Clinical evaluation of BD Veritor SARS-CoV-2 point-of-care test performance compared to PCR-based testing and versus the Sofia SARS Antigen point-of-care test.

Objectives The clinical performance of the BD Veritor System for Rapid Detection of SARS-CoV-2 antigen (Veritor test), a chromatographic immunoassay that detects the SARS-CoV-2 nucleocapsid antigen, was evaluated on nasal specimens from patients with suspected symptoms of COVID-19. Methods and Materials Two studies were employed. In study 1, nasal specimens and either nasopharyngeal or oropharyngeal specimens from 251 participants with COVID-19 symptoms (<=7 days from symptom onset [DSO]), >=18 years of age, were utilized to compare the Veritor test to the Lyra SARS-CoV-2 PCR Assay (Lyra assay). In study 2, nasal specimens from 361 participants with COVID-19 symptoms (<=5 DSO), >=18 years of age, were utilized to compare performance of the Veritor test to that of the Sofia SARS Antigen FIA test (Sofia test). Positive, negative, and overall percent agreement (PPA, NPA, and OPA, respectively) were the primary outcomes Results In study 1, the Veritor test had a PPA with the Lyra assay ranging from 81.8%-87.5% for the 0-1 through the 0-6 DSO ranges. In study 2, the Veritor test had a PPA, NPA, and OPA of 97.4%, 98.1%, and 98.1%, respectively with the Sofia test. Discordant testing showed one Lyra positive missed by the Veritor test and five Lyra positives missed by Sofia; one Veritor positive result was negative by Lyra. Conclusions The Veritor test met FDA-EUA acceptance criteria for SARS-CoV-2 antigen testing (>=80% PPA point estimate) for the 0-5 and 0-6 DSO ranges. In addition, the Veritor and Sofia tests showed a high degree of agreement for detection of SARS-CoV-2. The Veritor test should facilitate rapid and reliable results for COVID-19 diagnosis.


Conclusions
The Veritor test met FDA-EUA acceptance criteria for SARS-CoV-2 antigen testing (≥80% PPA point estimate) for the 0-5 and 0-6 DSO ranges. In addition, the Veritor and Sofia tests showed a high degree of agreement for detection of SARS-CoV-2. The Veritor test should facilitate rapid and reliable results for COVID-19 diagnosis.
. CC-BY-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.

INTRODUCTION (332 words)
In response to the COVID-19 pandemic, an emphasis has been placed on SARS-CoV-2 diagnostic testing for symptomatic individuals.
[1] Although laboratory-based PCR testing is considered the clinical reference standard for COVID-19 diagnosis, it is associated with some drawbacks, including false-negative reporting. [2][3][4] Also, limitations in capacity have been documented for PCR-based testing, [5,6] which can lead to prolonged time to result (at best 24 hours when sample shipment is considered); and in most cases, dedicated staff and automated platforms are required to provide effective turn-around-time and optimized patient management. [7] In February 2020, the World Health Organization identified point-of-care (POC) testing as a number one priority to address the COVID-19 pandemic.
[8] The relatively small investment in resources and expertise required to perform POC testing makes it ideal for use in decentralized health care settings. [7] Antigen-based immunoassay POC tests for SARS-CoV-2 can target multiple viral antigens, including spike or nucleocapsid protein in a cartridge-based, lateral flow format. Although it is too early to determine whether one target is advantageous over another, evidence supports the efficacy of nucleocapsid detection in these types of antigen-based assays. [9,10] Reports involving SARS and SARS-CoV-2 have demonstrated that the nucleocapsid protein is produced at high levels relative to the other viral proteins. [11,12] In addition, nucleocapsid detection was recently shown, albeit in a serology-based test, to result in higher sensitivity for detection of SARS-CoV-2 compared to spike protein detection. [13] . CC-BY-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.

US-FDA Emergency Use Authorization (EUA) was recently granted for the BD Veritor™
System for Rapid Detection of SARS-CoV-2 (henceforth referred to as "Veritor test"), a POC, chromatographic immunoassay that detects the SARS-CoV-2 nucleocapsid antigen. This report presents the performance data for the Veritor test using nasal swab specimens from  symptomatic individuals compared to the Lyra ® SARS-CoV-2 Assay (henceforth referred to as "Lyra assay"), which was utilized as the clinical reference standard. In a sub-population, Veritor test results were compared with results from another FDA-EUA authorized nucleoprotein antigen test, the Quidel Sofia ® SARS Antigen FIA test (henceforth referred to as "Sofia test").
. CC-BY-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.

