Association of Varying Clinical Manifestations and Positive Anti–SARS-CoV-2 IgG Antibodies: A Cross-Sectional Observational Study

Background The complex relationship between clinical manifestations of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) and individual immune responses is not fully elucidated. Objective To examine phenotypes of symptomatology and their relationship with positive anti–SARS-CoV-2 IgG antibody responses. Methods An observational study was performed of adults (≥18 years) from 5 US states. Participants completed an electronic survey and underwent testing to anti–SARS-CoV-2 nucleocapsid protein IgG antibody between May and July 2020. Latent class analysis was used to identify characteristic symptom clusters. Results Overall, 9507 adults (mean age, 39.6 ± 15.0 years) completed the survey; 6665 (70.1%) underwent antibody testing for anti–SARS-CoV-2 IgG. Positive SARS-CoV-2 antibodies were associated with self-reported positive SARS-CoV-2 nasal swab result (bivariable logistic regression; odds ratio [95% CI], 5.98 [4.83-7.41]), household with 6 or more members (1.27 [1.14-1.41]) and sick contact (3.65 [3.19-4.17]), and older age (50-69 years: 1.55 [1.37-1.76]; ≥70 years: 1.52 [1.16-1.99]), but inversely associated with female sex (0.61 [0.55-0.68]). Latent class analysis revealed 8 latent classes of symptoms. Latent classes 1 (all symptoms) and 4 (fever, cough, muscle ache, anosmia, dysgeusia, and headache) were associated with the highest proportion (62.0% and 57.4%) of positive antibodies, whereas classes 6 (fever, cough, muscle ache, headache) and 8 (anosmia, dysgeusia) had intermediate proportions (48.2% and 40.5%), and classes 3 (headache, diarrhea, stomach pain) and 7 (no symptoms) had the lowest proportion (7.8% and 8.5%) of positive antibodies. Conclusions SARS-CoV-2 infections manifest with substantial diversity of symptoms, which are associated with variable anti–SARS-CoV-2 IgG antibody responses. Prolonged fever, anosmia, and receiving supplemental oxygen therapy had strongest associations with positive SARS-CoV-2 IgG.


INTRODUCTION
Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection induces variable humoral immune responses. Although most patients with coronavirus disease 2019 (COVID-19) develop SARS-CoV-2 IgA, IgM, and IgG antibodies over days and weeks following infection, target antigens and quantitative titers can vary. 1 The importance of antibody titers was demonstrated in 2 recent studies. One study showed an inverse correlation between IgG levels and persistence of viral shedding. 2 Another study showed a dose-dependent relationship between titers of antieSARS-CoV-2 spike-protein IgG in transfused convalescent plasma and patient survival in COVID-19. 3 Assay-and antibody-dependent factors may impact antibody measurements and their use in determining individual immunity or population-level seroprevalence. 4,5 Importantly, such levels may also correlate with patient characteristics such as age and severity of illness in hospitalized patients. 2,6 Little is known about the clinical and demographic predictors of positive antieSARS-CoV-2 IgG antibody responses in patients with mild SARS-CoV-2 infections. We hypothesize that age, sex, and symptom severity among other factors impact the strength of antieSARS-CoV-2 nucleocapsid protein IgG antibody response. In addition, symptoms of COVID-19 exhibit substantial variation. 7,8 The significance of such variation is unclear, particularly as it relates to variation in antieSARS-CoV-2 IgG antibody responses. We additionally hypothesize that specific phenotypes of SARS-CoV-2 symptoms are predictors of IgG antibody responses to SARS-CoV-2 nucleocapsid protein. In this large-scale study, we examine the diversity of SARS-CoV-2 symptomatology and its relationship to IgG antibody responses in a convalesced population with a high SARS-CoV-2 seroprevalence.

Study design
The study involved a 2-stage sampling design as previously described. 9 Stage 1 was designed to determine the self-reported symptoms and outcomes of SARS-CoV-2 in adults. Subjects were recruited by local not-for-profit and social service organizations within orthodox Jewish communities across 5 states (California, Connecticut, Michigan, New Jersey, and New York) between May 13 and July 6, 2020. A cross-sectional survey invitation was sent to adults; 12,626 individuals began the survey process, with 9,507 adults completing the survey (completion rate, 75.4%). In stage 2, a subset of 6665 adults (70.1% response rate) had antibody testing performed shortly after completing the survey. Electronic informed consent was taken and disclosure of the study purpose was done before beginning the survey. The study was open to all participants and did not require participants to have SARS-CoV-2 symptoms or exposures to participate. The study was approved by IntegReview institutional review board.

