key: cord-0887710-9a57ecob
authors: Cotton, Dale M.; Liu, Liyan; Vinson, David R.; Ballard, Dustin W.; Sax, Dana R.; Hofmann, Erik R.; Lin, James S.; Durant, Edward J.; Kene, Mamata V.; Casey, Scott D.; Ghiya, Meena; Shan, Judy; Bouvet, Sean C.; McLachlan, Ian D.; Rauchwerger, Adina S.; Mark, Dustin G.; Reed, Mary E.
title: Clinical characteristics of COVID‐19 patients evaluated in the emergency department: A retrospective cohort study of 801 cases
date: 2021-08-23
journal: J Am Coll Emerg Physicians Open
DOI: 10.1002/emp2.12538
sha: beb8cc00880f29f959f57a22899365a00bf5fc69
doc_id: 887710
cord_uid: 9a57ecob

BACKGROUND: Coronavirus disease 2019 (COVID‐19), the disease caused by severe acute respiratory syndrome coronavirus 2 (SARS‐CoV‐2), has extracted devastating tolls. Despite its pervasiveness, robust information on disease characteristics in the emergency department (ED) and how that information predicts clinical course remain limited. METHODS: We conducted a retrospective cohort study of the first ED visit from SARS‐CoV‐2‐positive patients in our health system, from February 21, 2020 to April 5, 2020. We reviewed each patient's ED visit(s) and included the first visit with symptoms consistent with COVID‐19. We collected demographic, clinical, and treatment variables from electronic health records and structured manual chart review. We used multivariable logistic regression to examine the association between patient characteristics and 2 primary outcomes: a critical outcome and hospitalization from index visit. Our critical outcome was defined as death or advanced respiratory support (high flow nasal cannula or greater) within 21 days. RESULTS: Of the first 1030 encounters, 801 met our inclusion criteria: 15% were over age 75 years, 47% were female, and 24% were non‐Hispanic white. We found 161 (20%) had a critical outcome and 393 (49%) were hospitalized. Independent predictors of a critical outcome included a history of hypertension, abnormal chest x‐ray, elevated neutrophil to lymphocyte ratio, elevated blood urea nitrogen (BUN), measured fever, and abnormal respiratory vital signs (respiratory rate, oxygen saturation). Independent predictors of hospitalization included abnormal pulmonary auscultation, elevated BUN, measured fever, and abnormal respiratory vital signs. CONCLUSIONS: In this large, diverse study of ED patients with COVID‐19, we have identified numerous clinical characteristics that have independent associations with critical illness and hospitalization.

Coronavirus disease 2019 (COVID- 19) , the disease caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), has extracted incalculable tolls in healthcare settings and society at large.

By the end of 2020, over a third of a million Americans had COVID-19

confirmed deaths and more than 100,000 excess deaths had been associated with the pandemic, not including the anticipated mortality impacts of lost educational and economic attainment. 1-3 Despite significant scientific inquiry in response to this disease burden, no therapies have emerged as curative and the modest benefits seen in the best studied therapies are marginal. 4, 5 For these reasons, supportive care remains the mainstay of therapy, and decision-making around use of these supportive care resources (ie, admission to hospital) is the keystone in the provision of COVID-19 care.

Although significant COVID-19 disease is a condition primarily seen in emergency department (ED) and hospital settings, little has been published on characteristics of the disease in the ED. Most studies involving ED patients include them indirectly, by observing patients who have already been hospitalized. [6] [7] [8] [9] [10] [11] [12] When ED patients have been directly studied, the study population also included non-ED patients or included in the analysis a majority of patients not confirmed to have COVID-19. [13] [14] [15] Many models have examined population-level associations, or included variables driven primarily by data-collection convenience, emphasizing comorbidities or numerous laboratory values over the patient-level clinical features, examination findings, and diagnostic data typically available in the ED. 16, 17 These approaches are understandable given the urgent need to address COVID- 19, but it is anticipated that use of more granular ED data will offer stronger associations between patient characteristics and clinically important outcomes.

