We studied 177 patients who provided 645 distinct excess diagnostic material samples (Table 1). Patients were from diverse ethnic backgrounds (34% White, 35% non-White, 31% unreported; Appendix 1), and the median age was 64 years (interquartile range [IQR] 52–77 years). Fifty-seven percent were male, and 73% had >1 concurrent condition. Nineteen percent were asymptomatic and did not report respiratory symptoms at admission; these patients tested positive for SARS-CoV-2 infection while receiving treatment for other conditions. Among the 143 symptomatic patients, the median time from symptom onset to testing was 6 days (IQR 3–9 days). Of the 177 patients, 166 (94%) were hospitalized, 7 (4%) were staff, and 4 (2%) were outpatients. Of the hospitalized patients, 44 (27%) died (median time to death was 19.1 days [IQR 14.8–24.8 days]), 108 (65%) were discharged (median length of stay was 19.3 days [IQR 10.6–31.1 days]), and 14 (8%) remained hospitalized at the end of the study. Sixty-three (38%) patients were admitted to intensive care during the study.

Figure 1

Figure 1. Antibody dynamics in patients with severe acute respiratory syndrome coronavirus 2, United Kingdom, 2020. A) NOD by days after first positive PCR result. Blue indicates seroconverting patients; red indicates nonseroconverting...

We normalized optical densities proportional to levels of SARS-CoV-2 IgG (Figure 1, panel A). Of the 177 patients, 149 (84% [95% CI 78%–89%]) had already seroconverted at the time of the first serologic test, 13 (7.3% [95% CI 4.3%–12.1%]) seroconverted after the first serologic test, and 15 (8.5 % [95% CI 5.2–13.5%]) did not seroconvert during the entire follow-up period (Appendix 1). Of the 15 patients who did not seroconvert, samples from beyond day 20 were available for 4 patients (26%); we did not detect IgG in these samples. This finding suggests that 2.0%–8.5% of patients might not develop detectable IgG against SARS-CoV-2.

We plotted NODs by time after a patient’s first positive swab sample (Figure 1, panel B) and after symptom onset (Figure 1, panel C). NODs plateaued »12 days after PCR and »19 days after symptom onset; this time difference is consistent with the median time of 6 days between symptom onset and PCR. After seroconversion, mean NODs remained stable over the course of the study (up to »60 days after symptom onset).

We assessed whether the rate of seroconversion was associated with patient age (<70 or >70 years), sex, or respiratory symptoms. None of these variables were discernably associated with seroconversion rates (Appendix 1). NOD IgG levels were not associated with sex or the presence of respiratory symptoms. (Appendix 1).

Patients of non-White race had higher mean NODs than those of White race (1.06 vs. 0.85; F = 1.61, df = 119; p = 0.04 by unpaired Student t-test) (Appendix 1). No other differences were associated with race. We used a multivariate analysis to identify variables independently associated with NODs; the mean NOD was associated only with age, peak CRP, and race. Although age, sex, peak CRP, number of concurrent conditions, respiratory symptoms, and race were associated with patient outcome in the univariate analysis, only peak CRP was associated with poor outcome in the multivariate analysis (Appendix 1).

Persons who seroconverted were older than those who did not (median age 65.5 vs. 41.0 years; p<0.01 by Mann-Whitney test) and more likely to have >1 concurrent condition (124/130 vs. 38/47; p<0.01 by Fisher exact test). History of hypertension was associated with a higher probability of seroconversion (74/75 persons with hypertension vs. 88/102 persons without hypertension; p<0.01 by Fisher exact test). Body mass index was higher among the group who seroconverted (25.7 vs. 21.2; p = 0.03 by Mann-Whitney test).

Figure 2

Figure 2. Relationships between CRP levels, symptoms, outcomes, and NODs of patients with severe acute respiratory syndrome coronavirus 2, United Kingdom, 2020. A–C) CRPs at diagnosis for A) 113 symptomatic (open circles)...

Unlike other markers of inflammation, CRP is routinely measured in patients with COVID-19. Rising CRP levels are indicators of a poor prognosis (if other causes are excluded), and are associated with cytokine release syndrome (5; Y. Woo, unpub. data, https://osf.io/mxsvw). CRP levels were significantly higher in patients with respiratory symptoms at diagnosis than in those without symptoms (Figure 2, panel A). Patients who died or required intensive care during the study period had higher CRP levels than patients who did not die or require intensive care (Figure 2, panel B). Patients who did not seroconvert had lower CRP levels than those who did (Figure 2, panel C). Peak CRP levels had more pronounced associations with outcomes and seroconversion than did CRP levels at the time of the first positive swab sample result (Figure 2, panels D–F). CRP levels peaked a median of 12 days (IQR 8–17 days) after symptom onset and 4 days (IQR 1–11 days) after the first positive PCR result. Other inflammatory markers, such as peak D-dimer, fibrinogen, and ferritin, were also higher in patients with respiratory symptoms at diagnosis. However, these data were available for fewer patients (Table 2; Appendix 1).

