6) Four respondents had poverty status categories that were inconsistent with their poverty status level. (7) Nine respondents were coded as deliveries in the hospital file but had inconsistent values for other hospital stay data. (8) One respondent had duplicate hospital stay records. Details of the changes made to correct these problems may be obtained from NCHS. General information on the problems and changes is outlined below. Detailed descriptions of the specific changes are provided in the NMCUES report series by Lepkowski, etal. (to be published). 1) Records for 68 newborns were incorrectly coded as eligible for the entire survey period (all 366 days) despite a birth after January 1, 1980. These errors were corrected by changing the eligible time-adjustment factor and the person time-adjusted weight for each of the 68 records. 2) After careful examination, the University of Michigan and NCHS determined that six hospital stay records with charges of at least $90,000 were incorrect and should be changed. These six records and related information in the per- son file (e.g., hospital stay charges, total charges) were changed to conform with records in the Medicare best estimate file or with other information about each of the 6 respondents’ hospitalizations contained in the hospital stay file. 3) Discrepancies between source of payment and health care coverage were noted in the course of analysis. All of the discrepancies involved Medicare coverage. Forty-seven respondents reporting Medicare as a source of payment in " the medical visit, hospital stay, or prescribed medicine files were not properly coded as covered by Medicare. Health care coverage for the respondents was reclassified strictly according to source of payment data. Respondents originally coded as covered by private insurance but not showing private insurance as a source of payment for any services were coded as having Medicare and private insurance coverage. Where reassignment based on imputed data for source of payment would conflict with real data for health care coverage, the real data were used in preference to the imputed data. 4) For 173 cases, the value for hospital nights was greater than the value for bed-disability days. According to interviewer instructions for the NMCUES questionnaire, hospital nights should be included in bed-disability days, except for newborns. Therefore, the value of bed-disability days was adjusted to equal hospital nights for these 173 cases, a procedure used in the Health Interview Survey proc- essing. However, it does not fully compensate for the errors in recording or computing bed-disability days. It is likely that after the edit, bed-disability days are still underestimated for these 173 cases. The edit was performed without regard to the imputation status of either bed-disability days or hospital nights. 5) Four cases with discrepancies between bed-disability days and hospital days also had improperly coded hospital admission dates causing excessively long lengths of stay. In these cases, the admission dates and hospital nights were - corrected and the bed-disability days edit was not necessary. 6) Comparison of the continuous and the categorical poverty status variables on the public use file identified four respondents whose categorical poverty status was inconsistent with their continuous poverty status value. The categorical variable was changed for these four respondents to correspond to their poverty status on the continuous variable. 7) A variety of problems were discovered on nine records coded as deliveries in the hospital stay file. (a) Two deliveries were attributed to male respondents. Examination of the data files suggested that the sex variable was incorrectly coded in these two cases. The sex variable was, therefore, recoded to female. A third male delivery was actually that of the respondent’s spouse. In this case, the hospital record was reassigned and appropriate changes made in the person file for both respondents. 71 72 (b) Four hospitalizations for newborns were incorrectly coded as deliveries. These were recoded in the hospital stay file. A fifth newborn’s hospital record was attri- buted to its mother. In this case, the hospital rec- ord was transferred to the newborn, and appro- priate changes were made in the person file for both respondents. (c) One delivery was attributed to a 74-year-old woman. Following an NCHS recommendation, the response was recoded to reflect signs, symptoms, and ill-defined conditions as the admitting condition. 8) Two sets of duplicate records (four records in total) in the hospital stay file were discovered for one respondent. The two duplicates were deleted in the hospital stay file, and necessary changes were made in the person file. Three of the four records had been imputed to another respondent for reasons of attrition. No changes were made in the records for the respondent receiving the attrition-imputed records. Appendix lll Analytical Strategies Notion of an Average Population The NMCUES was a panel survey in which members of the population were followed during the panel period (i.e., calendar year 1980). The nature of a dynamic population over time influences the rules used to determine who should be followed and for how long. It also has significant implica- tions for the form of estimators for characteristics of the population during the panel period. Before discussing estima- tion strategies for NMCUES data, it is useful to review the nature of a dynamic population over time. Figure I illustrates the nature of a longitudinal population as members move in and out of eligibility. Stable members of the population appear at the beginning and at every time point during the life of the longitudinal time period. Even though these persons are termed “stable,” they may of course change residences during the panel period and may be quite Figure | Dynamic population for 12 time period panel survey Stable Leaver e——————————————————— rrr Entrant sere Mixed difficult to trace. Leavers are persons who are eligible at the beginning of a time period, but then become ineligible at some later time. Leaving may occur through events such as death, institutionalization, or moving outside the geographic boundary of the population. At the same time, new members may enter the population (i.e., entrants) through births or returns from institutions or from outside the geographic bound- ary of the population. Finally, there also will be population elements that are both entrants and leavers from the population during different time periods. The majority of the population typically will be stable in nature, but it is the entrants and leavers, persons who may be experiencing major changes in their lives, who are often of particular interest to analysts of panel survey data. In order to assure adequate coverage of all of the elements in the dynamic population considered over the entire time period, NMCUES followup rules were carefully specified to include entrants, leavers, and mixed population elements properly. As an illustration, consider a member of the Armed Forces who was in the Armed Forces on January 1, 1980, was discharged on June 1, 1980, and then became a key person (i.e., one to be followed for the rest of the year while eligible) in the NMCUES panel. Because NMCUES was designed to provide information about the civilian popula- tion, medical care use and charges during the first 5 months of 1980 for this person are outside the scope of the survey. Data about health care use and charges were not collected unless they occurred after June 1. At the same time, this person was eligible for only 7 months of the year, and he was also only “at risk” of incurring health care use or charges for 7 of the 12 months. This person thus contributes only 72 or 0.58 years of eligibility or “person year” to the study. This quantity is referred to as the “time-adjustment factor” in the documentation and throughout these appendixes. For those readers not familiar with the concept of “person years of risk,” it may be useful to consider briefly the rules that were used to determine eligibility for a given person at a given moment during 1980. There were essentially two ways of becoming eligible for or entering the NMCUES eligible population. The obvious way was to be a member of the U.S. civilian noninstitutionalized population on January 1, 1980, and hence a member of the original or base cohort about which inferences were to be made. The second way to become a member of the eligible population was to enter after January 1 through birth or through rejoining the civilian noninstitutionalized population during the year by returning 73 from an institution, from the Armed Forces, or from outside the United States. On the other hand, there were also several ways that persons who were eligible members of the population could become ineligible. Death obviously removes a person from further followup, as would institutionalization, joining the Armed Forces, or moving to residence outside the United States. For every person selected for NMCUES, information was collected to monitor the exact number of days they were eligible during the year. These eligibility periods are summarized by the time-adjustment factor on each record. The use of “person years” to form sample estimates requires careful assessment of the characteristic to be esti- mated. Estimates that use only data collected