Despite COVID-19's pervasiveness, robust information on ED clinical evaluation and how that information predicts clinical course remain limited. Therefore, we undertook a retrospective cohort study of a diverse group of ED patients diagnosed with COVID-19 in a multicenter, non-surge, community setting. Additionally, we used multivariable logistic regression to assess predictors of 2 clinically meaningful outcomes from index ED visit: hospitalization and a critical outcome of death or high flow nasal cannula (HFNC) or greater respiratory support within 21 days.

We conducted a retrospective cohort study of the first 1030 ED patient-visits with an in-system laboratory test positive for SARS-CoV-2 and an ED visit within 21 days before or after the test.

The study took place in the 21 community EDs of Kaiser Permanente Northern California (KPNC), an integrated health care system that provides comprehensive medical care for greater than 4 million members with approximately 1.2 million annual ED visits. KPNC members represent approximately one-third of the population in areas served and are highly representative of the ethnic and socioeconomic diversity of the surrounding population. 18 Figure S1 ).

Twelve practicing emergency medicine clinicians served as abstractors.

All received standardized training on data collection methods and use of the data collection tool, which was modified to its final form after an iterative process. 19 The principal investigator answered and arbitrated all coding questions and monitored data collection by regularly assessing abstractor performance and providing targeted feedback.

We asked abstractors to report if the possibility of COVID-19 was recognized by the treating ED clinician during the ED visit, either explicitly (ie, documented as such in the note or the discharge instructions) or implicitly (ie, ordered a SARS-CoV-2 test or ordered isolation precautions specific to COVID-19 practices). All encounters in which the initial abstractor reported that COVID-19 was "unrecognized" by the treating clinician were abstracted by a second abstractor and adjudicated by a third if the first 2 interpretations were incongruent. We also asked abstractors to determine if the ED visit was for symptoms poten-

In 801 patients who visited the emergency department with   COVID-19, 20% had a critical outcome, defined as death or   receipt of high flow nasal cannula, noninvasive ventilation, or invasive mechanical ventilation. Factors associated with a critical outcome included a history of hypertension, abnormal chest x-ray, elevated neutrophil to lymphocyte ratio, elevated BUN, fever, abnormal respiratory rate, or low oxygen saturation.

tially consistent with COVID-19 (see Table 3 for a list of symptoms we considered to be consistent with COVID-19). All encounters deemed incompatible with COVID-19 disease symptoms (eg, ankle pain after ankle trauma) were reviewed by a second abstractor and agreed on by consensus after discussion. To assess inter-rater agreement, 100 cases were randomly selected and independently reviewed by 2 abstractors;

we calculated a kappa on abstractor identification of the presence of shortness of breath as a symptom. We used accepted kappa ranges to assess the quality of our agreement. 20

We abstracted the following patient-level variables during manual 

A comprehensive list of data was collected from automated data sources including patient age, gender, body mass index (BMI), race, smoking history, KP health plan membership status, select comorbidities, and select chronic medication use. Chronic medication use was assessed by observing system-filled prescriptions in a 100-day window from 121 days prior to index visit to 21 days prior to index visit, and included bronchodilators, steroids, immunosuppressants, angiotensin converting enzyme (ACE) inhibitors, angiotensin receptor blockers (ARB), and other anti-hypertensives. Medications that may have been prescribed to treat a COVID-19 like illness were assessed by observing system-filled prescriptions in a 21-day window immediately prior to the index ED visit. These included medications at the time purported to treat COVID-19 (hydroxychloroquine, lopinavir), bronchodilators (beta agonists, inhaled steroids), systemic corticosteroids, antivirals, and antibiotics. Additionally, we identified the administration of these same medications in the ED. We electronically identified diagnostic results including index SARS-CoV-2 test order timing and ED laboratory values as well as ED vital signs. We also describe the highest level of respiratory support (HLRS) achieved during the index hospitalization, the time to achieve the HLRS, hospitalization length of stay, 30day all-cause mortality rates, and hospitalization within 21 days for those not initially hospitalized.

