Relative sensitivity of anterior nares and nasopharyngeal swabs for initial detection of SARS-CoV-2 in ambulatory patients: Rapid review and meta-analysis

Abstract

Nasopharyngeal (NP) swabs are considered the “gold standard” for diagnosing SARS-CoV-2 infections, but anterior nares or mid-turbinate swabs (nasal swabs) are often used. We performed a meta-analysis comparing the sensitivity of nasal and nasopharyngeal swabs against a composite reference standard for the initial diagnosis of SARS-CoV-2 infection in
ambulatory patients. The study is registered on PROSPERO (CRD42020221827). Data
sources included studies appearing between January 1, 2020, and March 20, 2021, identified by searches of PubMed, medRxiv and bioRxiv. Studies included at least 20 subjects who simultaneously provided nasal and nasopharyngeal specimens for reverse transcription-polymerase chain reaction testing, and for which confusion matrices could be constructed. Authors individually assessed studies for inclusion and compared assessments.
Each author independently extracted all data elements; differences were reconciled by a review of initial data sources. Extracted data included specimen site, patient characteristics,
collection site, and confusion matrices comparing results for nasal and nasopharyngeal
swabs. Assessed against a composite reference standard, anterior nares swabs are less
sensitive (82% – 88%) than nasopharyngeal swabs (98%). For populations with 10% specimen positivity, the negative predictive values of all swab types were greater than 98%. Midturbinate and anterior nares swabs seem to perform similarly. The lower sensitivity associated with nasal swab SARS-CoV-2 diagnosis is justified by the ability to screen more patients and reduced personal protective equipment requirements. Our conclusions are limited by the small number of studies and the significant heterogeneity of study designs and study outcome

Introduction

Rapid identification of SARS-CoV-2 viral infection is important for patient treatment, isola-tion of asymptomatic carriers, and following the community spread of infection [1]. Although flocked-swab nasopharyngeal samples (NPS) have constituted a “gold standard” upper respira-tory system specimens, samples obtained by swabbing the anterior nares (ANS, also known as “nasal”) are now often employed, because these samples can be self-collected, reducing both exposure and the need for personal protective equipment. Some studies have shown nearly identical performance for anterior nares swabs (ANS) and NPS [2]; others have suggested a significantly inferior performance for ANS, particularly later in the course of disease [3]. In addition, some authors have proposed mid-turbinate swabs (MTS) as an alternative to both ANS and NPS. In this paper we set out to answer the question: “When nucleic acid amplifica-
tion tests RT-PCR tests for SARS-CoV-2 are employed for initial diagnosis, what is the relative sensitivity for detection of virus when anterior nares or mid-turbinate samples are used rather
than nasopharyngeal swabs?” We focus on initial diagnosis to assure that the results of our analysis pertain to the way nucleic acid testing is currently used in SARS-CoV-2 patient man-agement, rather than for treatment monitoring or releasing patients from isolation or quaran-tine. To answer the question, we conducted a rapid systematic review based on PRISMA principles (a widely accepted set of guidelines for conducting systematic reviews) [4] and com-pared each sampling method with a composite reference standard [5] We included in our anal-ysis only studies in which NPS and ANS or MT swabs were obtained on the same day, and for which both specimens were analyzed using either a reverse transcription-polymerase chain reaction (RT-PCR) or transcription-mediated amplification (TMA) approach and included in our meta-analysis only studies that demonstrated a low-risk of patient selection bias.