from persons during periods of eligibility (e.g., total number of doctor visits, total charges for health care) do not need to account for time adjustments. Estimates for person characteristics (e.g., total population, proportion of the population in a given subgroup) must be based on person years to obtain estimates that correspond to those for health care estimates. Some estimates require the use of the time-adjustment factor in the denominator, but not in the numerator. For example, an estimate of the mean total charge for health care during 1980 must use the total charges for health care as a numerator, without time adjustment, but the denominator must be the number of person years that the U.S. population was exposed to the risk of such charges during 1980, a time-adjusted or person-years measure. The mean in this case is actually a rate of health care charges per person year of exposure for the eligible population in 1980. When making estimates in which person years are impor- tant, the effect of the time-adjustment factor will vary depend- ing on the subpopulation of interest (see Table II). A cross-sec- tional cohort of N persons selected from the U.S. population on January 1, 1980 and followed for the entire year will contribute a total number of person years for 1980 that is smaller than N because of removals (i.e. , deaths, institutionali- zation, and so on). If entrants are added to the initial cohort during the year, the person years contributed by the initial cohort and the entrants may well exceed N, but it will still be less than the number of original cohort members plus the number of entrants. The difference between persons and person years will vary by subgroups as well. Females 20-29 years of age on January 1 are a cohort for which few additions are expected due to returns from institutions, the Armed Forces, or living abroad. And few removals are expected due to death, in- stitutionalization joining the Armed Forces, or moving abroad. On the other hand, males 80 years of age or over on January 1 will contribute a much smaller number of person years to the population than the total number of persons in the cohort at the beginning of the year, because a large number of the cohort will die during the year. Role of Weights and Imputation Estimated means and sampling errors from NMCUES for bed-disability days, work-loss days, work-loss days in bed, cut-down days, and restricted-activity days are presented in Table III. For each survey measure, separate estimates were computed using all data (i.e., both real and imputed) and using only the real data. The unweighted and weighted mean, unweighted and weighted simple random sampling standard error of the mean, and the weighted complex standard error which accounts for the stratified, multistage nature of the design are presented. For each measure, the weighted means computed using all the data and using only the real data are quite similar. This similarity is not unexpected given that the weighted hot deck imputation procedure is designed to preserve the weighted mean for overall sample estimates. The simple random sampling standard errors, however, are smaller when all data are used simply because the simple random sampling variance is inversely related to the sample size. For the complex standard error, three of the five measures have smaller standard errors when all data are used, and the other two measures show the opposite relationship. Weighting and imputation for the disability measures have little or no effect on estimated means or their standard errors for the total popula- tion because the amount of missing data for these measures is small (approximately 7 or 8 percent). For other measures that have larger amounts of missing data, imputation has larger effects. Consider the means and standard errors for total charge for a hospital outpatient depart- ment visit shown in Table IV. There were 9,529 hospital outpatient department visits (real visit records plus those generated from the attrition imputation process), and 4,841 of these have a total charge that was imputed from one of the other hospital outpatient department visit records. Thus, more than one-half of the total charges were missing for this particular medical event. Despite the large amount of missing data, the weighted means using all the data and using only real values are quite similar; weighting does not Table li Effect of person-year adjustment on counts and sampling weights, by 4 population groups: United States, 1980 Sum of sampling weights Basic weight Adjusted weight Population group Sample size Person years in thousands in thousands Total population ............................ 17,123 16,862.84 226,368 222,824 Females, 25-29 yearsofage ..................... 702 699.39 9,529 9,494 Males, 80 years of age and over .................. 113 104.05 1,384 1,274 All persons born during 1980 ..................... 251 121.02 3,560 1,713 74 Table lll Sample size, means, and standard errors for 5 disability measures, by all and real data subgroups: United States, 1980 Unweighted estimates Weighted estimates Simple Simple random random sampling sampling Complex Disability measure Sample standard standard standard and data type size Mean error Mean error error Bed-disability days ANGER o.oo vimana mimes wesw REE 17,123 5.303 0.1279 5.268 0.1269 0.1540 RoAIdata ........covvrnnerrnnnrnansnnseanns 15,777 5.253 0.1326 5.228 0.1319 0.1599 Work-loss days ANAta ... ovis 13,069 3.614 0.1221 3.696 0.1220 0.1629 Real data ..........coonvevvnnmrnnnennnsanns 11,537 3.510 0.1284 3.574 0.1277 0.1716 Work-loss days in bed AAA ......oovnrrennrrraarennr arenas 13,069 . 1.516 0.0508 1.568 0.0518 0.0592 ROAITAIA ...:us:5 evr vemmnvmemmasasitisinenee 10,970 1.530 0.0556 1.578 0.0568 0.0652 Cut-down days ARORA. cov vvnnis wns Ho vRaT Eww SRE 88 17,123 6.831 0.1681 6.881 0.1697 0.3343 ROBI AERA ..cvvnissammmmmmnnnnssbsfssbisesnss 15,724 6.609 0.1721 6.639 0.1735 0.3322 Restricted-activity days ANDAR ......cvvvveeenvrarrrensvnarsrannnns 17,213 13.746 0.2559 13.805 0.2573 0.4716 Real data ..........coveeenmnonnnennnnsennns 14,049 13.036 0.2732 13.064 0.2742 0.4658 Table IV Sample size, means, standard errors, and element variance for total charge for a hospital outpatient department visit, by data type: United States, 1980 Unweighted estimates Weighted estimates Simple Simple random random sampling sampling Complex Element Sample standard standard standard variance Datatype size Mean error Mean error error (x10~ 8) Alldata .........oonierninienans 9,529 51.86 1.030 51.61 1.018 1.914 9.87 Realdataonly ..............coonveee 4,688 52.28 1.436 52.27 1.430 2.936 9.59 imputed data ......... oceans 4,841 51.45 1.476 50.98 1.447 1.60 10.14 Real data NOLAONOT ...ovvvevnnnnnnn ee eennnenns 929 47.83 2.108 48.53 2.117 3.935 4.17 DONOF ONOB oc. sivis luisa ium mia sie imoswonm wowims fis 2,789 55.85 2.016 55.76 1.982 3.386 11.00 DONOFIWICE «oo vvvveannnneennnnnnnnns 841 48.61 3.525 49.37 3.579 4.879 10.78 DONOr3-5tmes .......oovvvevennnnnnnns 120 29.45 7.340 28.97 7.987 11.64 7.66 affect the estimated means. However, sampling errors are changed substantially when imputed values are added to real values to form an estimate. The weighted and unweighted simple random sampling standard errors are markedly smaller for all data than for the real data. To investigate whether this decrease in sampling error is due to changes in sample size, changes in the element variance, or both, the element variances were computed by multiplying the weighted simple random sampling variances by the sample sizes. Inspection of Table IV suggests that the element variances are quite similar using all data and real data; the differences in standard error when all data and only real data are used can mostly be attributed to the loss in sample size when going from all data to real data. Not all of the real data were used as donors for imputation, and some of the real values were used as donors several times. Table IV also suggests that those real values not used as donors have a lower mean total charge than those used as donors, but values used as donors more than twice 75 tend to have even smaller mean total charges. The means for donors used once, twice, or more frequently are a function of the use of imputation classes within which the mean total charge and the amount of missing data varied. The difference in complex standard errors between all data and the real data in Table IV illustrates the large effects of imputation. However, neither the complex standard error computed using all the data nor that computed using only the real data is the correct standard error of the weighted mean estimated using all the data. The mean computed using all data includes 4,841 values that were actually subsampled with replacement from the 4,688 real values. In addition, the imputations were made across the primary sampling units and strata used in the sample selection process and in the variance estimation procedure. The assumption that the obser- vations were selected independently between primary sampling units and strata, which is needed to justify the variance estimation procedure, is incorrect. Hence, the complex stand- ard error for all data shown in Table IV fails to account for two sources of variability present in estimates based on all data: The double sampling used to select values for imputation and the correlation between primary sampling units