key: cord-0693167-jdc9fqsi
authors: Rivas, A. L.; Hoogesteijn, A. L.; Hittner, J.; van Regenmortel, M. H.; Kempaiah, P.; Vogazianos, P.; Antoniades, A.; Febles, J. P.; Fasina, F. O.
title: Toward a COVID-19 testing policy: where and how to test when the purpose is to isolate silent spreaders
date: 2020-12-24
journal: nan
DOI: 10.1101/2020.12.22.20223651
sha: d3ad47e287ce9bb827372a06ba4a81c8a72497de
doc_id: 693167
cord_uid: jdc9fqsi

Background To stop pandemics, such as COVID 19, infected individuals should be detected, treated if needed, and to prevent contacts with susceptible individuals- isolated. Because most infected individuals may be asymptomatic, when testing misses such cases, epidemics may growth exponentially, inducing a high number of deaths. In contrast, a relatively low number of COVID 19 related deaths may occur when both symptomatic and asymptomatic cases are tested. Methods To evaluate these hypotheses, a method composed of three elements was evaluated, which included: (i) county and country level geo-referenced data, (ii) cost benefit related considerations, and (iii) temporal data on mortality or test positivity (TP). TP is the percentage of infections found among tested individuals. Temporal TP data were compared to the tests/case ratio (T/C ratio) as well as the number of tests performed/million inhabitants (tests/mi) and COVID-19 related deaths/million inhabitants (deaths/mi). Findings Two temporal TP profiles were distinguished, which, early, displayed low (~ 1 %) and/or decreasing TP percentages or the opposite pattern, respectively. Countries that exhibited >10 TP % expressed at least ten times more COVID-19 related deaths/mi than low TP countries. An intermediate pattern was identified when the T/C ratio was explored. Georeferenced, TP based analysis discovered municipalities where selective testing would be more cost effective than alternatives. Interpretations When TP is low and/or the T/C ratio is high, testing detects asymptomatic cases and the number of COVID 19 related deaths/mi is low. Georeferenced TP data can support cost-effective, site specific policies. TP promotes the prompt cessation of epidemics and fosters science based testing policies.

Evidence before this study 24

To map this field, bibliographic searches were conducted in the Web of Science, which included 25 the following results: (i) COVID-19 (95,133 hits), (ii) SARS COV-2 (33,680 hits), (iii) testing 26 policy and COVID-19 (939 hits), (iv) testing policy and SARS COV-2 (340 hits), (v) testing 27 policy and COVID-19 and asymptomatic (80 hits), (vi) testing policy and SARS COV-2 and 28 asymptomatic (54 hits); (vii) test positivity and COVID-19 and validation (7 hits), and (viii) test 29 positivity and SARS CoV-2 and validation (5 hits). Therefore, before this study, testing policy in 30 relation to asymptomatic cases as well as test positivity represented a very low proportion 31 (between ~1 thousandth to ~ 1 ten thousandth) of all publications. While many articles 32 distinguished between diagnostic and screening tests, no paper was found in which testing policy 33 is mentioned as part of a process ultimately designed to isolate all infected individuals. The few 34 articles that mentioned test positivity only investigated symptomatic cases. These 35 quanti/qualitative assessments led the authors to infer that neither testing policy nor test 36 positivity had been adequately validated and/or investigated. 37 38 Added value of this study 39

We provide the first validation of test positivity as an estimate of disease prevalence under 40 rapidly changing conditions: in pandemics, disease prevalence may vary markedly within short 41 periods of time. We also address a double limitation of control campaigns against COVID-19, 42 namely: it is unknown who and where to test. Asymptomatic cases are not likely to seek medical 43 assistance: while they feel well, they silently spread this pandemic. Because they represent 44 approximately half of all infected individuals, they are a large, moving, and invisible target. 45 distributed but geographically clustered, and, up to now less than four persons per thousand 48 inhabitants are tested on a given day. However, by combining geo-referenced test positivity data 49 with cost-benefit considerations, we generate approaches not only likely to induce high benefits 50 without increasing costs but also free of assumptions: we measure bio-geography as it is. 51 52 Implications of all the available evidence 53

The fact that asymptomatic cases were not tested in many countries may explain the exponential 54 growth and much higher number of deaths observed in those countries. Ineffective testing (and, 55 therefore, ineffective isolation) can also result from the absence of geo-referenced data analysis. 56