Methods

The study is registered on PROSPERO (CRD42020221827). The protocol was not published. We searched PubMed, medRxiv and bioRxiv for studies appearing between January 1, 2020, and March 20, 2021. Preliminary searches showed that the search strategy [nose OR (nasal swab) OR nares) AND (sensitivity OR comparison OR diagnosis) AND (covid19 OR SARS-CoV-2)] was able to capture nearly all relevant papers. In addition, reference lists from review papers and papers identified in the search were also examined for possibly relevant work. Following the initial identification of papers, the titles and abstracts were screened to eliminate papers not meeting the prespecified inclusion criteria. Inclusion criteria required studies with at least 20 specimen pairs (or triads), each consisting of an ANS or MTS and a NPS obtained on the same day. Papers were not excluded based on title and abstract unless they indicated that the paper represented a review, or that they did not compare specimen types based upon testing using either RT-PCR or transcription-mediated amplification (TMA) methods. Papers remaining after this process were rescreened in their entireties. Papers were included only if they presented data in a manner that enabled construction of a confusion matrix of all possible combinations of test results: +/+, -/-, +/-, and -/+. Studies were excluded if they used a differ-ent diagnostic device for testing ANS/MTS and NPS. After screening 336 distinct papers, only 11 met inclusion criteria and were analyzed, as shown in the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) flow diagram [6] (Fig 1). No attempt was made to obtain data from the investigators involved in these published studies.

The potential for study-associated bias was evaluated using QUADAS2 [7, 8], which is used to assess patient selection bias and other biases that may be introduced by the manner in which testing is conducted. For example, patient selection bias or spectrum bias (case-mix bias), which occurs when test performance differs among patient populations, was assessed

Fig 1. Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) flow diagram. 􀀁Some papers excluded for more than one reason †Meta-analysis performed with and without studies that had risk of selection bias.

from the perspective of testing as an initial diagnostic method for ambulatory patients because site-specific viral load may change with time. Bias assessment does not constitute a judgment on the quality of the study, which may have been performed to demonstrate assay validity, assessment of recovery, or purposes different than that for which we evaluated potential bias. A predetermined data extraction form included author, study type, inclusion and exclusion criteria, setting, swab and sample types, transport medium, description or manufacturer of nucleic acid amplification assays, and space to record confusion matrices. Evaluations were performed by both authors, who resolved several minor differences regarding the characterization of potential bias after rereading and discussion. Meta-analysis calculations were performed by TJO.

Three papers (4 cohorts) with a low risk of patient selection bias were deemed appropriate to include in a meta-analysis based upon ANS and NPS [2, 9, 10] and these studies were analyzed using a diagnostic effects model (der Simion–Laird) [11] as implemented by