Study design
Both studies described here involved a prospective collection of upper respiratory specimens.
Eligible participants were ≥18 years of age and presented with one or more self-reported COVID-19 signs or symptoms. Individuals were excluded if a nasal swab was collected as part of standard of care (SOC). Demographic and healthcare-related information was collected (e.g. symptomology, health history, etc.). No study procedures were performed without an informed consent process or signature of a consent form. This research was performed in accordance with Good Clinical Practice guidelines and the Declaration of Helsinki. This article was prepared according to STARD guidelines for diagnostic accuracy studies reporting. [14] Specimen collection Study 1 (EUA Veritor/Lyra comparison) The first study was utilized to determine whether the Veritor test met FDA-EUA criteria for detection of SARS-CoV-2 in COVID-19 symptomatic individuals (within ≤7 DSO). Collection of specimens from 260 participants occurred across 21 geographically diverse study sites between June 5-11, 2020. Specimens for the Veritor test were from clinician-collected nasal specimens using regular-tipped flocked swabs (Becton, Dickinson and Company, BD Life Sciences-Integrated Diagnostics Solutions, Sparks, MD, USA) inserted approximately 2.5 cm up the nostril (from the edge of the nostril). The swab was rolled five times along the mucosa of the nostril to ensure that sufficient mucus and cells were collected; the process was repeated in the other nostril using the same swab.
. CC-BY-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) The copyright holder for this preprint this version posted September 3, 2020. . https://doi.org/10.1101/2020.09.01.20185777 doi: medRxiv preprint Lyra assay specimens came from nasopharyngeal (NP) or oropharyngeal (OP) swabs; SOC OP or NP swabs were taken before any study swabs. If an NP swab was collected as part of SOC, the participant had the option of having an OP study swab taken in lieu of a second NP swab. All NP (n=217) or OP (n=34) specimens were clinician-collected. Reference testing was performed at TriCore Reference Laboratories while the Veritor testing was performed internally at BD (San Diego, CA, USA).

Study 2 (Veritor/Sofia comparison)
The second study involved comparison of Veritor test performance to the Sofia test for SARS-CoV-2 detection run with Sofia 2 analyzer. Collection occurred from 377 participants with symptoms of COVID-19 (≤5 DSO) from five study sites in the USA. Specimen collection for Veritor testing was performed as described above. For Sofia testing, clinician-collected nasal specimens occurred using methods and swabs described in the IFU (Puritan ® regular foam swabs [Puritan, Guilford, ME, USA]). The specimens were obtained from a single nostril (with the most visible secretion) using gentle rotation. In some cases (n=76 specimens; approximately 20.2% of specimens in study 2), due to an update in the Sofia instructions for use (IFU), participants were instructed to blow their nose prior to nasal swab specimen collection. NP swab specimen collection for the Lyra assay (only for Veritor/Sofia discordant testing) was performed as described above. Testing for Veritor, Sofia2, and discordant Lyra assay, was performed at TriCore Reference Laboratories.