Because the geographical location where people reside, work, study, or shop is not a random 57 event, the analysis of small greographical areas is essential. Only when actual geographical 58 relationships are observed, optimal (cost-benefit oriented) testing policies can be devised. 59 60 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

The copyright holder for this this version posted December 24, 2020. ; https://doi.org/10.1101/2020.12.22.20223651 doi: medRxiv preprint

Albert Bartlett partially dedicated his life to teaching the expressions and consequences 85 of the exponential function [1] . Epidemics -including the COVID-19 pandemic− tend to grow or 86 decline exponentially. Given the catastrophic consequences of the epidemic exponential growth 87 [2 Rudan ], the first priority in epidemic control is to avoid contacts: when it takes just 60 days to 88 increase from 1 to > 60,000 COVID-19 related deaths (as observed in this pandemic), 89 'coexistence with the virus' or 'flattening the epidemic curve' is not a likely outcome. 90

Because blocking contacts between infected and susceptible individuals prevents the 91 exponential growth of an epidemic, testing is the first priority of a Therefore, a fourth action should be added to the three proposed by WHO: identification. 103

The determination of the geographical sites where asymptomatic cases are likely to be located is 104 the linchpin of epidemic control. However, only a minor proportion of the population is tested on 105 a given day (usually less than four per thousand inhabitants [8]). Given the scarce resources 106 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

The copyright holder for this this version posted December 24, 2020. ; https://doi.org/10.1101/2020.12.22.20223651 doi: medRxiv preprint available, the limited testing, and the need to remove viral spreaders before the epidemic 107 exponential growth overwhelms the anti-epidemic responses, identification should be precise, 108 rapidly implemented, and cost-effective [9] [10] [11] In this report, TP is viewed as a metric that may evaluate the efficacy of testing programs 118 aimed at detecting asymptomatic COVID-19 cases; i.e., test positivity percentages may indirectly 119 help to elucidate whether testing is or is not capturing asymptomatic cases, even when the 120 number of tests performed per million inhabitants is low. In this context, TP is a component of 121 an epidemiologic strategy used in a rapidly changing environment with the purpose of detecting 122 and isolating infected individuals before the epidemic exponential process consolidates. 123

To the best of our knowledge, the validity of the TP metric to detect asymptomatic cases 124 when testing is limited has not yet been demonstrated. To understand why TP percentages may 125 achieve that goal, the history of epidemics is worth revisiting. When disease prevalence is zero 126 (before an epidemic starts), the TP % is zero, too. The TP % is expected to be non-zero (but 127 relatively low) after epidemic onset. That is so because even the worst epidemics on record have 128 infected only a minor percentage of the population -for instance, the highest estimates on 129 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

The copyright holder for this this version posted December 24, 2020. institutions of some countries [16] [17] [18] ). This hypothesis is further supported when the alternative 140 is associated with promising outcomes, that is, when lower TP values are found in countries 141

where the testing policy explicitly captures asymptomatic cases and, later, much fewer deaths per 142 million inhabitants are reported than in countries reporting high TP percentages. This means that 143 low TP percentages may reflect the prompt removal of SARS CoV-2-positive individuals -a 144 likely outcome of policies that test asymptomatic cases. Thus, empirically elucidating whether 145 high TP percentages are found when asymptomatic cases are excluded in the testing policy as 146 well as the alternative hypothesis −low TP percentages are associated with testing programs 147 that detect asymptomatic cases− are pre-requisites of scientifically sound testing policies meant 148 to identify and isolate the spreaders of epidemics. 149

Other metrics of potential interest include: (i) the number of new cases, (ii) the number 150 of COVID-19 related deaths/million inhabitants (deaths/mi), and (iii) the ratio between the 151 number of tests and the number of cases detected (the T/C ratio). The time when the highest 152 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