OpenMetaAnalyst software program [12]. In order to meet the continuity condition for the calculation, 0.5 was added to all cells.
Because choice of any particular sample type as a “gold standard” yields a biased estimate of relative sensitivity compared to all other sample types, a composite reference standard (CRS)[5] was computed for each study. Neither equivocal results nor assay failures were included in sensitivity calculations, nor in construction of the CRS. Confidence limits for sensitivity
(reported in Table 1) were computed using Newcombe’s efficient score method, as imple-mented in the Vassarstats Clinical Calculator 1 (http://vassarstats.net/). Predictive value calcu-lations were carried out using the MedCalc diagnostic test evaluation calculator (https://www. medcalc.org/calc/diagnostic_test.php). Pre-specified criteria for inclusion in the formal meta-analysis were studies with low-risk of spectrum bias, similar sample types (ANS, MTS or NPS), and similar clinical environments (drive-through / community health center or hospital). Sev-eral exploratory (not pre-specified) meta-analyses were also performed.
Results
Three hundred thirty-six papers were considered for inclusion; of these, 11 studies comprising 12 cohorts met inclusion criteria (Fig 1) [2, 3, 9, 10, 13–17]. A brief summary of the studies is included in this review (Table 1), and more detailed notes about each paper (including the results used in this review) are available in the S1 File. The included cohorts ranged from 38 to 498 patients, with six having fewer than 100 patients. One study separately presented data for two cohorts [9]. The risk of spectrum bias associated with the study population, or method of recruit-ment, was rated as “high” for seven studies that included patients who had previously tested posi-tive [3, 13, 15–17]. One of these studies was also deemed to have a high risk for “flow and timing” bias [13]. No studies were believed to have a risk of index test bias or reference test bias, in part because the inclusion criteria minimized the opportunity for such bias. The prevalence of SARS-CoV-2 infection within the cohorts, as measured using the CRS, varied from 5.2% to 85%.
Estimates of NPS sensitivity ranged from 79% to 100%, with 7 of the 12 cohorts giving esti-mates of 98% or greater. Only one study, in which an estimate of 79% for NPS sensitivity was identified [16], estimated NPS sensitivity at less than 90%. Several factors may have contrib-uted to this low estimate. This study included subjects that had previously tested positive for SARS-CoV-2. As a result of study design, positive patients were probably in the later stages of infection. In addition to ANS and NPS, two different saliva samples were collected for these patients and used in computation of the CRS. Previous studies have suggested that saliva is more sensitive than NPS for SARS-CoV-2 detection in the latter stages of disease [18, 19], and exclusion of saliva from the CRS results in an estimate of 100% for NPS sensitivity. This effect is not seen in a study in which testing that included saliva specimens was conducted in a walk-up or drive-through setting that focused on initial diagnosis.
Eight studies examined the sensitivity of anterior nares specimens, arriving at sensitivity estimates ranging from 42 to 94%; five of these studies showed ANS sensitivity between 83 and 89%. In every study and cohort included in our analysis, ANS were inferior to NPS when com-pared to the composite reference standard, with a relative inferiority ranging from 4% to 46%. MTs was inferior to NPS in five of the six cohorts examined, and appears to be comparable to or greater than that of ANS, varying between 82% and 96% sensitivity [2, 13, 20–22]. Confi-dence intervals for sample types overlapped significantly for all studies included in this review.
Two papers (Callahan and the Griesemer Albany cohorts) included in this review showed dramatically lower sensitivity for ANS than for NPS [3, 9]. One of these studies (Callahan) found that the reduced sensitivity was independent of the swab type or transport approach [3]. The authors suggested that the sensitivity of ANS is dependent on the sensitivity of the

RT-PCR assay employed, and that SARS-CoV-2 patients with lower viral loads may thus get false-negative results when ANS is used in place of NPS. This conclusion was also reached by Pinninti [22]. Given that many of the patients in the Callahan study were tested as part of fol-low-up after laboratory-confirmed diagnosis, it is possible that more rapid reduction of viral loads in the anterior nares than in the nasopharynx with disease progression accounts for the results of these studies, and that heterogeneity in patient populations may account for differ-ences observed in other studies included in this review. Reduced viral loads in asyptomatic vs symptomatic patients could possibly also account for the reduced sensitivity for ANS in the Albany cohort, since Ct values tend to be higher in asymptomatic than symptomatic patients [23], but viral load information was not published for these patients.

