https://orcid.org/0000-0001-6876-6262
Abstract
A severe acute respiratory syndrome coronavirus 2 serosurvey among first responder/healthcare personnel showed that loss of taste/smell was most predictive of seropositivity; percent seropositivity increased with number of coronavirus disease 2019 symptoms. However, 22.9% with 9 symptoms were seronegative, and 8.3% with no symptoms were seropositive. These findings demonstrate limitations of symptom-based surveillance and importance of testing.
Detection of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection has often relied on symptom-based screening when coronavirus disease 2019 (COVID-19) testing resources are limited [1]. As a result, presymptomatic or asymptomatic transmission occurred and uninfected persons unnecessarily self-isolated [1, 2]. Although universal testing has been called for [2], it has not been widely implemented. Evaluations of how well symptoms predict infection have focused on association with COVID-19 test positivity, mainly reverse transcriptase polymerase chain reaction (RT-PCR) [1, 3–6]. However, RT-PCR accuracy is dependent on timing relative to infection—persons with asymptomatic infection may not present for testing, and symptomatic infections may not be identified if testing occurs too long after infection [2, 3]. Thus, the predictive power of symptoms could be under- or overestimated when evaluated by RT-PCR positivity compared with seropositivity.
We assessed the association between seropositivity and prior COVID-19 symptoms in a large serologic survey [7, 8] to inform COVID-19 surveillance and testing strategies. Serologic testing (conducted > 2 weeks after infection) [9–12] is a cumulative measure of infection over recent months and can more completely reveal the association between COVID-19 symptoms and infection.
METHODS
Between May 17 and July 2, 2020, serologic surveys were conducted among first responders and healthcare personnel in the Detroit, Michigan, metropolitan area and in New York City [7, 8, 12]. Region 2 South Healthcare Coalition (Michigan Department of Health and Human Services) and the New York City Department of Health and Mental Hygiene distributed survey recruitment materials to hospitals and emergency medical and public service agencies who shared materials with employees. Participation was voluntarily initiated by accessing a web-based questionnaire that included informed consent and eligibility screening for COVID-19 symptoms and RT-PCR positivity in the prior 2 weeks [9–11]. The survey collected data on presence of any of 9 COVID-19 symptoms since March 1 and other items (see Supplement). Upon survey completion, participants received information about providing a blood sample at or near their workplace within the next 1–7 days. Serologic testing was performed with the ORTHO Clinical Diagnostics VITROS Immunodiagnostic Products Anti-SARS-CoV-2 immunoglobulin G (IgG) test [9]. Individual results were not shared with employers. The Centers for Disease Control and Prevention (CDC) did not have access to personal identifiers. This activity was reviewed by the CDC and was conducted consistent with applicable federal law (45CFR6, 21CFR56; 42USC§241(d); 5USC§552a; 44USC§3501 et seq) and CDC policy.
After exclusions (invalid test results, n = 88, and implausible self-reported weight and/or height, n = 25), 40 938 participants were included. Unadjusted SARS-CoV-2 seropositivity rates were calculated with 95% confidence intervals (CI) using exact binomial models. Adjusted seropositivity rates were estimated using 2 logistic regression models with covariates shown in the Supplementary Table 1 (a priori model) [1, 7, 12, 13] plus either nonmutually exclusive dichotomous variables for 9 COVID-19 symptoms (model 1) or 1 variable for number of symptoms (model 2). There was no evidence of multicollinearity: variance inflation factors were < 1.9. A classification and regression trees (CART) approach was used to identify symptom combinations predictive of seropositivity. There were no meaningful symptom combinations that directly predicted seroprevalence. Therefore, infection severity was used as a surrogate response variable: seeking healthcare for COVID-19 symptoms and/or being hospitalized (n = 6351, 15.5%) vs not. The 9 COVID-19 symptoms were evaluated as risk predictors. The optimal binary tree resulted in a high predictive area under the curve (0.903) with a 5% error rate for predicting participants with less severe symptoms and a 50% error rate for predicting participants with more severe symptoms. Symptom combinations identified by CART predictive of ≥ 70% prevalence of more severe symptoms were then assessed for seropositivity prevalence. SAS 9.4 software (Research Triangle Institute) was used for all analyses.
RESULTS
Overall, 16.2% (95% CI, 15.9–16.6) of participants and 8.3% (95% CI, 7.9–8.6) of asymptomatic participants were seropositive. Seropositivity decreased with age, was higher among men vs women, lower among non-Hispanic White persons compared with other race/ethnic groups and increased with increasing weight status (Supplementary Table 1). Asymptomatic participants (n = 23 294) represented 56.9% of the study population and 28.9% of those who tested seropositive (n = 1921 of 6645). Participants reporting new loss of sense of taste or smell since March 1, 2020 (8.5% of participants), had the highest seropositivity (76.5%; 95% CI, 75.1–77.9) (Figure 1A). Seropositivity ranged from 14.1% (95% CI, 13.1–15.1) among participants reporting any 1 symptom to 77.1% (95% CI, 72.0–81.6) for all 9 symptoms (Figure 1B).
Crude and adjusted† percent seropositivity for SARS-CoV-2 (A) by symptom type, (B) by number of symptoms, and (C) crude seropositivity for specific symptom combinations. †Adjusted for age group, sex, race/ethnicity, weight status, jurisdiction, medical conditions, and exposure to a coronavirus 19–positive household member. *Symptom combinations identified using classified regression tree approach. Abbreviations: SARS-CoV-2, severe acute respiratory syndrome coronavirus 2; SOB, shortness of breath.