Our results illustrate serologic responses over the course of SARS-CoV-2 infection. Serologic tests can enhance diagnostic capability, especially during later infection (4,6,7), when viral loads might decrease. Serologic testing might also inform surveillance, seroepidemiologic studies, and contact tracing. Our study shows that a substantial proportion of COVID-19 patients require 3–6 weeks to generate antibodies. Furthermore, 2.0%–8.5% of patients do not have detectable antibodies within 60 days after infection. Most research on antibody dynamics came from China during the early stages of the pandemic (4,6,8). Here, we describe variables that influence IgG dynamics in SARS-CoV-2 infections in diverse populations.

The performance metrics of this ELISA (Appendix 2) are comparable to other validated assays. This first-generation ELISA might confirm infection in patients without a virologic diagnosis. We applied this test to study an ethnically and clinically diverse population. In most persons who seroconverted, the conversion was relatively rapid; NODs remained stable for weeks after infection (Figure 1). The probability of seroconversion was associated with increased age and concurrent conditions such as hypertension and increased body mass index. Higher NODs were associated with non-White race, admission to hospital, and higher peaks for inflammatory markers, such as CRP. Higher antibody titers are associated with clinical severity (i.e., death or admission to intensive care during study) in our cohort, in agreement with findings from other studies (4).

CRP is a sensitive marker of elevated proinflammatory cytokines, including interleukin 6. These cytokines might play a central role in cytokine release syndrome, which is associated with increased risk for death (Y. Woo, unpub. data, https://osf.io/mxsvw). Interventions such as tocilizumab, an interleukin-6 receptor antibody, interrupt the proinflammatory cascade. Such interventions might limit disease progression and reduce risk for death (9; E. Baker, unpub. data, https://osf.io/d2nh8); they are being studied in several randomized clinical trials. In our study, a small proportion of patients did not seroconvert within 20 days after testing positive for SARS-CoV-2 infection. Several mechanisms might explain this finding. First, these patients might never seroconvert. Second, their immune responses might be confined to other antigens or mediated through T cells. Another probable explanation is that some relatively mild infections might be restricted to the mucosal cells of the respiratory tract, where antibody responses are dominated by the secretory immune system. In this scenario, the systemic immune system might produce little or no IgG.

The association of higher NODs with elevated CRPs could indicate several potential pathways. For example, antibody responses might be closely related to cytokine response syndrome, which in turn is associated with more severe disease and death. Alternatively, elevated CRPs might indicate a more pronounced innate immune response in persons already at risk for severe disease and death. This heightened innate response might be associated with a higher viral load (potentially caused by enhanced viral replication mechanisms) and genetic interactions that influence innate inflammatory pathways. Therefore, a higher viral load might lead to higher NODs for antibodies in the acquired immune response pathways. Small trials on the potential therapeutic benefits of interventions using passive antibody transfer (10) suggest that the heightened innate response hypothesis is more probable (11). Higher antibody responses are also associated with higher doses of a nonreplicating Ad5-vectored vaccine for SARS-CoV-2 (12).

Limitations of PCR include difficulties with sampling; different sample types and techniques yield varying results. Furthermore, PCR demonstrates diminishing diagnostic yield for COVID-19 as respiratory viral loads fall and symptoms subside (8,13). It might also produce false positives caused by lingering viral nucleic acid, which is not infective yet can persist for weeks after infection. Contamination could also occur during sample handling; because PCR requires amplification steps, this assay has heightened risk for contamination. Serologic testing, and the ability to detect viral antigens, may increase diagnostic accuracy for COVID-19. Our findings support early studies suggesting that physicians should consider these diagnostic modalities in conjunction, especially when a patient has negative PCR results but has symptoms of COVID-19 (6). Many COVID-19 patients experience a delay in care, a trend that emphasizes the importance of containment strategies that encourage isolation.

One limitation of our study is that it is based mainly on hospitalized patients, of whom 1 in 5 did not have COVID-19 symptoms. Further studies should document antibody dynamics of patients with less severe infections, such as healthcare workers (14), and patients with low viral loads at the time of consultation. Our findings will complement the large cross-sectional and longitudinal serologic surveys, especially as high-quality tests become more widely available. NODs were within a limited dynamic range (we could not conduct dilution studies because of small sample volumes) but nevertheless associated with clinically relevant features of COVID-19. Prospective studies are assessing the relationships between viral loads and serologic responses in patients. Regular and long-term serologic assays will be essential to monitoring the duration of the humoral response and its protective role against SARS-CoV-2.

When interpreting serologic assays of COVID-19 patients, physicians should consider factors that can influence the probability of seroconversion. Our study elucidates some of these factors. We found that less severe infections and younger age were associated with reduced probability of seroconversion. Risk factors for more severe disease, such as non-White race, increased age, and hypertension, are also associated with increased inflammatory responses, higher normalized antibody titers, and probability of seroconversion.