and strata induced by imputation. At the same time, the complex standard error for the weighted mean computed using only the real data is an incorrect estimate of the standard error of the mean based on all the data. The actual sampling error of the weighted mean for all the data is probably larger than that shown for the mean estimated using all the data; it may even be larger than the sampling error computed using only the real data. As a final illustration of the effects that imputation can have on survey results, Figure II presents estimated mean charges per hospital outpatient department visit for four family income groups computed using all the data and using only the real data. For the real data, the mean charge per visit Figure II. Estimated mean charges per hospital outpatient department visit, by 4 family income classes for all and real data: United States, 1980 65 [ Data type 60 — | Al ~ O Real / & | / / 40 |— od 5 I Less than $5,000- $12,000- $35,000 $5,000 $11,999 $34,999 or more Family income 76 Figure lll Estimated mean charge per hospital outpatient department visit, by 16 imputation classes for all persons and for persons in families with income less than $5,000: United States, 1980 120 — Legend MW Total O Family income less than $5,000 P 8 | 3 & 8 3 / olivia Si 123 456 7 8 9 101112 13 14 15 16 Imputation class increases in a linear fashion as the family income increases. However, when all the data are used to estimate the mean charge per visit, the mean charge does not increase as rapidly with increasing family income. The strong relationship be- tween family income and mean charge per hospital outpatient department visit in the real data has been attenuated by the imputed values. The reason for this attenuation is shown in Figure III. Sixteen imputation classes were formed for the imputation of total charges for hospital outpatient department visits. Figure III shows mean charge for real data for the total sample and the subgroup with family incomes less than $5,000 in 1980. The low income group has lower mean charges than the total sample. Because family income was not one of the variables used to form imputation classes, low family income persons within an imputation class with missing hospi- tal outpatient department visit total charges were imputed a charge that was, on average, higher than the mean charge for low income persons with real data. This occurs in almost every imputation class. When the real and imputed data are combined for persons with family incomes less than $5,000, the effect of imputation is to increase the mean charge for this subgroup. Conversely, for persons with family incomes of $35,000 or more, total hospital outpatient depart- ment visit charges for persons with real data tend to be larger than values imputed to persons with missing charges. The overall impact of the imputation process on the relation- ship between charges for hospital outpatient department visits and family income is a regression toward the mean charge for real data for low and high income subgroups. The results in Tables III and IV and Figure II demonstrate the effect that imputation can have on estimated means, on estimated sampling errors, and on relationships between variables. Several strategies for handling imputation in estima- tion are suggested by these findings. It is beyond the scope of this discussion to evaluate various strategies and indicate the reasons why one was chosen for this report. The strategy used in preparing estimates in this report was, despite the sizeable effects due to imputation noted here, to use all the data in all estimates. This strategy means that estimated means and totals presented in the report have been adjusted for item nonresponse, but sampling errors and relationships among some variables may be adversely affected by the imputation process. The reader should keep in mind that sampling errors for estimates that are subject to large amounts of item nonresponse may be substantially underestimated, and the strength of relationships between a variable receiving imputed values and a variable that was not used to form imputation classes may be attenuated by the imputation process. Estimation Procedures Sample estimators from NMCUES data, regardless of whether they are totals, means, medians, proportions, or standard errors, must account for the complexity of the sample survey design. Totals, means, or other estimates must include sampling weights to compensate for unequal probabilities of selection, nonresponse, and undercoverage. Stratification, clustering, and weighting must also be accounted for in the estimation of sampling errors. In addition, one must consider time-adjustment factors to account for persons not eligible for the entire year and imputations which were made to compensate for missing items. The weighting adjustment fac- tors, the imputations, and the estimated sampling errors for estimates with imputed values affect the results shown in this report. A variety of estimators were used for the descriptive analyses. To illustrate the role of time adjustments, we consid- er six specific estimates that were used in the analysis: 1. An estimated total charge for a selected subgroup (e.g., high-volume users). An estimated total population. The mean charge per visit. The mean charge per person. The proportion of charges that fall in a certain range of charges. 6. The proportion of persons whose charges are less than or equal to a fixed level. To define these estimators, suppose we introduce the following notation for these quantities for the ith person: y; = total charges for health care in 1980, x; = total number of medical visits for 1980, w; = nonresponse and undercoverage adjusted person weight, 1; = time-adjustment factor (i.e., the proportion of days in 1980 that the person was an eligible member of the population), 1, if total charges are less than or equal to d= a fixed value, 0, otherwise, 1, if the total charge is between two fixed e= values, 0, otherwise, and 1, if the ith person is a member of a designated d= subgroup of the population, 0, otherwise. Estimating total charges or any quantity from NMCUES which was recorded only during periods when the person was a noninstitutionalized civilian in the United States, is a relatively straightforward task requiring only a weighted “sum of charge values. In particular, b= y ws, is the estimated total charge for a particular service for a selected subgroup. On the other hand, for estimates of total population, a time-adjusted estimator is required such as y= Y LA Thus, §’ denotes an estimate of the 1980 average subgroup population, while j denotes the 1980 charges by a subgroup of the noninstitutionalized civilian population. Estimated means may or may not need to include a time-adjustment factor in the denominator. For example, to estimate the mean charge per visit during 1980, no time adjustment is needed. Hence, y= Y wy y WX, can be used to estimate mean charge per visit. However, to estimate mean charge per person, a time adjustment is required in the denominator, because the denominator is actu- ally an estimate of the total average population in 1980. In particular, the estimator has the form Jy = Y wy,/ Y wit. Estimates of mean charges for subgroups have a similar form with the indicator variable &; included in the numerator and denominator for the appropriate subgroup of interest. Estimated proportions are, of course, means with an indicator variable in the numerator and a count variable in the denominator. Proportions may also have time adjustments not only in the denominator, but also in the numerator. For example, to estimate the proportion of persons who had charges less than or equal to a fixed value, an estimate of the form 77 = Y wit, / y wit, was used. Appropriate indicator variables were added to the numerator and denominator to make estimates for selected subgroups. On the other hand, the estimated proportion of total charges between two fixed levels of charges does not require time adjustments in the numerator or the denominator. In particular, p= y we / 3 wi is the estimated proportion of all charges for persons that occurred between two levels of charges. The Logistic Regression Model One of the statistical methods used in this report is logistic multiple regression. The methodology is used to iden- tify characteristics of individuals in the U.S. civilian nonin- stitutionalized population in 1980 that are predictive of use of hospital services or of high use of hospital services. Logistic regression is closely related to the standard regression methods but was used here because of the nature of the dependent variable in the models, a measure with only two possible outcomes (i.e., yes or no). In the standard regression situation, the dependent variable Y is intervally scaled (i.e., continuous in distribution). The regression methodology is used to examine independent vari- ables X,, X,, . . . | X, which are statistically important predictors of the dependent variable Y. The ordinary regression model predicts Y as a linear combination of the Xs, that is, Y=Bo+BXi + BX +... + BX, +¢ where & is a random variable with mean zero and equal variance for all sample observations. The parameters f, Bi... B,, are often referred to as the slopes or (partial) regression coefficients and are estimated in the regression analysis. Logistic regression methods are applied when the depen- dent