Examination of Table 1 shows no obvious relationship between the prevalence of disease in the tested populations and the sensitivity of ANS, MTS, or NPS identified in those populations, nor does it suggest a relationship between hospitalization and sensitivity. Perhaps because all the testing platforms have low limits of detection, neither was there an obvious relationship between testing platform and relative sensitivity of the swab types. Although both cohorts with poor sensitivity for ANS had these specimens collected by healthcare workers, Table 1 shows that professionally collected specimens provided high sensitivity in most cohorts. Similarly, the study summaries in the supplemental data do not suggest great differences in the perfor-mance of different nasal swabbing materials.
Fig 2 presents computer output for the meta-analyses based on studies of ANS having a low risk of recruitment bias. The pre-specified analysis estimated the sensitivity of NPS, as com-pared with the composite reference standard, at 98%, with lower and upper bounds of 94%
and 99% respectively (Fig 2A). There was no statistical suggestion of heterogeneity (τ2 = 0.000, Q[df = 2] = 0.284, p = 0.963, I2 = 0). In contrast, the sensitivity of ANS (Fig 2B) was estimated at 82% (95% CI 65% – 92%). There was significant heterogeneity (τ2 = 0.709, Q[df = 2] = 16.902, p<0.001, I2 = 81.357). An unplanned exploratory analysis (Fig 1C) that excluded the Griesemer Albany cohort [9] showed AN sensitivity of 88% (95% CI 83% – 91%), with little
suggestion of heterogeneity (τ2 = 0.000, Q[df = 2] = 1.750, p = 0.417, I2 = 0). There is no over-lap in 95% confidence intervals between ANS and NPS specimens.
Because visual inspection suggested little difference between the results obtained from stud-ies believed to have a low risk of recruitment bias and studies believed to have a higher risk, as well as little or no difference between ANS and MTS, we conducted a second unplanned exploratory meta-analysis that included all ANS and MTS cohorts. Assessment of NPS showed
at most a modest and not statistically significant suggestion of heterogeneity (τ2 = 0.337, Q
[df = 11] = 16.437, p = 0.126, I2 = 33.08), and yielded a sensitivity estimate for NPS of 95%
(95% CI 92–97%), as seen in Fig 3A. This result is quite similar to that seen when only cohorts with a low risk of recruitment bias were included in the analysis. When the results of ANS and MTS from all studies were combined, there was once again strong evidence for statistical het-
erogeneity (τ2 = 0.882, Q[df = 2] = 70.904, p<0.001, I2 = 84.486). Removing the Callahan [3] and Griesemer [9] Albany cohorts reduced heterogeneity (τ2 = 0.000, Q[df = 11] = 3.969, p = 0.913 I2 = 0.000) and yielded an estimate of ANS/MTS swab sensitivity of 86% (95% CI 82–89%) (Fig 3B). This point estimate is somewhat higher than that originating from cohorts hav-ing low risk of spectrum bias. For this exploratory analysis, there is no overlap between 95%confidence intervals for NPS and ANS/MTS.

Discussion

Nasopharyngeal swabs have been the “gold standard” for diagnosis of upper respiratory infec-tions, including SARS-CoV-2 infection. The studies we examined have reaffirmed this

Fig 2. Computer output showing sample sensitivity forest plots and meta-analysis results (studies with low risk of recruitment bias).(A) Results for nasopharyngeal specimens. (B) Results for anterior nares specimens. (C) Exploratory analysis in which the Albany cohort from Griesemer [9] was excluded.

approach. Only one study included here shows a diagnostic sensitivity less than 90% for NP swabs. Our meta-analysis of cohorts undergoing initial diagnostic testing gives an estimated 98% (95% CI 94% – 99%) sensitivity for NP swabs. These estimates are similar to that obtained by repeat testing [24]. This results in high confidence regarding NP swab sensitivity for initial diagnosis of SARS CoV-2 infection when assessed against the CRS. The meta-analysis suggests noticeably lower sensitivity (95% CI 82% – 88%) for ANS; review of the data in Table 1 suggests than MTS gives results comparable to those of ANS. These results are similar to those found in a recent review and meta-analysis of SARS-CoV-2 testing that used a different methodologic approach [25], and parallel the results seen in influenza testing [26–28]. In a testing environ-ment in which the “true” SARS-CoV-2 infection rate is 10%, the negative predictive value of ANS swabbing, based on the composite reference standard, is estimated as 98.1% – 99.3%, whereas that of NPS is estimated at 99.3% – 99.9%. Although somewhat less sensitive than NPS for detection of SARS-CoV-2, ANS and MTS sample types are more amenable to self-

Fig 3. Computer results from unplanned exploratory analysis to include all anterior nasal swab and middle turbinate samples. (A) Results for NPS from all cohorts included in this review. (B) ANS+MTS from all cohorts except the Griesemer [9] Albany cohort and the Callahan [3] cohort. Only ANS results are included from the Tu study [2].