The copyright holder for this this version posted December 24, 2020. To demonstrate that novel testing methods can capture asymptomatic cases even when 156 testing is limited, high-resolution, geo-referenced data may be required. When small 157 geographical areas report high TP percentages, intensive testing in such areas could induce 158 benefits that will include the larger surrounding area, generating high benefit-cost ratios [20] . 159 Accordingly, relationships between the TP percentage, its inverse (the number of tests 160 conducted per each detected case ratio or T/C ratio), and other metrics (the number of new cases 161 and the number of deaths per million inhabitants) were investigated across countries and −using 162 geo-referenced data of high resolution − across municipalities. This study aimed at elucidating 163 whether, when and where the TP percentage and/or the T/C ratio informed earlier and/or 164 provided more information than classic metrics in reference to asymptomatic cases and, if so 165 demonstrated, whether such metrics, once geo-referenced, could foster region-specific, cost-166 benefit oriented control policies. 167 168

Methods 170

Data on new (daily) tests performed, cases identified, test positivity (the percentage of positive 171 cases/tests performed) and its inverse (the test/case ratio), as well as the cumulative number of 172 tests or deaths per million inhabitants were collected from publicily available data sources and 173 compared across countries and across time. Countries were classified as either displaying high or 174 low test positivity (TP) when, at least three months after the initiation of the epidemic, they 175 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

The copyright holder for this this version posted December 24, 2020. ; https://doi.org/10.1101/2020.12.22.20223651 doi: medRxiv preprint showed non-overlapping intervals of TP and such classes also revealed non-overlapping intervals 176 of epidemic-related outcomes (deaths/million inhabitants). Differences in median deaths/million 177 inhabitants between low TP and high TP countries were investigated with the Mann-Whitney test 178 using a commercial statistical package (Minitab LLC, State College, PA, USA). Geographical 179 data were processed with ARC GIS software (ESRI, Redlands, Ca, USA). Thirteen countries 180 were evaluated because, together, they exhibited geographical and demographic diversity and 181 individually, they were affected by COVID-19 long enough (two or more months) so they could 182 CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

The copyright holder for this this version posted December 24, 2020. ; https://doi.org/10.1101/2020.12.22.20223651 doi: medRxiv preprint

Test positivity-related temporal patterns 200

Findings revealed two temporal patterns: (i) one characterized by an early TP peak (the 201 highest TP percentages took place in the first weeks), followed by a rapid decline to ~ 1% (four 202 countries, green rectangles, Figs. 1 A-D); and (ii) a pattern shared by six countries, which 203 showed later and increasing TP values, even >10%, over one or more weeks (red rectangles, 204

Figs. 1 E-J). In addition, the data of ten countries consistently demonstrated that the highest TP 205 percentages occurred earlier or at the same time as, but not later than the number of new cases 206 ( Fig. 1 A-J). The first pattern (low TP percentage) was associated with a quasi-complete 207 cessation of the epidemic within four months. The second pattern (high TP percentages) was 208

shown by countries where, four to six months into the epidemic, the median number of deaths/mi 209 was at least 10 times higher than in countries displaying the alternative pattern (p<0.02, Fig. 2) . is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

The copyright holder for this this version posted December 24, 2020. ; https://doi.org/10.1101/2020.12.22.20223651 doi: medRxiv preprint

To investigate the hypothesis that geo-referenced and temporal TP data support site-222 specific, cost-effective interventions, the test positivity percentages reported in all municipalities 223 of Puerto Rico on September 11, 2020, were investigated. One municipality reported a TP 224 >50%, which was surrounded by 17 municipalities that reported much lower TPs. Assuming that 225 testing resources (including tests) were distributed according to the size of the population, the 226 municipality with >50% TP received 1/42 of such resources and the remaining municipalities 227 obtained testing-related resources 42 times larger in magnitude (Fig. 5) . − a region barely affected by the epidemic and, therefore, highly vulnerable (Fig. 5) . 236

These data also estimated the costs of random testing and/or testing that ignores geo-237 temporal cost-benefit considerations. If the surrounding area (area 2) was affected as much as 238 area 1, the size of epidemic could increase 42 times (Fig. 5) . This epidemic growth could occur 239 even if testing continued, uninterruptedly, at the same level as previously performed. 240 241 Additional tools 242

Relationships between metrics elucidated when testing was or was not adequate. 243

Examples collected from 14 countries revealed when testing grew faster than cases (a desirable 244 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

The copyright holder for this this version posted December 24, 2020. ; https://doi.org/10.1101/2020.12.22.20223651 doi: medRxiv preprint outcome) or the alternative pattern suggested additional testing was needed (Fig. 6A-D) and also considers geo-referenced, cost-benefit oriented data analysis. 254