Survey
The survey was developed to determine the most common symptoms and outcomes of SARS-CoV-2 (see this article's Online Repository at www.jaci-inpractice.org). The survey included 81 data points including questions about patient demographics, contacts with other Covid-19einfected individuals in the household, symptoms of SARS-CoV-2, whether they tested positive for SARS-CoV-2 by nasal swab (yes/no), and required oxygen for SARS-CoV-2 throughout their illness. The survey was administered via the Health Insurance Portability and Accountability Act-compliant and secure Research Data Capture software.

Antibody measurement
AntieSARS-CoV-2 antibody measurements were performed at the Mayo Clinic Laboratory (Rochester, Minn) using the Epitope Diagnostics ELISA (San Diego, Calif), established and used for clinical reporting of qualitative test for detection of IgM or IgG antibodies to the nucleocapsid protein from SARS-CoV-2. For the purposes of this study, IgG results were reported as described previously, with index value thresholds of greater than or equal to 1.21, less than or equal to 1.01, and 1.01 to 1.20 for positive, negative, and indeterminate results. 10

Data analysis
Baseline characteristics were determined and summary statistics were estimated. Frequency and proportion of SARS-CoV-2 symptoms were estimated overall and in those with positive SARS-CoV-2 antibodies.
Latent class analysis (LCA) was used to examine phenotypical patterns of SARS-CoV-2 symptoms. LCA uses observed categorical or binary data to identify patterns, or latent classes. Conditional probabilities were estimated using maximum likelihood to characterize the latent classes by indicating the chance that a member would give a certain response (yes/no) for the specific symptom. Conditional probability plots are presented, where probabilities closer to 0 or 1 indicate lower or higher chances, respectively. LCA regression models examine the differential effects of individual variables across unobserved classes. The ideal number of latent classes and best-fitting models were selected by minimizing the corrected Akaike information criterion and Bayesian information criterion and interpretability. c 2 tests were used to test the associations of age, sex, household size (above or below the median household size), and the presence of sick contacts in the household with membership in the latent classes.
Among adults with positive SARS-CoV-2 antibodies, bivariable logistic regression models were constructed to determine whether demographic or household characteristics are associated with having at least 1 SARS-CoV-2 symptom or individual SARS-CoV-2 symptom (dependent variables). Crude odds ratio (OR) and 95% CI were estimated. Similarly, Poisson regression models were constructed to identify associations of the number of self-reported SARS-CoV-2 symptoms (dependent variable). Crude risk ratios and 95% CI were estimated. Multivariable models included sex (male/female), age (continuous), household size, and number of household sick contacts. Adjusted OR, relative risk, and 95% CI were estimated. Two-and 3-way statistical interactions were tested between covariables.
Bivariable logistic regression models were also constructed to determine whether demographic or household characteristics, SARS-CoV-2 symptoms, and latent classes of SARS-CoV-2 symptoms are associated with having positive SARS-CoV-2 antibody tests (binary dependent variables). OR and 95% CI were estimated. Multivariable models included all variables from the bivariable models (except for self-report of any symptoms or fever) and state of residence. Adjusted OR and 95% CI were estimated. Two-and 3way statistical interactions were tested between covariables.
Bivariable logistic regression models were constructed to elucidate the impact of household size overall and number of children age 0 to 3, 4 to 10, or 11 to 17 years (independent variables) on positive  J ALLERGY CLIN IMMUNOL PRACT VOLUME 9, NUMBER 9 SARS-CoV-2 antibodies (binary dependent variable). Multivariable models controlled for age, sex, and state of residence. All data processing and statistical analyses were performed in SAS version 9.4.3 (SAS Institute, Cary, NC). Complete data analysis was performed; that is, subjects with missing data were excluded. A 2-sided P value of less than .05 was considered statistically significant.

Predictors of SARS-CoV-2 symptoms
Among respondents with positive SARS-CoV-2 antibodies, 178 (8.9%) reported no SARS-CoV-2 symptoms. Self-report of any SARS-CoV-2 symptoms was associated with households with 6 or more members ( Table E3 in this article's Online Repository at www.jaci-inpractice.org). In multivariable models, the associations remained significant for age 70 years or more, female sex, household sick contacts, and having a positive COVID nasal swab result. There were no significant 2-or 3-way statistical interactions.