collection, which speeds specimen acquisition and reduces the use of personal protective equipment. This is a significant advantage when testing large numbers of individuals and when more frequent retesting is needed. Since frequent retesting may be more important than test sensitivity for controlling pandemic spread, the reduced sensitivity associated with ANS or MTS may be a reasonable trade-off for throughput. Nevertheless, our analysis strongly suggests that NPS remains the “gold standard” sample for SARS-CoV-2 when used in diagnostic (as opposed to screening) settings in which time and resources are not constrained.
Although variability in nasal swab construction and transport conditions cannot account for the very low sensitivity for ANS seen in two of the studies, it may account for some of the variation seen among the others. Flocked, foam, and spun polyester swabs all function reason-ably well, but none is optimal for all operators and transport conditions [29–31]. Nevertheless, the studies meeting inclusion criteria for our systematic review protocol do not provide evi-dence for the superiority/inferiority of any ANS or MTS swabbing device, transport approach, or for healthcare worker-collected vs self-collected ANS/MTS. Although the data suggest that, least with some assay systems, ANS may be less effective in subjects with low viral loads than it is in those with higher viral loads, the data that we examined does not provide a robust test of ANS efficacy in the identification of asymptomatic infection. Data in the literature strongly suggests that asymptomatic patients have lower viral loads than symptomatic patients, but also show that many asymptomatic patients are effectively diagnosed using ANS [23].
Our analysis is limited by the small number of cohorts which met inclusion criteria; inclusion of studies included fewer than 20 subjects might have provided additional information. The use of an abbreviated search strategy may miss publications that otherwise may have satisfied inclusion criteria. The use of a CRS, which defines the false-positive rate as zero for all assays, may introduce a downward bias in the estimates of sensitivity [32]. This bias varies somewhat based on study size and precise cell frequencies. The computational requirement to add a constant to cells containing zero also creates a downward bias in sensitivity estimates, which is greatest for studies in which the lowest sensitivities are identified. This may have resulted in a downward bias of up to 4% in the meta-analysis estimate of ANS sensitivity. Finally, the studies included in the review demon-strate considerable heterogeneity in assay design, patient population, and disease prevalence. Thus, this systematic review and meta-analysis may provide more insight into the range of possi-ble results than they do for the specific performance in any given setting.
This systematic review also has several strengths. It focuses on sample sets that were taken at the same time, which differs from most other systematic reviews. The number of samples is substantially greater than that of any individual study based on simultaneous sample compari-son, and allows sensitivity estimates based on meta-analysis. Furthermore, our meta-analysis focuses on initial diagnosis of infection in patients who were not previously tested positive for SARS-CoV-2 or hospitalized for treatment. This may yield a more accurate assessment of per-formance in community-based initial testing, and provides evidence that, while ANS is inferior to NPS for initial diagnosis, it nevertheless retains high negative predictive value in most popu-lations. In the absence of a “gold standard” for diagnosis of SARS-CoV-2 infection, the CRS is a relatively unbiased approach to a “reference standard” and is more likely to provide a better sense of the true performance difference associated with various sampling approaches than intrinsically biased comparisons. Our conclusions are similar to those found in a recently pub-lished systematic review and meta-analysis which was performed using somewhat different methodology, lending strength to our conclusions [25].
During this pandemic, the medical and scientific community has come together in an unprecedented collaborative manner to offer multiple testing options and publish research to inform laboratory testing decisions. By conducting this systematic review, we are able to pro-vide unbiased evidence to colleagues in medicine and public health, that despite minor differ-ences in diagnostic sensitivity, the ANS and MTS testing may be reasonable alternatives given the practical limitations associated with NPS.

Supporting information

S1 Checklist. PRISMA 2009 checklist.
(DOC)
S1 File. Supplemental data: Study summaries and confusion matrices.
(DOCX)

Acknowledgments

We wish to thank Yuan-Po Tu (The Everett Clinic-Part of Optum) for valuable discussions which led to this work, and Kirsten St. George of the New York State Department of Public Health Wadsworth Laboratory for providing information on the specimen collection approach used in reference 9.