variable is not continuous but is dichotomous (1: Y is only one of two possible values). Use of the standard regression methodology when the dependent variable is dichotomous is inappropriate for a variety of reasons: Besides violating a number of important assumptions of the regression model, predicted values for Y can fall outside the range of Y. The logistic regression approach addresses these deficien- cies by examining not the dependent variable ¥ but the logit of the probability that the dependent variable assumes one of the two possible values. Suppose that ¥ can assume only two values, 1 or 2: The logit is defined to be the logarithm of the ratio of the probabilities that Y is 1 or 2: Logit (Y) = In [Pty = 1}/PlY = 2] 78 where In[*] denotes the natural logarithm of the argument []. The logistic regression model then predicts the logit of ¥ as a linear combination of the X's: Logit (Y) = By + B,X, +8,X, +... + B,X, The B; are again slope or (partial) regression coefficients, but in this case denote the partial linear regression of the logit of ¥ on X; given that the other X’s are in the model. Several methods are used to obtain estimates of the coefficients including an iterative method to derive a maximum likelihood estimate. An alternative interpretation of these logistic regression coefficients can be made by observing that the logit of Y is actually the logarithm of the odds that an individual will be classified as ¥ = 1 rather than ¥ = 2. Suppose, for example, that Y = | corresponds to the event that an individual is hospitalized in 1980 and ¥ = 2 to the event that the individual is not hospitalized. Then P{Y = 1} / P{Y = 2} denotes the odds that an individual will be hospitalized during the year and the logit is simply the logarithm of the odds ratio. The logistic regression coefficients can be translated into statements about the ratios of odds for two different individuals with different characteristics X,. Consider the following three indicator variables: I, if the individual is under 35 years of age, X, = 0, otherwise, 1, if the individual is 55-74 years of age, X; = 0, otherwise, and 1, if the individual is 75 years of age and over, X; = 0, otherwise. These three indicators combined use the individuals 35-54 years of age as a reference or comparison group against which the odds of being hospitalized, for example, are con- trasted. The logistic regression coefficients represent the unit change in the log odds that occurs for an individual in one of the three age groups defined by these indicators relative to the age group 35-54 years. For example, in Table B the logistic regression coefficient corresponding to X, is 0.1510. Persons under 35 years of age have somewhat higher odds of being hospitalized than persons 35-54 years of age (the reference group) when all the other variables in the model in Table B are controlled at an average value. Similarly, persons 55-74 years of age and 75 years of age and over have higher odds of being hospitalized than persons 35-54 years of age. The coefficients can be used to predict the probability of hospitalization or high use of hospital care for a hypothetical person. The estimation of predicted probabilities requires three steps: 1. Identify characteristics of the hypothetical person in terms of characteristics in the model. 2. Identify coefficients that apply to that individual to obtain a predicted logarithm of the odds ratio. 3. Convert the predicted logarithm of the odds ratio to obtain the predicted probability. For example, suppose that it is desired to estimate the predicted probability of hospitalization in 1980 for a hypotheti- cal individual using the estimated coefficients in Table B. Following the characteristics in Table B, suppose the person is a male, white or other race, reported good health status, 35-54 years of age, income that is 200 to 400 percent of poverty level, with private health insurance, residing in the northeast region, and with a usual source of care. The logarithm of the odds ratio is estimated by summing the appropriate coefficients from the table. For characteristics that have two levels, such as sex, either a coefficient will be included in the sum or nothing will be added. For sex, for instance, males are the reference group, and the hypotheti- cal person would not have a coefficient added to the logarithm of the odds ratio sum. On the other hand, for race the reference group is the black race, and the hypothetical indi- vidual would have the coefficient 8 = 0.0948 would be added to the sum. For characteristics with several levels, the coefficient for the reference group in the table (enclosed in parentheses) is zero, while the coefficients for the other groups of the characteristic are nonzero. Thus, for the hypothetical indi- vidual reporting good health status, the reference group is good health status and the health status coefficient for the person is zero. On the other hand, for type of health care coverage, the reference group is none or some mixture of part-year coverages. Because the person has private coverage, the private insurance coefficient B = 0.3468 would be added to the sum. The coefficients for poverty level must be handled in a slightly different way. The coefficient for the group with income that is 700 percent or more of the poverty level is determined by adding the negative value of the sum of the other coefficients to the logarithm of the odds ratio. For the hypothetical person with income 200-400 percent of the poverty level, the contribution of poverty level to the logarithm of the odds ratio for that person would be simply the value of the coefficient for that group, i.e., B = —0.0728. If the hypothetical person had an income that was 700 percent or more of the poverty level the contribution of poverty level would be — (0.1483 + —0.0728 + —0.1191) = —0.0436. Once the coefficients of each separate characteristic are determined and combined, the estimated logarithm of the odds ratio is calculated by adding the sum of coefficients to the constant term. This coefficient represents the effect for a person who has all the reference group characteristics. Thus, the logarithm of the odds ratio for the hypothetical individual is computed as the sum of the coefficients for the constant term, for the white and other race category, for the category with income 200-400 percent of the poverty level, for private insurance coverage, and for persons with a usual source of care: (—2.9031) + (0.0984) + (— 0.0728) + (0.3468) + (— 0.0608) = —2.5915. The last step to obtain a predicted probability of hospitali- zation for the hypothetical individual is to take the natural exponent of the logarithm of the odds ratio to obtain the, estimated odds ratio: [p/(1 = p)] = exp (— 2.5915) = 0.0749. Finally, the estimated odds ratio is converted to an estimated probability as 0D ow 0.0736. = Il 1 + 0.0749 That is, the hypothetical individual has predicted probability of hospitalization in 1980 of 7.4 percent. The logistic regression coeffcients for these indicator vari- ables can also be interpreted in terms of a ratio of odds. For example, using the three indicator variables for age, the function e # is the ratio of the odds of hospitalization for an individual under 35 years of age to the corresponding odds that an individual 35-54 years of age is hospitalized. In Table B, this odds ratio is 1.1630 suggesting that persons under 35 years of age have approximately 16 percent higher odds of hospitalization than persons 35-54 years of age. The logistic regression coefficients presented in this report were computed using a program for logistic regression analysis in the OSIRIS statistical software system (Computer Support Group, 1982) called DREG (Dichotomous REGression). The program uses an iterative maximum likelihood method to estimate the logistic regression coefficients and has a feature that incorporates sampling weights directly into the estimates. The program also estimates standard errors for the logistic regression coefficients, but these estimated standard errors are computed under the assumptions of simple random selections. At present, there is no accessible software for estimating the standard errors of logistic regression coefficients under a complex sample design. An ad hoc procedure was used to adjust the estimated standard errors from the DREG program to account for the complex NMCUES sample design. For each logistic regression model, the identical model was esti- mated using standard regression methods for complex sample designs in which estimated standard errors are computed by balanced repeated replication methods (Frankel, 1974). For each coefficient, the ratio of the actual standard error to the corresponding standard error computed under the as- sumptions of independent sample selection is computed. As- suming that the same ratio for logistic regression coefficients is similar, the estimated standard errors from the DREG 79 program were multiplied by the ratio from the standard regres- sion method to obtain an adjusted standard error for the logistic regression coefficient. These adjusted standard errors were used to create confi- dence intervals for the estimated odds ratios computed from the logistic regression coefficients. For each coefficient, a 95-percent confidence interval was computed by adding to and subtracting from the estimated coefficient 1.96 times the adjusted standard error. The