Test procedures
. CC-BY-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) The copyright holder for this preprint this version posted September 3, 2020. . https://doi.org/10.1101/2020.09.01.20185777 doi: medRxiv preprint Swabs were shipped for testing on dry ice (-70°C); nasal swabs were shipped dry and OP/NP swabs were shipped in universal viral transport medium. All testing was conducted with all personnel blinded to all other test results.
The Veritor and Sofia tests were performed according to the manufacturer's IFU (Becton, Dickinson and Company, BD Life Sciences-Integrated Diagnostic Solutions, San Diego, CA [15] and Quidel Corporation, Athens, OH, [16] respectively). Swabs were removed from -70°C storage ≤5 hours prior to the time of testing. Swabs were placed at 2-8°C for ≥2 hours and then at room temperature for 10-30 minutes prior to testing.
For specimen extraction prior to Veritor or Sofia testing, the swabs were added to each respective extraction buffer tubes and mixed for at least 15-30 seconds or 1 minute, respectively.
The extraction buffer/specimen mixture from each test was then added to the sample well of the corresponding test cartridge to initiate the testing. After the assays proceeded for 15 minutes, the test cartridges were inserted into either the Veritor or Sofia 2 analyzer to obtain results.
The Lyra assay was performed according to the manufacturer's IFU (Quidel Corporation. Athens, OH). [17] When using the NucliSENS ® easyMAG ® and the Applied Biosystem 7500 Fast Dx Real-Time PCR instrument, the Lyra assay reports cycle number in a manner that omits the first 10 cycles; here the cycle numbers for the Lyra assay are reported with the first 10 cycles included. The BD MAX™ real time SARS-CoV-2 PCR assay (henceforth termed "MAX assay") was used for discordant testing on residual nasal swabs following Veritor and Lyra testing in . CC-BY-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) The copyright holder for this preprint this version posted September 3, 2020. . https://doi.org/10.1101/2020.09.01.20185777 doi: medRxiv preprint study 1. The MAX assay was performed according to the manufacturer's IFU (Becton, Dickinson and Company, BD Life Sciences-Integrated Diagnostic Solutions, Sparks, MD). [18] Data collection and statistical analyses The primary outcome measures for this study were positive, negative, and overall percent agreement (PPA, NPA, and OPA, respectively) point estimates for the Veritor test compared to results from the Lyra assay in study 1 and for the Veritor test compared to the Sofia test in study 2.
For study 1, the acceptance criteria was a point estimate of ≥80% PPA of the Veritor test when compared to the Lyra assay; clinical evaluation required contiguous enrolment to a minimum of 30 prospectively collected positive specimens as specified in the Antigen Template for Manufacturers (May 11, 2020) for EUA submissions to the US-FDA. [19] Based on an estimated 10% prevalence rate, it was necessary to enroll approximately 300 participants to achieve the required number of positives.
For study 1, positive predictive value, negative predictive value, and accuracy were also calculated as secondary outcomes. [20] Additionally, a t-test (2-tailed) was used to compare means between Lyra assay positive Ct values on specimens matched to Veritor negative and positive test results for SARS-CoV-2 in study 1.
. CC-BY-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) The copyright holder for this preprint this version posted September 3, 2020. . https://doi.org/10.1101/2020.09.01.20185777 doi: medRxiv preprint

Study 1 (EUA study)
Participant reconciliation, demographics, and COVID-19 symptomology The mean and median age of the participants (44.7 and 43 years, respectively) were close (Table   S1). More than half of the participants were female. By race, the largest proportion of participants were White, followed by Black, and then Asian. Approximately 40% were Hispanic or Latino. Cough was the most-reported symptom from participants, followed by muscle pain, and then headache. While the drive-through/tent and outpatient clinic collection site categories represented approximately three-fourths of the collection sites, the research clinic category had the highest positivity rate (22.2%). The mean for DSO among the participants was 3.2 (Table   S1). From 260 participants, specimen sets from nine participants were removed overall. One was removed due to a labeling error, two sets were from participants with no COVID-19 symptoms, four sets had enrollment errors, and two had invalid reference results. Thus, 251 evaluable nasal specimens (each paired with either OP or NP specimens) were included ( Figure S1a).

Veritor test performance and discordant reconciliation
Performance values for the Veritor test are shown by DSO for the participants providing valid specimens ( Table 1). The 0-5 DSO range was the shortest range tested to have a PPA value above 80% and include at least 30 reference positive results. The 0-6 DSO range also met PPA value acceptance criteria. The NPA for the Veritor test was 100% for the 0-1 to 0-5 DSO ranges; however, the NPA value for the 0-6 and 0-7 DSO ranges was 99.5% (95% CI: 97.4, 99.9) ( Table   1). The area under the curve (AUC) values associated with Veritor test performance for the 0-1 through the 0-6 DSO ranges were >0.9; the AUC value for the 0-7 DSO range was 0.88 (Table 1 . CC-BY-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)
The copyright holder for this preprint this version posted September 3, 2020. . https://doi.org/10.1101/2020.09.01.20185777 doi: medRxiv preprint and Figure 1). Performance values for the Veritor test compared to the Lyra Assay were analyzed by number of symptoms reported by participants during sample collection. As shown in Table 2, PPA point estimates were higher for the Veritor test when stratified by ≥2 symptoms versus 1 symptom for both the 0-5 DSO range (88.0% and 66.7%, respectively) and the 0-6 DSO range  (Table 3). From the remaining seven discordants, six were associated with a negative MAX assay result and one was associated with an unresolved result (no detection of internal control in the MAX assay).
. CC-BY-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)
The copyright holder for this preprint this version posted September 3, 2020. . https://doi.org/10.1101/2020.09.01.20185777 doi: medRxiv preprint Figure 4 shows the PPV, NPV, and accuracy associated with the Veritor test by DSO range. As shown, PPV values for the Veritor test were 100%, for the 0-1 through 0-5 DSO ranges. The single Veritor test positive discordant result for the Veritor test that occurred in the 0-6 DSO group resulted in a PPV point estimates of 96.6% and 96.7% for the 0-6 and 0-7 DSO ranges, respectively. The NPV values for the 0-1 to 0-6 DSO groups ranged from 96.8 to 97.2. At 0-7 DSO, the NPV was 95.9.