Testing symptomatic or asymptomatic cases may depend on medical and public health-255 related perspectives [7] . For instance, testing that emphasizes symptomatic cases differs from 256 screening-oriented testing. In the first case, it is assumed that epidemics can be stopped through 257

interventions conducted within hospitals -as when individuals that feel ill seek medical 258 assistance. It is expected to find high percentages of test positivity when confirmatory testing is 259 promoted (the type of testing physicians are likely to prescribe). While this type of testing (or 260 purpose) is necessary when a clinical perspective is considered, it does not contribute much to 261 prevent epidemic spread, which depends on early detection of asymptomatic cases [7,18]. In 262 contrast, in population medicine, epidemiologic testing is needed (screening or information 263 usable in epidemic control, not only in clinical medicine). Here it was hypothesized that lower 264 TP percentages predicted lower numbers of deaths per million inhabitants than when higher TP 265 percentages were observed -findings consistent with the hypothesis that asymptomatic cases are 266 detected when the percentage of TP is low. Findings supported these hypotheses, opening a path 267 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

The copyright holder for this this version posted December 24, 2020. ; https://doi.org/10.1101/2020.12.22.20223651 doi: medRxiv preprint toward scientifically grounded testing policies that seek to detect (and immediately isolate) 268 asymptomatic -not only symptomatic− cases [7, 18]. As expected, countries that displayed high 269 TP percentages in early epidemic stages, later reported many more deaths/mi than low TP 270 countries. These findings supported the notion that, in high TP countries, asymptomatic cases 271 were missed and, consequently, the epidemic grew exponentially and so did deaths. 272

The epidemics here investigated also showed that TP peaks occur earlier than the peak 273 number of new cases. Data turning points (which include peaks) help to conduct epidemiologic 274 forecasts [21] . 275

Findings also addressed the statement made by WHO on May 12, 2020, which suggested 276 that low (<5%) TP values over two consecutive weeks signal that an epidemic is under control 277

[22]. This study found that even 1% TP was associated with longer epidemics (Figs. 3 A-D) . 278

The fact that neither a 5% nor a 1 % TP cutoff predicted the immediate cessation of an 279 epidemic was expected: dichotomization of continuous data promotes errors. Such a procedure 280 assumes that the discrete (discontinuous) nature of a process can change below or above a cutoff 281 generated by continuous data. Yet, the evidence demonstrated that an 'epidemic-positive' 282 country does not become an 'epidemic-negative' one when TP changes from 5.00 to 4.99 %. 283

While pattern recognition can discriminates without numerical cutoffs (as Fig. 5 shows), the 284 dichotomization assumption tends to lose information and generates errors [23] . 285 Therefore, TP seems a robust and explanatory metric. Unlike the number of new cases, 286 TP may inform when (and, provided that geo-referenced information is also available, where) 287 spikes of its percentage will identify specific places that require additional testing so 288 asymptomatic cases can be rapidly detected, isolated, and treated [24] . However, to prevent such 289 patients from promoting exponential epidemic growth, early and geographically-specific 290 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this this version posted December 24, 2020. implemented in such situations, the epidemiologic condition of the surrounding region can 302 rapidly worsen. This epidemiologic 'flood' metaphor was empirically supported: rapid 303 interventions in small areas that threaten larger areas tend to be highly cost-effective and may 304 stop epidemics (Fig. 4) . 305

While aggregate (state-or country-level) data do not reveal spatial relationships and, 306 therefore, are not very useful [27], non-aggregate geo-referenced information -e.g., data on 307 neighborhoods− are epidemiologically usable, as shown with municipality-level information 308 collected in Puerto Rico (Fig. 5) . When geographical-epidemiological relationships are 309 considered, it is possible to identify policies that benefit not only the area of direct intervention 310 but also the surrounding area -as commonly practiced in flood protection-related policies [28] . 311 This approach has also been described as the 'wrong pocket problem', i.e., the area where the 312 investment is made does not only benefit from it: a larger area is benefited as well [29] . One 313 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this this version posted December 24, 2020. ; https://doi.org/10.1101/2020.12.22.20223651 doi: medRxiv preprint example of this geographically grounded, cost-benefit oriented approach has recently been 314 applied in an emergency vaccination against rabies conducted in Tanzania, where a limited 315 intervention in a small ring that surrounded a large city also benefitted the large city [11] . 316