We assessed 2 primary outcomes: a critical outcome within 21 days of index ED visit and hospital admission from index ED visits. A critical outcome was defined as death from any cause or use of advanced respiratory support (HFNC, non-invasive ventilation including continuous or biphasic positive airway pressure, or invasive ventilation).

Our critical outcome measure is congruent with the World Health 

We reported baseline characteristics among patients that did and did not meet 1 of our 2 primary outcomes, allowing for calculation of an unadjusted odds ratio (OR). We conducted a multivariable logistic regression with facility level random effect to assess patient characteristics associated with each outcome and report ORs and 95% confidence intervals (CIs). 24 We initially included variables considered to be clinically meaningful that also had a statistically significant association (P < 0.05) with our outcomes on unadjusted analysis. We then narrowed our selections prioritizing variables associated with our critical outcome as well as laboratory tests that were more prevalent as an approach likely to maximize the clinical interpretability of our model. We then pruned our variable selection based on a variable to outcome ratio of 10:1 to pre-vent over fitting. For simplicity and to aid in comparison, we used the same variables in both outcomes. We examined model fit and performance by examining the area under the receiver operator curve (ROC) curves and reporting the P-values of the Hosmer-Lemeshow goodness of fit test.

We performed an analysis in our multivariable model using heart rate in lieu of temperature, given a known association between the variables. We also performed another analysis using our critical outcome with a 7-day instead of the 21-day post ED visit window. All analyses were performed using SAS (Cary, NC) 9.4 and R Studio (Boston, MA) 3.14.

Of the 1030 encounters examined, 801 met our inclusion criteria: the median age was 55 years (interquartile range [IQR] = 42-67): 47%

were female, 24% were non-Hispanic white, and 70% were obese or overweight (Table 1) . 

We found 22% of patients arrived at the ED by ambulance, including a large minority (46%) of those who met our critical outcome ( Table 2) .

The most common chief complaints were shortness of breath (41%), cough (37%), and fever (32%). Less common were gastrointestinal chief complaints (10%) or systemic chief complaints like weakness (5%). We found 15% were known to have COVID-19 at the time of triage or initial clinical assessment. In triage, 59% patients were identified as having a fever by history, 10% as having an upper respiratory symptom, and 35% as having a symptom of systemic illness. Being transported via ambulance and registering with chief complaints of weakness or shortness of breath was all more likely in patients with a critical outcome.

The most commonly endorsed type of complaint was a lower respiratory symptom (90% of patients), the most common being cough (82%) (Table S1 ).

In our cohort, vital sign abnormalities were common ( 

Laboratory testing was common in patients: 78% of patients had bloodwork analyzed, including the majority (57%) of patients discharged home from index visit (Tables 5 and S2 ). 

Of patients with chronic use of select medications, the most commonly used were antihypertensives: 8% used an ACE inhibitor, 3% used an ARB, and 30% used another anti-hypertensive (Table S3 ). In the 21 days before the index visit, we found that use of medications to potentially treat COVID-19 symptoms was uncommon. In the ED, patients were commonly given antibiotics (at least 50% of patients) and bronchodilators (35%), and to a lesser degree, corticosteroids (10%) and Abbreviations: ALT, alanine aminotransferase; ANC, absolute neutrophil count; AST, aspartate aminotransferase; BNP, b natriuretic peptide; BUN, blood urea nitrogen; CRP, C-reactive protein; NLR, neutrophil to lymphocyte ratio; WBC, white blood cell. hydroxychloroquine (19%). For patients discharged home these same medications were uncommonly prescribed.

Of 408 patients discharged home from the ED after index visit, 116 (28%) returned to the ED within 21 days (Table S4) 

Our final multivariable models included 16 variables (Table 7) . Independent predictors of a critical outcome included a history of hyperten- 

We performed an analysis looking at our critical outcome at 7 days instead of 21 days; we found minimal change in our modelling results. Additionally, our assessment of interrater agreement between 2 reviewers on 109 charts for the identification of shortness of breath as a symptom demonstrated substantial agreement, κ = 0.74 (95% CI 0.61-0.87) and 87% agreement.