Study 2 (Veritor/Sofia test comparison study)
Participant reconciliation, demographics, and COVID-19 symptomology From 377 participants, four specimen sets were removed due to inclusion/exclusion criteria and 16 were removed due to non-compliant specimens or invalid test results; 361 evaluable specimens were utilized for this study ( Figure S1b). The mean and median age of the participants (45.3 and 44 years, respectively) were similar. Fever, cough, headache, sore throat, and shortness of breath were the five most common symptoms reported (Table S2) Table 4). Of the seven discordant results, one was Veritor negative/Sofia positive and was positive by the Lyra assay; six were Veritor positive/Sofia negative, with 5 being positive by the Lyra assay and one being negative by the Lyra assay.
. CC-BY-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)
The copyright holder for this preprint this version posted September 3, 2020. . https://doi.org/10.1101/2020.09.01.20185777 doi: medRxiv preprint DISCUSSION (1,300 words) The Veritor test was required to achieve ≥80% PPA relative to the reference standard (with at least 30 positive specimens by reference) in order to be considered acceptable for FDA-EUA.
The Veritor test showed 83.9% and 82.4% PPA for specimens from COVID-19 symptomatic participants that were 0-5 and 0-6 DSO, respectively. In addition, the AUC values for the 0-1 through the 0-6 DSO ranges were excellent (ranging from 0.91-0.94). The results presented here suggest that the Veritor test should be effective in settings that would benefit from POC testing (e.g. decentralized health care settings) in order to classify 0-5 or 0-6 DSO individuals as positive or negative for SARS-CoV-2 infection to support patient management.
Discordant analysis for the 0-1 DSO through 0-6 DSO specimens revealed one false negative result (Participant D from Table 3) (Table S3). Reflex testing (e.g. PCR-based testing) may . CC-BY-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)
The copyright holder for this preprint this version posted September 3, 2020. Eight of the nine false-negative Veritor test results here were matched with Lyra assay Ct values that were above the mean Ct value for the 38 Lyra assay positive results (four were approximately ten cycles above). This, combined with the significant difference in Lyra-matched . CC-BY-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)
The copyright holder for this preprint this version posted September 3, 2020. . https://doi.org/10.1101/2020.09.01.20185777 doi: medRxiv preprint Ct values for the 29 Veritor test true positive and 9 Veritor test false negative specimens, suggests that Veritor-to-Lyra concordance is indirectly proportional to the Lyra assay Ct score.
While PCR-based testing is sensitive for target detection, other testing modalities (such as antigen-based testing) may also be informative and may help clinicians determine the peak time frame during which infections are transmissible. However, more data is needed to establish the efficacy of antigen-based tests, such as Veritor or Sofia, for identifying contagious individualsespecially in the asymptomatic population. The Veritor and Sofia tests are currently only authorized for individuals suspected of having a SARS-CoV-2 infection at 0-5 DSO. In addition, the high level of agreement observed between the Veritor and Sofia tests as the tests is consistent with reported, similar limits of detection for SARS-CoV-2. [15,16] The difference in EUA labeled sensitivity for Sofia (96.7%) vs Veritor (84%) was not supported by this study, probably due to spectrum differences in patient populations in this study versus the Sofia EUA study. The patient population chosen for this study was intended to reflect the performance of the Veritor test in clinical settings where decentralized point-of-care testing such as antigen testing would be most appropriate. The study data presented here included a large proportion of specimens collected from clinical settings (such as drive-through testing, tents, and outpatient clinics), largely include individuals with milder severity illness, compared with study populations that have been used to generate sensitivity estimates for EUA antigen tests where enrollment included Emergency Department patients and hospitalized patients. Several publications have demonstrated an association between severe disease and higher viral loads which could inflate antigen test sensitivity performance estimates when compared to . CC-BY-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)
The copyright holder for this preprint this version posted September 3, 2020. . https://doi.org/10.1101/2020.09.01.20185777 doi: medRxiv preprint performance estimates generated in patients with milder disease.[23-27] The finding in this study of an observed Ct score shift for subjects with 1 symptom vs ≥2 symptoms also supports the possibility that there may even be differences in viral load according to disease severity even amongst patients with milder disease. Analyses here (Table 2 and Figure 2) suggest that ≥2 symptoms lead to a higher PPA than 1 symptom alone, which is reflected by a trend towards lower Ct scores (higher viral load) for specimens from participants with ≥2 symptoms.