upper and lower confidence limits for the logistic regression coefficients, 8, and Br, were then converted to upper and lower confidence limits for the odds ratio by the transformations e’’ and e™. For example, in Table B, the estimated odds ratio for persons under 35 years of age (relative to persons 35-54 years of age) is 1.1630 with 95-percent confidence limits from 1.0321 to 1.3105. Thus, with 95-percent confidence one can conclude that the odds of being hospitalized for an individual under 35 years of age are greater than for an individual 35-54 years of age. Finally, in standard regression analysis a useful measure for assessing the goodness of fit of the model to the data is the proportion of variance explained by the model, referred to as the multiple correlation coefficient or R2. A parallel measure can be developed for the logistic regression methodol- ogy. For each individual in the sample, the probability that that individual was hospitalized, for example, can be estimated from the logistic regression coefficients by substituting the individual’s values for each X; into the model, computing the log odds of hospitalization for that individual, and convert- ing the log odds into the corresponding probability. For a particular individual, suppose that they were hospitalized, 1e., Y = 1. The probability that this individual was hos- pitalized under the logistic regression model can then be computed. Similarly, for every other individual in the sample, the probability that they were hospitalized or not hospitalized, as the case may be for each person, can be computed. 80 Probabilities close to one denote that the predictive power of the model is good for that individual, while probabilities close to zero suggest that the predictive power of the model is poor. As an overall assessment of the predictive power of the model, a mean of the predictive powers for all sample individuals can be computed. In this case, a simple average is not appropriate since the logistic regression model operates on a logarithmic scale rather than a linear one as in the standard regression method. Let P(Y,=y) denote the predicted probability from the logistic model that for the ith individual y; is the observed value. The predictive power of the model over all individuals can be computed as the geometric mean of these probabilities, f= [ mP(Y;=y,) I where n denotes the sample size. If all individuals’ observed values are predicted well from the model, the probabilities By, = yi) will be close to one and 7 will be close to one. Since 7 is a measure of how well the model fits the data, then i, = 1 — 7 can be used as a measure of the error associated with the model. Without any of the predictors X,, X,, . . . , X,, the logit of Y would be predicted by an intercept-only model. The importance of the X/s to prediction of the logit of Y can be assessed by examining how much of the predictive error for the mean-only model is accounted for when the predictors Xin Xa, «+. , X,, are added to the model. The proportion of predictive error m, accounted for by the addition of these predictors is a measure corresponding to the multiple correla- tion coefficient R* in standard regression analysis. Such a measure was used in this report to examine alternative models and assess whether the predictors were adding predictive power to the model. Appendix IV Sampling Errors The NMCUES sample was one of a large number of samples that could have been selected from the U.S. civilian noninstitutionalized population using the same sampling proce- dures. Each of the possible samples could have provided estimates that differ from sample to sample. The variability among the estimates from all the possible samples that could have been selected is defined to be the standard error of the estimate, or the sampling error. The standard error may be used to assess the precision of the estimate itself by creating a confidence interval. These intervals have a specified probability that the average estimate over all possible samples selected from the population using the same sampling proce- dures will be in the interval. Preparation of sampling errors for every estimate in this report is a sizeable task. A more difficult task, though, is to find a way to present sampling error estimates that would not greatly increase the length of the report or would not make it difficult to distinguish the estimates from their standard errors in a single table. Rather than compute and display standard errors for every estimate in this report, standard errors were computed for a subset of estimates. A set of functions was fitted to these estimated standard errors to determine whether a model could be identified that would allow computation of standard error using the function that would be reasonably close to the estimated standard error. This appendix is designed to provide the reader of the report with these summary formulae derived from the esti- mated standard errors, which can be used to approximate the standard error for any given estimate in the report. The formulae have been designed to allow the reader to compute an estimated standard error using an electronic calculator with basic arithmetic operators and a square root function. The computed estimate will be an average or smoothed esti- mate of the actual standard error of the estimate. The formulae for standard error estimates are presented for three types of estimates found in the report: 1. Totals or aggregates (e.g., total charges for ambulatory visits made during 1980; total person years for males). 2. Means (e.g., mean charge per ambulatory visit). 3. Proportions and percents (e.g., percent of persons having no hospitalizations during 1980). The reader also may be interested in making comparisons between point estimates from two different subgroups of the population. Formulae are also given for computing stan- dard errors for two types of comparisons that are made in this report: a. Comparisons of two mutually exclusive subgroups (e.g., comparing the percent of females who were low users of hospital services and the percent of females who were high users, where low-user and high-user subgroups have no members in common). b. Comparisons between a subgroup and a larger group in which the subgroup is contained (e.g., comparing the mean charge per hospital day for high users of hospital services and for all users of hospital services). The standard error of a difference is based on the standard error of the totals, means, proportions, or percents of interest and ignores certain covariances between estimates that typi- cally are small relative to the standard errors of the estimates themselves. The standard errors calculated from the formulae in this appendix can be used to form intervals for which confidence statements can be made for estimates from all possible samples drawn in exactly the same way as NMCUES. The confidence level is determined by multiplying the estimated standard error by a constant derived from the standardized normal probability distribution. In particular, for the estimate 6 with estimated standard error Sj, the upper limit for a (1 — a) X 100-percent confidence interval can be formed by adding z,, times Sg to #; the lower limit is formed by subtracting z 42 times Sg from #. The value of z,,, is obtained from the standard normal probability distribution. For example, a 95-percent confidence interval corresponding to a = 0.05 can be formed with zg 02s = 1.96; a 99-percent confidence interval (i.e., a = 0.01) uses zg 00s = 2.346. Illustrations of these calculations are provided in the discussion section for each formula. : A final feature of such confidence intervals for com- parisons of estimates between two subgroups is the ability to make inference about whether the difference is statistically significant. If a (I — a) X 100-percent confidence interval does not include the value zero, one can conclude that the difference is significantly different from zero. Totals Let § denote the estimated total or aggregate for which a standard error is desired. The standard error for the estimate can be calculated by the expression oo” 0 62 |1/2 Sy= [as +5" 81 where a and b are constants chosen from Table V for the particular estimate of interest. This formula was derived from a study of the relationship between the estimated total § and its standard error Ss; in which a parabolic or quadratic relation- ship was observed. As an illustration of the use of this formula, suppose that the standard error of the estimated number of persons having no ambulatory visits during 1980 is needed. From Table 1, y = 46,716,000, the estimated total number of person years accumulated in 1980 by persons having no am- bulatory visits during 1980. From Table V, we obtain the coefficients a = 25,011 and b = 0.00048043 to use in the formula to calculate the standard error of y. The estimated standard error is then computed as Sp = [@5.011)46,716,000 + (0.00048043)(46,716,000] - [1.1684 x 1012) + (1.0484 x 103] — 1,488,925. This estimated standard error for the total y can be used to create confidence intervals for the number of persons having no ambulatory visits. For example, a 68 percent confidence interval can be obtained by adding and subtracting the standard error from the estimate. In this case, in 68 out of 100 samples drawn exactly in the same way as the NMCUES, the estimated number of persons having no ambulatory visits during 1980 will be between 45,227,075 and 48,204,925. Similarly, a 95 percent confidence interval can be obtained by adding and subtracting from the estimate 1.96 times the standard error. Thus, for 95 out of 100 samples drawn in the same way as the NMCUES, the estimated number of persons having no ambulatory visits would be between 43,797,707 and 49,634,293. Table V Coefficients for standard error formula for estimated aggregates or totals Coefficient Estimator a b Personyears ............ 