One limitation of our study is that it disproportionately includes critical COVID-19 illness, because these patients are more likely to be identified due to a need for acute medical care. Additionally, early testing recommendations favored patients being or already hospitalized, making it more likely to capture critical cases. Despite this limitation, a meaningful fraction of our cohort was not admitted and did not meet our critical outcome (ie, had mild or moderate disease), which offers important characteristic data to report associations.

Examining an outcome of hospitalization has many limitations.

Although we feel this is an appropriate and vital perspective to capture, taken on its own it is an unreliable measure due to variability from numerous influences. Despite this, hospitalization captures an otherwise unobservable summation of factors that includes clinical gestalt.

This complex but meaningful contribution has unique value, especially when taken in the context of other data.

Another limitation of our study is the possibility that some of our study patients followed up in other facilities. Despite this possibility, the vast majority of patients were KP health plan members (85%), and we were able to capture deaths occurring outside of the KPNC system using state mortality databases, mitigating this limitation.

The retrospective nature of our study is another limitation that may impact our ability to reliably capture information. Despite the limitations of a retrospective approach, we did employ manual chart review techniques, which improved our granular data extraction compared to studies that relied entirely on automated data collection. Additionally, our study was conducted with typical methodology standards used for clinical research, without suspended standards due to the pandemic. 28 Given the pitfalls of using exceptions for research standards in COVID-19, we feel our study's contribution is meaningful despite a retrospective limitation. 28 [8%] for COVAS), limiting the utility in understanding prognosticators among patients with confirmed COVID-19 disease. The tools also did not incorporate physical examination findings or laboratory diagnostics, the latter which has been well-described in hospitalized patients to be associated with critical illness and the former that is not well studied.

Most COVID-19 studies that examine diagnostic data as predictors of critical illness include already hospitalized patients. 9,30-34 Similar to our analysis, several of these studies have observed independent associations with abnormal CXR findings. 33, 35 Studies that examined diagnostic laboratory results have also found independent associations, including elevated NLR, BUN, glucose, C-reactive protein, D-Dimer, troponin, and others. These findings are supported by the results of our multivariable model (NLR, BUN, and glucose). Physical examination findings (specifically pulmonary auscultation) were not examined as predictors in these studies. 36 This is presumably because they were not reliably available using only electronic data capture.

Another interesting consideration is the variation in body temperatures association with outcomes between studies. This could be due to different temperature obtainment practices, such as non-oral temperatures. In our non-surge setting, 96% of temperatures acquired were from an oral or core route, which would be expected to aid in accurate temperature representation and allow for observation of predictive value if present. In our multivariate model, an elevated temperature was associated with increased odds of a critical outcome. In a study by Garibaldi et al 9 that had many of the same conditions as our study (eg, multi-center, non-surge), they also noted meaningful independent association with measured fever, although like most studies they do not describe their temperature acquisition method.

Our study shifts the focus from hospitalized to ED patients, pro- A separate interesting and diagnostic testing observation is the 100% negative influenza testing rate (from 165 tests) for our cohort. The time lag between our dataset and the initial reported data out of Wuhan, China is only weeks to months, suggesting that confounding concomitant illness in a novel pandemic may rapidly approach negligible rates as risk mitigation behaviors are deployed in the public sphere. 37, 38 Although previously described for other medical conditions but not well addressed in clinical decision-making surrounding COVID-19, there is a distinction between the population risk factors that describe who is at risk for severe disease prior to clinical assessment from those at risk of severe disease after clinical data are available. Although age and comorbidities have strong associations with severe COVID-19 disease and hospitalization, these population-level factors appear to not carry significant independent value when compared to granular clinical data such as vital signs, laboratory results, examination findings, and imaging results. For example, our model suggests that a clinician should be more concerned about a COVID-19-positive 35-year-old female with hypertension, measured fever, and an elevated BUN (with otherwise normal evaluation) than a COVID-19-positive 65-year-old male former smoker with hyperlipidemia (and otherwise normal evaluation). After an ED assessment, the population level associates of critical illness such as age and certain comorbidities no longer carry the predictive value that they did prior to clinical evaluation. For patients without overt indication for admission to hospital, our data suggests characteristics of their ED evaluation that may influence a decision to hospitalize, help determine a level of outpatient monitoring, or inform clinician-patient conversations about risk of progression.