Limitations
This study had some limitations. First, the Veritor test was performed on nasal swab specimens; however, the Lyra assay was performed on either NP (or OP) swab specimens per FDA-EUA

requirements. Other EUA submissions (the LumiraDx SARS-CoV-2 Ag Test ["Luminar test"]
and the Abbott BinaxNOW COVID-19 Ag CARD ["Abbott test"]) utilized nasal swab specimens for both the antigen test and the reference PCR assay. Furthermore, MAX from the remnant Veritor nasal swab in this report agreed with negative Veritor results in 7 of 9 discordant specimens. Improved PPA for Veritor versus Lyra may have been achieved through the use of paired nasal swab specimens in the EUA study.
The Sofia assay in study 2 was performed on nasal swabs that were collected either with (Table   S4) or without (Table S5) a nose blowing step prior to collection. The nose-blowing step was an addition to the Sofia test IFU intended only to reduce the frequency of invalid results, and was not included in order to alter the performance of the Sofia test. Although the n is low for specimens with a pre-nose blowing step in study 2, here, the results suggest that the noseblowing step did not alter the overall performance of the Sofia test in relation to the Veritor test.
. CC-BY-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)
The copyright holder for this preprint this version posted September 3, 2020. . CC-BY-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.

This study was funded by Becton, Dickinson and Company; BD Life Sciences-Integrated
Diagnostics Solutions. Non-BD employee authors received research funds as part of this work.

CONFLICTS OF INTEREST
The authors disclose the following conflicts of interest: . CC-BY-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)
The copyright holder for this preprint this version posted September 3, 2020. . CC-BY-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)
The copyright holder for this preprint this version posted September 3, 2020. . CC-BY-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)
The copyright holder for this preprint this version posted September 3, 2020. . CC-BY-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)
The copyright holder for this preprint this version posted September 3, 2020.  . CC-BY-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)
The copyright holder for this preprint this version posted September 3, 2020. . https://doi.org/10.1101/2020.09.01.20185777 doi: medRxiv preprint  . CC-BY-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)
The copyright holder for this preprint this version posted September 3, 2020. . https://doi.org/10.1101/2020.09.01.20185777 doi: medRxiv preprint TABLES    . CC-BY-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)
The copyright holder for this preprint this version posted September 3, 2020. . https://doi.org/10.1101/2020.09.01.20185777 doi: medRxiv preprint . CC-BY-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)
The copyright holder for this preprint this version posted September 3, 2020. . https://doi.org/10.1101/2020.09.01.20185777 doi: medRxiv preprint . CC-BY-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)
The copyright holder for this preprint this version posted September 3, 2020. . https://doi.org/10.1101/2020.09.01.20185777 doi: medRxiv preprint CC-BY-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) The copyright holder for this preprint this version posted September 3, 2020. . https://doi.org/10.1101/2020.09.01.20185777 doi: medRxiv preprint  CC-BY-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) The copyright holder for this preprint this version posted September 3, 2020. . https://doi.org/10.1101/2020.09.01.20185777 doi: medRxiv preprint CC-BY-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) The copyright holder for this preprint this version posted September 3, 2020. . https://doi.org/10.1101/2020.09.01.20185777 doi: medRxiv preprint FIGURE S1 Figure S1. (a) Reconciliation during enrollment of swab specimens from participants, ≥18 years of age, with signs or symptoms of COVID-19 for study 1. Abbreviations: ID, identification; DSO, days from symptom onset (b) Reconciliation during enrollment of swab specimens from participants, ≥18 years of age, with signs or symptoms of COVID-19 for study 2.
Abbreviations: DSO, days from symptom onset . CC-BY-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) The copyright holder for this preprint this version posted September 3, 2020. . https://doi.org/10.1101/2020.09.01.20185777 doi: medRxiv preprint