Patterns of SARS-CoV-2 symptoms
To identify statistically significant homogeneous patterns of SARS-CoV-2 symptoms (latent classes) among subjects based on their observed binary reporting of symptoms (n ¼ 9311 with complete symptom data), we used LCA. The best-fit model had 8 classes. Conditional probabilities of having different SARS-CoV-2 symptoms are plotted in Figure 1, A.
Class 7 had the highest membership probability (27.3% of the survey cohort) and consisted of very low probabilities of any symptom (Table I). Class 2 had the next highest membership probability (18.8% of the survey cohort) and had higher probabilities of muscle ache and headache. Class 4 (17.8% of the survey cohort) had higher probabilities of fever, cough, muscle ache, anosmia, dysgeusia, and headache. Class 6 (15.0% of the survey cohort) had higher probabilities of fever, cough, muscle ache, and headache. Class 1 (6.2% of the survey cohort) had higher probabilities of all symptoms. Class 8 (5.9% of the survey cohort) had higher probability of anosmia and dysgeusia. Class 5 (5.4% of the survey cohort) had higher probabilities of fever, cough, muscle ache, headache, diarrhea, and stomach pain. Class 3 (3.7% of the survey cohort) had higher probabilities of headache, diarrhea, and stomach pain.
There were significant associations of latent class membership with age, sex, household size, and the presence of sick contacts in the household (c 2 , P < .0001 for all) (Figure 1, B). Membership   (Table III). These associations remained significant in multivariable models controlling for age, sex, and state of residence.