2.5011 x10* 4.8043x10 Charges ................ 1.0986 x 10° 4.5524x10 4 Visits or acquisitions . . . .. .. 4.6408 x 10° 5.7634x10 Means A number of means for different types of measures are presented in this report. Despite the variety of measures pre- sented, a single formula is recommended for calculating an estimated standard error for a mean. The formula given here is based on the assumption that the standard error of the mean is determined by two quantities, the population variance and the effect of the sample design on the variances. The population variance for weighted survey data with weights w; is estimated as 82 2 ) wy; =)? BR emissions: § u-n3 where n is the size of the sample, y; denotes the value of the characteristic ¥ for the ith sample person, and y is the weighted sample mean. The effect of the sample design on the variance of a sample mean is called the design effect or ‘deff’ (Kish, 1965), and is often expressed as deff = (1 + [(n/a) — 1] roh), where a is the number of clusters in the sample design and roh is a measure of within cluster similarity among observa- tions from the same cluster. The formula recommended for estimating the standard error of a mean in this report is a function of both the estimated population variance and the design effect. In particular, the estimated standard error for a mean y can be calculated as a2 Sy = [ def 3 ] - n a2 _ I 8277 =[ a+ gs ger - row 5-1" where 7 is the estimated population total for the subgroup under consideration and 1,795,637 represents the number of clusters (@ = 138) times the average basic person weight. Con- sequently, 7/1,795,637 is an estimator for n/a in the expression for deff. The values of roh and §° for a variety of means appearing in this report can be obtained from Table VI. The table provides, for example, values of roh and §* for mean charges and mean utilization measures of various types. As an illustration, suppose that the standard error of the mean number of hospital days for high users of hospital services is needed. From Table 1, y = 38.5, and from Table VI under the entry for ‘Mean visits per person, Hospital days’ the values roh = 0.013098 and §*> = 8.5018 Xx 10° are obtained. There were an estimated 74 = 3,837,000 persons who were high users of hospital services. Substituting these values into the expression for S;, a a 3,837,000 8.5018 x.10° 1/2 s=[ [1 + (795637 ~ 1)(0.013098)] 5000 | [U0 + (2.1368 — 1)(0.013098)] 0.22159] [(1.0149) (0.22157)]'/2 = 0.47420. That is, the standard error of the mean number of hospital days for persons with high use of hospital services is 0.5. Approximate confidence intervals may be constructed for the population mean by adding to and subtracting from the estimated mean a constant times the estimated standard error. For example, to form a 95-percent confidence interval for the estimated mean number of hospital days for persons with Table VI Values of roh and §* for standard error formula for estimated means Estimator roh § Mean charge per visit Ambulatory visits... ................ 0.018777 2.4613x107 HospRal days: cus sussmmvmenpnmpas 0.018777 1.9938 x 10° Prescribed medications .............. 0.018777 3.8099 x 10° Mean visits per person Ambulatory visits. .................. 0.048246 1.6398 x 10° Hospital days ...................... 0.013098 8.5018 x 10° Prescribed medications .............. 0.048246 1.6651 x 10° Mean charge per person Ambulatory visits. .................. 0.029644 2.5650 x 10° HOSA SAYS vis mmmmammny ive «5 v » 0.029644 6.1652x10'° Prescribed medications .............. 0.029644 2.4323x10°® high use of hospital services, 1.96 times the estimated standard error is added to and subtracted from the estimated mean y = 38.5. In this case, the 95-percent interval ranges from 37.6t039.4. When the estimated sample size is about the same size or smaller than the constant 1,795,637 in the standard error formula, the design effect effectively becomes equal to one. Thus, when i < 1,795,000, the design effect portion of the standard error formula is not necessary, and the estimated standard error can be calculated simply as Sp= [$71] where §? is again chosen from Table VI. Proportions and Percents The standard error of a proportion is computed using a formula similar to that recommended for the standard error of a mean. Let p denote the estimated proportion for which a standard error is needed. The standard error for p is calculated as A — nn — $3=[ (1+ (Tos. where 7 is the estimated sample size on which the proportion is based, roh is a value selected from Table VII, and the constant 13,012 is the average time-adjusted weight for all persons in the sample. For proportions, the population variance can be estimated simply as 1) roh) A301 a = he, §2=p(-D), and hence, can be estimated directly from the sample propor- tions themselves (i.e., no value of §% is needed in Table VII). The design effect, the ratio of the actual sampling variance for the estimated proportion to the variance that would be achieved for a simple random sample of the same size, is calculated for proportions in the same way it was calculated for means. As an illustration of the use of the formula for Sj, consider obtaining the standard error for the proportion of persons (p = 0.572) who had no ambulatory visits during 1980 and rate their health as excellent (see Table 18). To calculate the standard error for percents, the same formula may be used as for proportions, after the percent has been divided by 100. There are an estimated i = 46,716,000 persons having no ambulatory visits (see Table 18), and roh = 0.0057805 is obtained from Table VII. Substituting these values into the formula for §;, 46,716,000 Sym [[ 1+ (e700 1) (0.0057805) | 13,012-(0.572)(1 — 0.572) | 1/2 46,716,000 _ [[ 1+ (25.016)(0.0057805) | 3,185.5 12 46,716,000 = [1.1446) (6.8189 x 10-9] =0.0088345. Because S;=0.0088345 is the estimated standard error for the proportion p=0.572, simply multiply S; by 100 for a standard error of 0.88345 for the percent 57.2. An approximate 95-percent confidence interval for the percent can now be calculated by adding to and subtracting from the estimated percent 1.96 times the estimated standard error. In this case, the 95-percent interval ranges from 55.5 to 58.9 percent of those persons having no ambulatory visits during 1980 rating their health as excellent. When the estimated sample size is less than or equal to 1,795,637, the design effect is close to one and the formula can be simplified to [13.0125 (1=p)12 sp=[ 2552] n as described for the standard error of a mean in the previous section. For example, 88.1 percent of persons 65 years of age or over having high use of hospital services during 1980 received Medicare supplemented by other health care Table VII Values of roh for standard error formula for estimated proportions Estimator roh Person years Proportion of users of hospital days ..................... 0.0057805 Proportion of users of ambulatory services ............... 0.0057805 Proportion of users of prescribed medicines .............. 0.0057805 Demographic subgroups (e.g., age, race, sex) ............ 0.069992 83 coverage (see Table 14). For the iA = 1,580,000 estimated persons in this subgroup (see Table 14), the standard error of the proportion associated with this percent is estimated as 13,012-(0.881)(1 — 0.881) 1/2 _ [ 1,580,000 ] = 0.029384 . A 95 percent confidence interval for the estimated percent is calculated by multiplying this estimated standard error by 100-(1.96) = 196 and adding the result to and subtracting the result from the percent. Thus, the 95 percent interval ranges from 82.3 t0 93.9 percent. Mutually Exclusive Subgroup Differences Many comparisons between the same estimate for two different subgroups in the population are made in this report. Let d=6, — 6, denote the difference between two subgroup estimates, where , and 6, are the estimates for the two subgroups. For example, suppose that the proportion of per- sons with high use of hospital services having family incomes less than $15,000 is to be compared with the proportion of persons with low use of hospital services having family incomes less than $15,000 (see Figure 5). Then, 6)=p,=0.564 for high users, 6,=p,=0.408 for low users, and d= Pi —P2=0.156. The standard error of this difference is computed as Sg = [5 +8 | & where 5%, and 5%; are the estimated sampling variances for 6, and 6,, respectively. (This formula ignores the nonzero covariance between 6, and 8, that arises in complex samples such as the NMCUES. This covariance is typically positive and small relative to the variances themselves. Ignoring the covariance will result in standard errors for differences that are on average somewhat larger than the actual standard errors.) From Table 5, 4, = 3,837,000 and A, = 5,242,000, and from Table VII, roh =0.0057805. Hence, _ 3,837,000 Sp, = [[ + (T75'637 — 1)(0.0057805) | 13,012-(0.564)(1 — 0.564) 1/2 3,837,000 = 0.028972 . 