Author Contributions

Conceptualization: Yaolin Zhou, Timothy J. O’Leary.
Data curation: Yaolin Zhou, Timothy J. O’Leary.
Formal analysis: Yaolin Zhou, Timothy J. O’Leary.
Investigation: Yaolin Zhou.
Methodology: Timothy J. O’Leary.
Writing – original draft: Timothy J. O’Leary.
Writing – review & editing: Yaolin Zhou, Timothy J. O’Leary.

References

  1. Tu Y-P, O’Leary TJ. Testing for severe acute respiratory syndrome–coronavirus 2: challenges in getting good specimens, choosing the right test, and interpreting the results. Critical Care Medicine. 2020; 48: 1680–1689. https://doi.org/10.1097/CCM.0000000000004594 PMID: 32826428
  2. Tu Y-PP, Jennings R, Hart B, Cangelosi GGA, Wood RC, Wehber K, et al. Swabs collected by patients or health care workers for SARS-CoV-2 testing. The New England Journal of Medicine. 2020; 383: 494–496. https://doi.org/10.1056/NEJMc2016321 PMID: 32492294
  3. Callahan C, Lee R, Lee G, Zulauf KE, Kirby JE, Arnaout R. Nasal-swab testing misses patients with low SARS-CoV-2 viral loads. medRxiv [Preprint]: 2020; 2020.06.12.20128736. Available from https://www. medrxiv.org/content/10.1101/2020.06.12.20128736v1.full.pdf PMID: 32587981
  4. McGrath TA, Alabousi M, Skidmore B, Korevaar DA, Bossuyt PMM, Moher D, et al. Recommendations for reporting of systematic reviews and meta-analyses of diagnostic test accuracy: A systematic review. Systematic Reviews. 2017;6. https://doi.org/10.1186/s13643-017-0407-9 PMID: 28088237
  5. Baughman AL, Bisgard KM, Cortese MM, Thompson WW, Sanden GN, Strebel PM. Utility of composite reference standards and latent class analysis in evaluating the clinical accuracy of diagnostic tests for pertussis. Clinical and Vaccine Immunology. 2008; 15: 106–114. https://doi.org/10.1128/CVI.00223-07 PMID: 17989336
  6. Moher D, Liberati A, Tetzlaff J, Altman DG. Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statement. PLoS Medicine. 2009; 6: e1000097. https://doi.org/10.1371/journal. pmed.1000097 PMID: 19621072
  7. Schueler S, Schuetz GM, Dewey M. The revised QUADAS-2 tool. Annals of Internal Medicine. 2012; 156: 323; author reply 323–4. https://doi.org/10.7326/0003-4819-156-4-201202210-00018 PMID: 22351721
  8. Whiting PF, Rutjes AWS, Westwood ME, Mallett S, Deeks JJ, Reitsma JB, et al. Quadas-2: A revised tool for the quality assessment of diagnostic accuracy studies. Annals of Internal Medicine. 2011;
    155:529–536. https://doi.org/10.7326/0003-4819-155-8-201110180-00009 PMID: 22007046
  9. Griesemer SB, van Slyke G, Ehrbar D, Strle K, Yildirim T, Centurioni DA, et al. Evaluation of specimen types and saliva stabilization solutions for SARS-CoV-2 testing. Journal of Clinical Microbiology.
    2021;59. https://doi.org/10.1128/JCM.01418-20 PMID: 33674284
  10. Hanson KE, Barker AP, Hillyard DR, Gilmore N, Barrett JW, Orlandi RR, et al. Self-collected anterior nasal and saliva specimens versus health care worker-collected nasopharyngeal swabs for the molecu-lar detection of SARS-CoV-2. Journal of Clinical Microbiology. 2020;58. https://doi.org/10.1128/JCM. 01824-20 PMID: 32817233