DISCUSSION
In this large-scale observational cohort, we demonstrate substantial diversity of symptoms from SARS-CoV-2 infection. The 5 most common symptoms reported in the survey cohort were headache, myalgia, cough, anosmia, and fever, whereas in patients with serologically confirmed COVID-19, the most common symptoms were similarly myalgia, headache, cough, anosmia, and fever. Certain clinical characteristics have particularly strong relationships with antieSARS-CoV-2 IgG antibody response, including prolonged fever, anosmia, and receiving supplemental oxygen therapy, which is consistent with previous reports. 13 We used LCA to further elucidate the COVID symptom complex and its relationship with IgG antibody responses. LCA is a statistical method used to identify a set of discrete subgroups or latent classes of individuals based on their responses to a set of categorical variables. 14 Adults who experienced all symptoms (class 1) and those who specifically had fever, cough, muscle Bivariable logistic regression models were constructed with SARS-CoV-2 IgG test results (positive vs negative/indeterminate) as the dependent variable and age, sex, household size, household sick contacts, any symptoms, any fever, peak fever, duration of fever, other individual symptoms, receiving supplemental oxygen therapy, self-report of a positive SARS-CoV-2 nasal swab test result, and duration of overall illness as the independent variables. Crude ORs and 95% CI were estimated. Multivariable regression model 1 included all variables from the bivariable models (except for self-report of any symptoms or fever), and state of residence. Adjusted OR and 95% CI were estimated. Bold indicates statistical significance (P < .05).
ache, anosmia, dysgeusia, and headache (class 4) were most likely to have positive antieSARS-CoV-2 antibodies and self-reported SARS-CoV-2 PCR nasal swabs. Interestingly, anosmia and dysgeusia alone (class 8) were not associated with as robust an immune response as fever, cough, muscle ache, and headache (class 6). This appears to be in contrast to a previous study in health care workers, and may suggest an association with a more robust immunophenotype with class 6. 15 Similar to a previous study, 16 we found that headache was reported in all but 2 latent classes (7 and 8). These symptom patterns may be useful to predict the likelihood of having SARS-CoV-2 infection and related outcomes based on symptoms alone, and potentially guide occupational and public health recommendations regarding resource allocation. In addition, these symptom patterns can help predict which patients likely have positive IgG antibodies to SARS-CoV-2 and guide clinical recommendations regarding vaccination and social distancing requirements. We found that households with more than 5 residents had more SARS-CoV-2 symptoms and positive antibodies in bivariable analyses. These associations were attenuated after controlling for sick contacts in multivariable models. That is, people living in larger households have more potential sick contacts during an outbreak. In particular, the presence of adolescents was associated with higher antibody positivity, but not younger children. These results have important ramifications for public health and school policy. First, sociocultural groups with more persons per household may be prone to higher rates of infections. Second, households with more persons, particularly adolescents, may warrant more caution with respect to mitigation strategies for preventing community-based spread of SARS-CoV-2 infection. Third, as schools and workplaces around the world develop policies to reopen in-person, it is important to distinguish between regions and sociocultural groups with typically larger versus smaller household sizes. Adolescents and adults who might become infected with SARS-CoV-2 in school and work can transmit the virus to far more people residing in a larger versus smaller households, thereby potentially increasing community spread.
Strengths of this study include the large sample size, inclusion of a wide age range, and spectrum of disease severity, including a fairly high proportion of younger adults and milder symptoms. This allowed for comparison of symptoms and IgG antibody responses as a function of age and symptom severity. However, there are limitations. This ambulatory cohort may not accurately reflect the symptomatology and serologic profiles of patients with more severe disease. Viral PCR positivity was assessed via survey rather than direct testing. Moreover, this cohort included persons from communities with early COVID-19 outbreaks, when PCR testing was not yet widely accessible across the United States. Many participants were unable to get PCR testing, leading to a low proportion (6%) who reported having a positive PCR test result; that is, patients who reported not having a positive PCR test result may not have been tested. Thus, it is possible that some participants who experienced COVID-19 symptoms had other viral illnesses. We analyzed antibodies to the nucleocapsid protein but not spike protein, which may have led to lower rates of antibody positivity. Data on travel history and contact tracing were not available. This was a largely Ashkenazi Jewish population. COVID-19 hit the Orthodox Jewish community in the United States particularly hard, especially in the early days when much was unknown. At that time of great loss, Jewish communities around the United States rallied to participate in research to help the millions of other people impacted by the pandemic. Although the cohort included broad representation of age and sex, there was limited racial diversity. It is additionally unclear whether seroconversion using currently available assays reflects immunity to SARS-CoV-2, especially because the absence of T-cell data in a large portion of our population overlooks a phenotype of immunity that may be especially important in asymptomatic patients. 17,18 This latter aspect Bivariable logistic regression models were constructed with SARS-CoV-2 IgG test results (positive vs negative/indeterminate) as the dependent variable and presence of children ages 0-3, 4-10, or 11-17 y (0, !1) and number of other adults in the household (0, 1, 2-5, >5) as the independent variables. Crude ORs and 95% CIs were estimated. Multivariable regression models included age (continuous) and sex (male, female) as fixed-effects variables, and state of residence. Adjusted ORs and 95% CI were estimated. The age cutoffs were selected a priori on the basis of previous reports suggesting that younger children are least likely to transmit COVID, followed by older children, then adolescents and greatest transmission occurring in adults. 11,12 Bold indicates statistical significance (P < .05).
J ALLERGY CLIN IMMUNOL PRACT VOLUME 9, NUMBER 9 regarding cellular immunity patterns is an aspect that we are following up in future studies.

CONCLUSIONS
SARS-CoV-2 was associated with a heterogeneous profile of symptoms. Adults who experienced prolonged fevers and anosmia and received supplemental oxygen therapy, as well as those who experienced a multitude of symptoms in combination, had the highest odds of positive SARS-CoV-2 antinucleocapsid IgG. Future studies should examine the impact of these characteristics on other aspects of immunity to SARS-CoV-2. Analyses were limited to adults with positive SARS-CoV-2 IgG test results. Bivariable logistic regression models were constructed with any SARS-CoV-2 symptoms (yes vs no) as the dependent variable and age, sex, household size, household sick contacts, and self-report of a positive SARS-CoV-2 nasal swab test result as the independent variables. Crude ORs and 95% CI were estimated. Multivariable regression models included all variables from the bivariable models, and state of residence. Adjusted ORs and 95% CI were estimated. Bold indicates statistical significance (P < .05). Analyses were limited to adults with positive SARS-CoV-2 IgG test results. Bivariable Poisson regression models were constructed with number of self-report of any SARS-CoV-2 symptoms as the continuous dependent variable and age, sex, household size, and household sick contacts as the independent variables. Crude RR and 95% CI were estimated. Multivariable models included all variables from the bivariable models, and state of residence. Adjusted RR and 95% CI were estimated. Bold indicates statistical significance (P < .05).