5, = [[! + (242,000 1) (0.005780) | by 1,795,637 13,012+(0.408)(1 —0.408)7 1/2 5,242,000 = 0.024621 . Hence, the standard error of the difference is computed as 84 Sy [00289722 + (0.024621)”] 2 = 0.038021 . This standard error can be used to form an approximate confidence interval for the difference in the same manner described previously for estimates of totals, means, propor- tions, and percents. In this instance, the 95 percent confidence interval is from 0.081 to 0.231. Since this interval does not include the value zero, one could conclude with 95-percent confidence that the proportion of persons having family in- comes less than $15,000 differ for the two use groups. In other words, the chances are only 5 out of 100 that the difference over a large number of identical surveys will be equal to zero. : Subgroup to Total Group Differences Another type of comparison made in this report is between an estimate for a subgroup and the same estimate for a group which contains the subgroup. Let d= 6, — denote the differ- ence between a subgroup estimate and the estimate for a group in which the subgroup is contained, where 6, is the subgroup estimate and 6; is the estimate for the larger group. The standard error of the difference is computed as Sy = Si! = (ir /p)] 2, where Sg denotes the standard error of the estimator 6,, and fi, and Ar denote the estimated sample sizes for the subgroup and for the larger group, respectively. (This formula is based on an assumption that the covariance between 6, and 6, is the same as the variance of 6, G.e., 53). This assumption re- sults in an estimated standard error for the difference that is on average somewhat larger than the actual standard error.) For example, suppose that the standard error of the differ- ence between the proportion of high users of ambulatory care living in the South and the proportion of all persons living in the South is needed. From Table 9, 6,=p,=0.208, 6r=pr=0.312, A, = 10,024,000, and A, = 222,824,000. Using the formula for estimating the standard error of the proportion and the value from Table VII (i.e., roh = 0.0057805), 10,024,000 $5, = [[! + (ED. 1)(0.0057805) | 13,012-(0.208)(1 —0.208)] 1/2 10,024,000 = 0.014816 . Hence, the standard error of the difference, d=0.208 —0.312= —0.104, is computed as $= 0.014816 1 - (10,024,000 / 222,824,000) = 0.014479 ‘ A 95-percent confidence interval can be constructed for the difference by adding to and subtracting from the estimated difference 1.96 times the estimated standard error of the differ- ence. In this instance, the 95-percent confidence interyal is from — 0.132 to — 0.076. One would conclude with 95-percent confidence, that fewer high users of ambulatory care live in the South, because this confidence interval does not include Zero. 85 Appendix V Definition of Terms Age—The age of the person as of January 1, 1980. Babies born during the survey period were included in the youngest age category. Ambulatory care visit—A direct personal exchange be- tween an ambulatory patient and a health care provider. The visit may have taken place in the provider's office, hospital outpatient department, emergency room, clinic, health center, or the patient’s home. Services may have been rendered by a physician, chiropractor, podiatrist, optometrist, psychologist, social worker, nurse, or other ancillary personnel. Average length of stay—The average length of stay is the total number of hospital days accumulated at time of discharge by patients discharged during the year divided by the number of patients discharged. Bed-disability day—A bed-disability day is one in which a person stays in bed for more than half of the daylight hours because of a specific illness or injury. All hospital days for inpatients are considered to be bed-disability days even if patient was not actually in the bed at the hospital. Condition—Any entry on the questionnaire that describes a departure from a state of physical or mental well-being. It is any illness, injury, complaint, impairment, or problem perceived by the respondent as inhibiting usual activities or requiring medical treatment. Pregnancy, vasectomy, and tubal ligation were not considered to be conditions; however, related medical care was recorded as if they were conditions. Neo- plasms were classified without regard to site. Conditions, except impairments, are classified by type according to the Ninth Revision of the International Classification of Diseases (World Health Organization, 1977) as modified by the National Health Interview Survey Medical Coding Manual (NCHS, © 1979); these modifications make the code more suitable for a household interview survey. Impairments are chronic or permanent defects, usually static in nature, that result from disease, injury, or congenital malformation. They represent decrease or loss of ability to perform various functions, particu- larly those of the musculoskeletal system and the sense organs. Impairments are classified by using a supplementary code specified in the coding manual. In the supplementary code, impairments are grouped according to type of functional im- pairment and etiology. Disability—Disability is the general term used to describe any temporary or long-term reduction of a person’s activity as aresult of an acute or chronic condition. Disability day—Short-term disability days are classified according to whether they are days of restricted activity, bed- disability days, hospital days, or work-loss days. All hospital 86 days are by definition days of bed disability; all days of bed disability are by definition days of restricted activity. The converse form of these statements is, of course, not true. Days lost from work applies only to the working population, but these too are days of restricted activity. Hence, restricted-activity days is the most inclusive term used to describe disability days. Education of head of family—The years of school com- pleted by the head of family, when the family head was 17 years of age and over. Only years completed in regular schools, where persons are given a formal education, were included. A “regular” school is one that advances a person toward an elementary or high school diploma or a college, university, or professional school degree. Thus, education in vocation, trade, or business schools outside the regular school system was not counted in determining the highest grade of school completed. Family—A group of people living together related to each other by blood, marriage, adoption, or foster care status. An unmarried student 17-22 years of age living away from home was also considered part of the family even though his or her residence was in a different location during the school year. Family head—At the time of the first interview, the respondent for the family was asked to designate a “family head.” If no head was designated or this information was missing, a family head was imputed. Family income in 1980—Each member of a family is classified according to the total income of the family of which he or she is a member. Because some persons changed families during the year, their family income is defined as the income of the family they were in the longest. If a family did not exist for the entire year, the family income is adjusted to an annual basis by dividing actual income by the proportion of the year the family existed. Unrelated persons are classified according to their own income. For each person, 12 categories of income were collected, including income from employment for persons 14 years of age and over and income from various government programs, pen- sions, alimony or child support, interest, and net rental in- come. Where information was missing it was imputed. For persons who were members of more than one family, their total income was allocated to each family in proportion to the amount of time they were in that family. Health care coverage—Twelve mutually exclusive categories of health care coverage were developed. Because of the importance and extent of Medicare coverage for persons 65 years of age and over, the population was first divided into those under 65 years of age and those 65 years of age and over. For persons under 65 years of age, coverage is divided into four mutually exclusive categories: Coverage all year from a single source, coverage all year from a mixture of sources, coverage only part of the year, and no health care coverage. For those under 65 years of age and covered all year from a single source, three subcategories of coverage were designated: Private insurance only, such as commercial carrier or Blue Cross; Medicaid only; and other public programs including Medicare, CHAMPUS/ CHAMPVA, Indian Health Service, and other programs cover- ing the cost of health care. Persons in the part-year-coverage category had health care coverage under either a private insurance policy or a public program, but the coverage did not extend throughout the year. People 65 years of age and over are partitioned into two major coverage categories: Those covered by Medicare and not covered by Medicare. The former group is subdivided into persons having only Medicare coverage, those who have supplemented their Medicare with private policies, and those who are covered not only by Medicare but also by Medicaid, the Indian Health Service, or other public program. The second subgroup, those not having Medicare, is divided into persons who have some other type of