  11. DerSimonian R, Laird N. Meta-analysis in clinical trials. Controlled Clinical Trials. 1986; 7: 177–188. https://doi.org/10.1016/0197-2456(86)90046-2 PMID: 3802833
  12. Wallace BC, Dahabreh IJ, Trikalinos TA, Lau J, Trow P, Schmid CH. Closing the gap between method-ologists and end-users: R as a computational back-end. Journal of Statistical Software. 2012; 49: 1–15. https://doi.org/10.18637/jss.v049.i05
  13. McCulloch DJ, Kim AE, Wilcox NC, Logue JK, Greninger AL, Englund JA, et al. Comparison of unsuper-vised home self-collected sidnasal Swabs with clinician-collected nasopharyngeal swabs for detectionof SARS-CoV-2 infection. JAMA Network Open. 2020; 3: e2016382. https://doi.org/10.1001/jamanetworkopen.2020.16382 PMID: 32697321
  14. Pe´re´ H, Podglajen I, Wack M, Flamarion E, Mirault T, Goudot G, et al. Nasal swab sampling for SARS-CoV-2: A convenient alternative in time of nasopharyngeal swab shortage. Journal of Clinical Microbiol-ogy. 2020. https://doi.org/10.1128/jcm.00721-20 PMID: 32295896
  15. Berenger BM, Fonseca K, Schneider AR, Hu J, Zelyas N. Sensitivity of nasopharyngeal, nasal and throat swab for the Detection of SARS-CoV-2. medRxiv [Preprint]: 2020; 2020.05.05.20084889. Avail-able from https://www.medrxiv.org/content/10.1101/2020.05.05.20084889v1.full.pdf
  16. Kojima N, Turner F, Slepnev V, Bacelar A, Deming L, Kodeboyina S, et al. Self-collected oral fluid and nasal swab specimens demonstrate comparable sensitivity to clinician-collected nasopharyngeal swab specimens for the detection of SARS-CoV-2. Clinical Infectious Diseases. 2020. https://doi.org/10. 1093/cid/ciaa1589 PMID: 33075138
  17. Federman DG, Gupta S, Stack G, Campbell SM, Peaper DR, Dembry LM, et al. SARS-CoV-2 detection in setting of viral swabs scarcity: Are mrsa swabs and viral swabs equivalent? PLoS ONE. 2020; 15: e0237127. https://doi.org/10.1371/journal.pone.0237127 PMID: 32756602
  18. SoRelle JA, Mahimainathan L, McCormick-Baw C, Cavuoti D, Lee F, Thomas A, et al. Saliva for use with a point of care assay for the rapid diagnosis of COVID-19. Clinica Chimica Acta. Elsevier B.V.; 2020. pp. 685–686. https://doi.org/10.1016/j.cca.2020.09.001 PMID: 32910978
  19. Dogan OA, Kose B, Agaoglu NB, Yildiz J, Alkurt G, Demirkol YK, et al. Does sampling saliva increase detection of SARS-CoV-2 by RT-PCR? Comparing saliva with oro-nasopharyngeal swabs. Journal of Virological Methods. 2021;290. https://doi.org/10.1016/j.jviromet.2020.114049 PMID: 33387561
  20. Federman DG, Gupta S, Stack G, Campbell SM, Peaper DR, Dembry LM, et al. SARS-CoV-2 detection in setting of viral swabs scarcity: Are mrsa swabs and viral swabs equivalent? PLoS ONE. 2020; 15: e0237127. https://doi.org/10.1371/journal.pone.0237127 PMID: 32756602
  21. Liu M, Li Q, Zhou J, Ai W, Zheng X, Zeng J, et al. Value of swab types and collection time on SARS-COV-2 detection using RT-PCR assay. Journal of Virological Methods. 2020;286. https://doi.org/10. 1016/j.jviromet.2020.113974 PMID: 32949663
  22. Pinninti S, Trieu C, Pati SK, Latting M, Cooper J, Seleme MC, et al. Comparing nasopharyngeal and midturbinate nasal swab testing for the identification of severe acute respiratory syndrome coronavirus