Seroprevalence by symptom type was attenuated after adjustment, but the pattern of increasing seroprevalence with increasing number of symptoms did not appreciably change (Figure 1B). Loss of taste/smell had the strongest association with seropositivity (55.6% adjusted seroprevalence; 95% CI, 53.5–57.7), followed by fever, chills, muscle aches, and cough (18.6%–26.0%). Adjusted seroprevalence among asymptomatic participants was 14.5% (95% CI, 13.9–15.1), a level similar for sore throat (12.0%; 95% CI, 11.5–12.5%), diarrhea (14.4%; 95% CI, 13.6–15.1), and headache (15.7%; 95% CI, 15.1–16.4).
Seropositivity among participants reporting symptom combinations identified by CART analysis is shown in Figure 1C. Seropositivity was 62.8% among participants reporting a combination of fever, shortness of breath (SOB) and chills, and 82.1% among participants with fever, SOB, and loss of taste/smell. The combination of fever, SOB, chills, and headache had a seroprevalence similar to fever, SOB, and chills (63.6% vs 62.8%).
DISCUSSION
New onset of loss of taste/smell had the strongest association with seropositivity and was more common among younger participants, as seen in other studies [1, 4, 13]. Seropositivity increased with number of symptoms. However, ~25% of participants reporting either loss of taste/smell or all nine symptoms were seronegative. Seropositivity associated with symptom clusters was high, but seronegativity was 18% for the combination with the highest seropositivity (fever, SOB, and loss of taste/smell). Moreover, prevalence of this combination was low, limiting usefulness in symptom screening. Asymptomatic participants had adjusted seroprevalence similar to those reporting each of several other nonspecific symptoms commonly associated with other infections and/or conditions (sore throat, diarrhea, and headache). Asymptomatic participants represented more than one-half of the study population and nearly 30% of those who tested seropositive. Previous studies have found ~20% of persons with positive RT-PCR tests remain asymptomatic [5].
These findings suggest that recalled presence or absence of symptoms is insufficient screening criteria to accurately predict infection status. Other studies have noted the strong association of new onset loss of taste/smell with RT-PCR positivity [1, 4, 5]. Loss of smell in the absence of blocked nasal passages (which may be more indicative of upper respiratory infection or allergies) has been noted to have higher predictive value for RT-PCR positivity [14]. This is one of the first reports showing a strong association with seropositivity [15]. Although obtaining accurate RT-PCR test results is dependent on testing close to time of infection, serology indicates infection over the past few months [12, 16], and may identify asymptomatic people who did not present for RT-PCR testing. This study demonstrates that symptom association with seropositivity is consistent with earlier associations found with RT-PCR positivity. Yet with both types of diagnostic tests, a substantial percentage of those with symptoms suggestive of COVID-19 have negative results. Although it is possible that some are false negatives, findings from a previous analysis in a subsample of this serology survey with previously confirmed infections (2547 participants with state health department-confirmed RT-PCR positive results > 2 weeks before study participation) showed that seronegativity was very low: 3% among those with loss of taste/smell and 1.8% among those with 9 symptoms [12]. In addition, supplemental laboratory testing showed high agreement between the Ortho IgG assay used in this study and the Ortho pan IgG assay and CDC pan IgG assay. All 3 of these assays demonstrated increased sensitivity compared with a fourth assay, suggesting false negatives are not common with the assay used in this study [12]. Thus, in the current study among the entire study population in which most did not have confirmed prior infection, seronegativity most likely represents lack of previous infection rather than false negatives or failure to produce antibody.
Limitations include biases inherent in analysis of first responders and healthcare personnel who were likely healthier and younger compared with the general adult population. Additionally, results from this convenience sample may not be generalizable. Recall bias likely affected symptom reporting with a recall period starting March 1, 2020, for a survey administered May–July 2020. We are unable to differentiate between false-negative results, loss of antibodies, [17] and failure to develop antibodies [12]. The contribution of these 3 patterns to observed seronegativity could vary by symptoms, especially if certain symptoms are more strongly associated with more severe illness.
Overall, our findings highlight the limitations of symptom-based screening. Increasing the number of specific symptoms to improve specificity for seropositivity results in a lower proportion of persons with those combinations and thus, lower utility for screening. The corollary—that there were relatively large percentages of infected people with less specific symptom combinations or no symptoms at all—implies that many infected persons will be missed by symptom-based screening. In addition to principal mitigation measures (e.g., physical distancing), regular RT-PCR or other viral testing of persons at high exposure risk or with continuing contact with patients or the community can augment symptom-based strategies used to prevent transmission.
Supplementary Data
Supplementary materials are available at Clinical Infectious Diseases online. Consisting of data provided by the authors to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the authors, so questions or comments should be addressed to the corresponding author.
Notes
Acknowledgments. The authors thank Addie Crawley, Demetre Daskalakis, Rebecca A. Henseler, Beth Maldin Morgenthau, Preeti Pathela, and Don Weiss of the New York City Department of Health and Mental Hygiene, Queens, New York.
Financial support. Data and specimen collection activities and specimen testing was funded by a US Department of Health and Human Services contract (75P00120C00036). The authors are federal employees (Centers for Disease Control and Prevention), Michigan Department of Health and Human Services Region 2 South Healthcare Coalition, or BIO5 Institute, University of Arizona employees and received no outside funding for their work.
Potential conflicts of interest. The authors: No reported conflicts of interest. All authors have submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest.
Disclaimer. The findings and conclusions in this report are those of the authors and do not necessarily represent the official position of the Centers for Disease Control and Prevention.