health care coverage, whether private or public, and those who have no coverage _atall. For the multivariate analyses, categories that do not distin- guish between age groups were formed, as follows: e Multiple Public: Coverage by more than one public pro- gram during the year; in most, but not all cases denotes simultaneous coverage by two programs, such as coverage by both Medicare amd Medicaid at the same time. e Single Public: Coverage by only one public program, either Medicare, Medicaid, Indian Health Service, CHAMPUS, or other government program. e Private and Public: Coverage during the year by at least one private insurance plan and at least one public program; in most cases both types of coverage overlay for at least part of the year. e Private Only: Coverage during the entire year by private insurance only. e None or other: Includes those with no health care coverage during the year; those with coverage for part of the year by only one source, and, for those 65 years of age and over, coverage other than Medicare. Hospital admission—The formal acceptance by a hospital of a patient who is provided room, board, and regular nursing care in a unit of the hospital. Included as a hospital admission is a patient admitted to the hospital and discharged on the same day. Also included is a hospital stay resulting from an emergency department visit. Hospital days—The total number of inpatient days ac- cumulated at time of discharge by patients discharged from short-stay hospitals during a year constitute hospital days. A stay of less than 1 day (patient admission and discharge on the same day) is counted as 0 days in the summation of hospital days. For patients admitted and discharged on different days, the number of days of care is computed by counting all days from (and including) the date of admission to (but not including) the date of discharge. Hospital outpatient department visit—A face-to-face en- counter between an ambulatory patient and a medical person. The patient comes to a hospital-based ambulatory care facility to receive services and departs on the same day. If more than one department or clinic is visited on a single trip, each depart- ment or clinic visted is counted as a separate visit. Household—Occupants of a housing unit or group quarters that was included in the sample. This could have been one per- son, a family of related people, a number of unrelated people, or a combination of related and unrelated people. Housing unit—A group of rooms or a single room oc- -cupied or intended for occupancy as separate living quarters: that is, (1) the occupants did not live and eat with any other per- sons in the structure, and (2) there was either direct access from the outside or through a common hall, or there were complete kitchen facilities for the use of the occupants only. Key person—A key person was (1) an occupant of a na- tional household sample housing unit or group quarters at the time of the first interview; (2) a person related to and living with a State Medicaid household case member at the time of the first interview; (3) an unmarried student 17-22 years of age living away from home and related to a person in one of the first two groups; (4) a related person who had lived with a person in the first two groups between January 1, 1980, and the round 1 inter- view, but was deceased or had been institutionalized; (5) a baby born to a key person during 1980; or (6) a person who was liv- ing outside the United States, was in the Armed Forces, or was in an institution at the time of the round 1 interview but who had joined a related key person. Limitation of activity—A functional limitation score was developed for classifying limitation of activity. It ranges from 0, indicating no limitation of activity, to 8, meaning severe activity limitation, and 9, indicating death during the survey period. The functional limitation score was developed from responses to a battery of questions designed to assess ability to perform various common functions such as walking, driving a car, and climbing stairs. For NMCUES, these questions were asked of persons 17 years of age and over. Mean charge per unit of service—The arithmetic mean calculated from charges reported by the household respondent without consideration for the amount actually paid or the source of payment. Zero charges were assigned to service units that the household reported as free from the provider in response to three separate questions. Nonkey person—A person related to a key person who joined him or her after the round 1 interview but was part of the civilian noninstitutionalized population of the United States at the date of the first interview. Patient—A person formally admitted to the inpatient ser- vice of a short-stay hospital for observation, care, diagnosis, or treatment. In this report the number of patients refers to the number of discharges during the year, including any multi- ple discharges of the same individual from one or more short- stay hospitals. The terms “patient” and “inpatient” are used synonymously. 87 Per capita charges—Calculated by dividing the total charges by the number of people in the reference population. Perceived health status—The family respondent’s judgment of the health of the person compared with others the same age, as reported at the time of the first interview. The categories were excellent, good, fair, or poor. Poverty status—The poverty status in 1980 was calculated by dividing the person’s family income in 1980 by the appro- priate 1980 nonfarm poverty level threshold and converting it to percent. These thresholds, as used by the U.S. Bureau of Census, are determined by the age and sex of the family head and the average number of persons in the family. Prescribed medicine acquisitions—The number of times a person had a prescription filled, regardless of whether it was an initial filling or a refill of a prescription. Race—The race of people 17 years of age and over reported by the family respondent; the race of those under 17 years of age derived from the race of other family members. If the head of the family was male and had a wife who was living in the household, her race was assigned to any children under 17 years of age. In all other cases, the race of the head of the family (male or female) was assigned to any children under 17 years of age. Race is classified as “white,” “black,” or “other.” The “other” race category includes Ameri- can Indian, Alaskan Native, Asian, Pacific Islander. The cate- gory “white and other” includes the categories “white” and “other.” Region—NORTHEAST: Maine, New Hampshire, Ver- mont, Massachusetts, Rhode Island, Connecticut, New York, New Jersey, Pennsylvania, NORTH CENTRAL: Michigan, 88 Wisconsin, Ohio, Indiana, Illinois, Minnesota, Iowa, Mis- souri, North Dakota, South Dakota, Nebraska, Kansas; SOUTH: Delaware, Maryland, District of Columbia, Virginia, West Virginia, North Carolina, South Carolina, Georgia, Florida, Kentucky, Tennessee, Alabama, Mississippi, Arkan- sas, Louisiana, Oklahoma, Texas; WEST: Montana, Idaho, Wyoming, Colorado, New Mexico, Arizona, Utah, Nevada, Washington, Oregon, California, Alaska, Hawaii. Reporting unit—The basic unit for reporting data in the household components of NMCUES. A reporting unit con- sisted of all related people residing in the same housing unit or group quarters. One person could give information for all mem- bers of the reporting unit. Restricted-activity day—A restricted-activity day is one on which a person cuts down on his usual activities for the whole of that day because of an illness or an injury. The term “usual activities” for any day means the things that the person would ordinarily do on that day. A day spent in bed or a day home from work because of illness or injury is, of course, arestricted-activity day. Round—A round was the administrative term used to de- signate all interviews that occurred within a given period of time and that used the same instruments and procedures. Work-loss day—A work-loss day is a day on which a person did not work at his or her job or business because of a specific illness or injury. The number of days lost from work is determined only for persons 17 years of age and over who reported that at any time during the survey period they either worked at or had a job or business. Department of Health and Human Services Margaret M. Heckler, Secretary Health Care Financing Administration C. McClain Haddow, Acting Administrator Office of Research and Demonstrations David J. Butler, Acting Director Office of Research Allen Dobson, Ph.D., Director Division of Program Studies Carl Josephson, Director Surveys Studies Branch Herbert A. Silverman, Ph.D., Chief Public Health Service James O. Mason, Acting Assistant Secretary for Health National Center for Health Statistics Manning Feinleib, M.D., Dr.P.H., Director Office of Interview and Examination Statistics Program E. Earl Bryant, Associate Director Division of Health Interview Statistics Robert R. Fuchsberg, Director Utilization and Expenditure Statistics Branch Robert A. Wright, Chief fo U.S. DEPARTMENT OF HEALTH AND HUMAN SERVICES Public Health Service National Center for Health Statistics 3700 East-West Highway Hyattsville, Maryland 20782 THIRD CLASS MAIL BULK RATE POSTAGE & FEES PAID PHS/NCHS PERMIT No. G-281 OFFICIAL BUSINESS PENALTY FOR PRIVATE USE, $300 DHHS Publication No. 86-20402 GENERAL LIBRARY - UC. BERKELEY BODOL4925)