  23. Clinical Infectious Diseases. 2020 [cited 26 Mar 2021]. https://doi.org/10.1093/cid/ciaa882 PMID: 32596725
  24. Tu YP, Iqbal J, O’Leary T. Sensitivity of ID NOW and RT–PCR for detection of SARS-COV-2 in an ambulatory population. eLife. 2021;10. https://doi.org/10.7554/ELIFE.65726 PMID: 33876726
  25. Long DR, Gombar S, Hogan CA, Greninger AL, O’Reilly-Shah V, Bryson-Cahn C, et al. Occurrence and timing of subsequent severe acute respiratory syndrome coronavirus 2 reverse-transcription polymer-ase chain reaction positivity among initially negative patients. Clinical Infectious Diseases. 2021; 72: 323–326. https://doi.org/10.1093/cid/ciaa722 PMID: 33501950
  26. Lee RA, Herigon JC, Benedetti A, Pollock NR, Denkinger CM. Performance of saliva, oropharyngeal swabs, and nasal swabs for SARS-CoV-2 molecular detection: a systematic review and meta-analysis. Journal of Clinical Microbiology. 2021 [cited 5 Apr 2021]. https://doi.org/10.1128/jcm.02881-20 PMID: 33504593
  27. Irving SA, Vandermause MF, Shay DK, Belongia EA. Comparison of nasal and nasopharyngeal swabs for influenza detection in adults. Clinical Medicine and Research. 2012; 10: 215–218. https://doi.org/10. 3121/cmr.2012.1084 PMID: 22723469
  28. Malosh RE, Petrie JG, Callear AP, Monto AS, Martin ET. Home collection of nasal swabs for detection of influenza in the Household Influenza Vaccine Evaluation Study. Influenza and other respiratory viruses. 2020 [cited 1 Feb 2021]. https://doi.org/10.1111/irv.12822 PMID: 33107200
  29. Frazee BW, Rodrı´guez-Hoces de la Guardia A, Alter H, Chen CG, Fuentes EL, Holzer AK, et al. Accu-racy and discomfort of different types of intranasal specimen collection methods for molecular influenza testing in emergency department patients. Annals of Emergency Medicine. 2018; 71: 509–517.e1. https://doi.org/10.1016/j.annemergmed.2017.09.010 PMID: 29174837
  30. Elson D, Daivaa N, Park C, Desind S, Kwiatkowski R, Chakravorty S, et al. Evaluation of sample collec-tion and transport strategies to enhance yield, accessibility, and biosafety of COVID-19 RT-PCR testing. medRxiv [Preprint]: 2021; 2021.03.03.21251172. Available from https://www.medrxiv.org/content/10. 1101/2021.03.03.21251172v1.full.pdf
  31. Hart B, Tu YP, Jennings R, Verma P, Padgett LR, Rains D, et al. A comparison of health care worker-collected foam and polyester nasal swabs in convalescent COVID-19 patients. PLoS ONE. 2020;15. https://doi.org/10.1371/journal.pone.0241100 PMID: 33108384
  32. Padgett LR, Kennington LA, Ahls CL, Samarasinghe DK, Tu YP, Wallander ML, et al. Polyester nasal swabs collected in a dry tube are a robust and inexpensive, minimal self-collection kit for SARS-CoV-2 testing. PLoS ONE. 2021; 16: e0245423. https://doi.org/10.1371/journal.pone.0245423 PMID: 33852576
  33. Takwoingi Y, Guo B, Riley RD, Deeks JJ. Performance of methods for meta-analysis of diagnostic test accuracy with few studies or sparse data. Statistical Methods in Medical Research. 2017; 26: 1896–1911. https://doi.org/10.1177/0962280215592269 PMID: 26116616

Readers found these studies helpful