Findings support the view that, to stop epidemic exponential growth, testing policies should 317 aim at detecting all infected individuals (including asymptomatic cases). In spite of its relevance, 318 testing policy seems an area that lacks interdisciplinary scientific expertise. Approaches that 319 include temporal and geo-referenced test positivity data, as well as cost-benefit oriented 320 considerations, may contribute toward filling that gap. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

The copyright holder for this this version posted December 24, 2020. ; https://doi.org/10.1101/2020.12.22.20223651 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

The copyright holder for this this version posted December 24, 2020. ; peaks, increasing percentages of the TP over two or more weeks, and/or TP percentages higher 478 than 5% (red rectangles, E-J). In all ten countries investigated, the peak TP percentage occurred 479 earlier than the peak number of new cases. Source: Worldometer. 480 481 482

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

The copyright holder for this this version posted December 24, 2020. investigated, which partially included the set reported in Fig. 1 . Four countries showed high TP 496

percentages but few or no high tests/case ratios (A-D). One country displayed a hybrid pattern, 497 which was characterized by a few high TP percentages and a few high tests/case ratios (E). The 498 remaining four countries predominantly exhibited high tests/case ratios (F-I). 499 500 he re TP rn, he . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

The copyright holder for this this version posted December 24, 2020. ; https://doi.org/10.1101/2020.12.22.20223651 doi: medRxiv preprint 501 Fig. 4 . Pattern-vs. threshold-based evaluations of epidemic control measures. Four 502 countries located in three continents showed new cases even when the percentage of test 503 positivity (TP) was < 1% (ovals, A-D). Because new cases were observed over protracted (more 504 than two-week long) periods of time, it is concluded that no numerical cutoff (e.g., TP < 5%) 505

indicates an epidemic is under control. In contrast, patterns -such as a perpendicular data 506 inflection-promote inferences, regardless of any cutoff value. 507 508 est re %) ata . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

The copyright holder for this this version posted December 24, 2020. ; https://doi.org/10.1101/2020.12.22.20223651 doi: medRxiv preprint 509 510 Fig. 5 . Geo-epidemiologically specific, cost-benefit effective, differential testing. The test 511 positivity (TP) percentages reported in all municipalities of Puerto Rico, on Sept 11, 2020, are 512 depicted. Area 1 shows the municipality of Jayuya. Area 2 identifies a larger, surrounding region 513 that includes ~42 times more people and covers an area ~ 8.5 times larger than those of Jayuya. 514

Given Jayuya +50% TP, a greater testing effort conducted in this municipality may yield large 515 benefits, which may also include the surrounding area (Area 2) and, indirectly, benefit the 516 western half of the island. Vice versa, keeping the same level of testing performed before may 517 lead to long-lasting, costly consequences: if the virus circulating within Jayuya reached the 518 surrounding area and the prevalence of COVID-19 became similar to the one affecting Jayuya 519 (which is estimated by test positivity), then a much larger number of people (up to 42 times 520 larger) could become infected, who would reside in an area 8.5 times larger, i.e., control would 521 then be much harder, longer and costlier. 522 523 est re on ya. ge he ay he ya es ld . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. tests/million inhabitants. Testing policies may be adjusted based on patterns generated by 526 cases (A-D) and fatalities (E-H). When testing grows slower than cases or deaths (when the 527 curvature approaches a vertical pattern), testing is likely to be insufficient or inadequate. 528 of by he . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

The copyright holder for this this version posted December 24, 2020. ; https://doi.org/10.1101/2020.12.22.20223651 doi: medRxiv preprint

Population and Energy -a talk by Al Bartlett

Critical response time (time available to 361 implement effective measures for epidemic control): model building and evaluation

Connecting network properties of rapidly 364 disseminating epizoonotics

Where and when to vaccinate? 367 Interdisciplinary design and evaluation of the 2018 Tanzanian anti-rabies campaign

World Health Organization. COVID-19-virtual press conference -30

Screening in Public Health and Clinical Care: 374 similarities and differences in definitions, types, and aims -a systematic review

The Spanish flu (1918-20): The global impact of the largest influenza pandemic in 378 history