In summary, we identified patient-level clinical variables from ED encounters that carry independent associations with severe COVID-19 and hospitalization. Our study adds significantly to the understanding of the clinical trajectory of COVID-19 through inclusion of a diverse, community-based, multi-center cohort of patients. and MER oversaw the study as a whole. DMC drafted the manuscript and all authors contributed substantively to its critical revision and its final approval. DMC takes responsibility for the paper as a whole.

The authors declare no conflicts of interest.

Excess deaths associated with COVID-19, by age and race and ethnicity-United States

Estimation of US children's educational attainment and years of life lost associated with primary school closures during the coronavirus disease 2019 pandemic

An interactive web-based dashboard to track COVID-19 in real time

Update to living WHO guideline on drugs for COVID-19

Understanding clinical decision-making during the Covid-19 pandemic: a cross-sectional worldwide survey

Presenting characteristics, comorbidities, and outcomes among 5700 patients hospitalized with Covid-19 in the new york city area

Functional characterization of the cannabinoid receptors 1 and 2 in zebrafish larvae using behavioral analysis

Development and external validation of a Covid-19 mortality risk prediction algorithm: a multicentre retrospective cohort study

Patient trajectories among persons hospitalized for Covid-19: a cohort study

Characteristics, outcomes, and trends of patients with Covid-19-related critical illness at a learning health system in the united states

Epidemiological and clinical characteristics of 99 cases of 2019 novel coronavirus pneumonia in Wuhan, China: a descriptive study

Clinical characteristics of coronavirus disease 2019 in china

CoVA: an acuity score for outpatient screening that predicts coronavirus disease 2019 prognosis

Characteristics, risk factors and outcomes among the first consecutive 1096 patients diagnosed with Covid-19 in Kuwait

Identifying patients with symptoms suspicious for Covid-19 at elevated risk of adverse events: the COVAS score

Prediction models for diagnosis and prognosis of Covid-19 infection: systematic review and critical appraisal

Characterization and clinical course of 1000 patients with coronavirus disease 2019 in New York: retrospective case series

The Kaiser Permanente northern California Adult Member Health Survey

Looking through the retrospectoscope: reducing bias in emergency medicine chart review studies

The measurement of observer agreement for categorical data

Frequency and distribution of chest radiographic findings in patients positive for Covid-19

The role of chest imaging in patient management during the Covid-19 pandemic: a multinational consensus statement from the fleischner society

A minimal common outcome measure set for COVID-19 clinical research

Reporting research results: recommendations for improving communication

Risk Factors associated with mortality among patients with Covid-19 in intensive care units in Lombardy, Italy

Prevalence of comorbidities among individuals with Covid-19: a rapid review of current literature

Prevalence of comorbidities and its effects in patients infected with SARS-CoV-2: a systematic review and meta-analysis

Against pandemic research exceptionalism

Expression of concern: Hydroxychloroquine or chloroquine with or without a macrolide for treatment of Covid-19: a multinational registry analysis

Factors associated with hospital admission and critical illness among 5279 people with coronavirus disease 2019 in New York City: prospective cohort study

Neutrophil-to-lymphocyte ratio predicts critical illness patients with 2019 coronavirus disease in the early stage

Machine learning to predict mortality and critical events in a cohort of patients with Covid-19 in new york city: model development and validation

Development and validation of a clinical risk score to predict the occurrence of critical illness in hospitalized patients with Covid-19

Clinical course and risk factors for mortality of adult inpatients with COVID-19 in Wuhan, China: a retrospective cohort study

Model-based prediction of critical illness in hospitalized patients with covid-19

Auscultation of the respiratory system

Clinical features, diagnostics, and outcomes of patients presenting with acute respiratory illness: a comparison of patients with and without COVID-19. medRxiv

Influenza and COVID-19 coinfection: report of six cases and review of the literature