Decision letter: Quantifying antibody kinetics and RNA detection during early-phase SARS-CoV-2 infection by time since symptom onset

Abstract

Article Figures and data Abstract Introduction Results Discussion Materials and methods Appendix 1 Data availability References Decision letter Author response Article and author information Metrics Abstract Understanding and mitigating SARS-CoV-2 transmission hinges on antibody and viral RNA data that inform exposure and shedding, but extensive variation in assays, study group demographics and laboratory protocols across published studies confounds inference of true biological patterns. Our meta-analysis leverages 3214 datapoints from 516 individuals in 21 studies to reveal that seroconversion of both IgG and IgM occurs around 12 days post-symptom onset (range 1–40), with extensive individual variation that is not significantly associated with disease severity. IgG and IgM detection probabilities increase from roughly 10% at symptom onset to 98–100% by day 22, after which IgM wanes while IgG remains reliably detectable. RNA detection probability decreases from roughly 90% to zero by day 30, and is highest in feces and lower respiratory tract samples. Our findings provide a coherent evidence base for interpreting clinical diagnostics, and for the mathematical models and serological surveys that underpin public health policies. Introduction Since its emergence in December 2019, the SARS-CoV-2 pandemic has been the subject of intense research assessing all facets of the pathogen and its rapid global spread. Serology – the measurement of serum antibodies – provides crucial data for understanding key aspects of infection and epidemiology (Weitz et al., 2020). At the level of populations, serologic data can provide insights into virus spread by enabling estimation of the overall attack rate, and seroprevalence estimates can elucidate the potential for herd immunity (Stringhini et al., 2020; Bryant et al., 2020). In addition, these estimates are essential for developing accurate mathematical models of virus transmission dynamics, which provide the foundation for policies to reopen societies (Krsak et al., 2020; Angulo et al., 2020; Kissler et al., 2020). At the level of individuals, the presence and concentration of antibodies against SARS-CoV-2 are indicators of past exposure, providing insights over a much wider temporal window than other metrics. When considered jointly with PCR testing to detect viral RNA, antibodies substantially improve the probability of detecting present and past infections (Prager et al., 2019). This improvement is highly valuable because RNA detection is typically limited to a relatively brief period of infection, and because PCR sensitivity varies considerably with infection severity and biological sample type (Azkur et al., 2020; Yongchen et al., 2020). Assessment of the levels of different antibody types (e.g. IgG, IgM) may even be used to infer approximately when individuals became infected (Azkur et al., 2020; Chang et al., 2005; Du et al., 2020; Borremans et al., 2016), while detection of neutralizing antibodies may indicate protection from reinfection (Ni et al., 2020). These applications of serologic data depend critically on knowing when different antibodies against the pathogen become detectable (seroconversion time), how their concentrations change over time (antibody level kinetics) and how long they last (antibody decay) (Lipsitch et al., 2020). When these key factors are known, serologic data become a powerful tool for inferring infection attack rate and transmission dynamics in the population (Bryant et al., 2020; Winter and Hegde, 2020). Five months into the pandemic, a remarkable number of serologic studies on the initial immune response against SARS-CoV-2 had been published. These studies were conducted in different laboratories, used different assays and sampling methods, and were performed on different patient groups that showed different clinical manifestations of SARS-CoV-2 infection (Whitman et al., 2020; Lassaunière et al., 2020; Kontou et al., 2020). This extensive variation arising from many sources creates substantial challenges for integrating existing data into one coherent picture of antibody kinetics and viral RNA detection following SARS-CoV-2 infection. In 21 studies reporting the kinetics of anti-SARS-CoV-2 antibodies, we found the use of 8 different antibody assays, 10 different target antigens, and 9 different reported antibody level units (studies are listed in the Materials and methods section). Additionally, the temporal resolution at which studies collect data is highly variable: while some studies report antibody measurements for specific days, many bin results into periods of multiple days or even weeks. Integrated analysis of such diverse data is challenging, and requires statistical methods specifically developed for this purpose. Yet this type of integration is essential to capitalize on the limited and precious data available, to assess to what degree antibody and RNA detection patterns are affected by assay type and target antigen choice, and to establish consensus patterns. For example, a properly integrated analysis would better enable us to test whether antibody patterns depend on disease severity (Huang et al., 2020; Tan et al., 2020). In this study, we quantified IgG and IgM antibody kinetics and RNA detection probability during SARS-CoV-2 infection (up to 60 days post-symptom onset) by aggregating data from published sources. We formally characterized IgG and IgM seroconversion times, detection probabilities over time and antibody level kinetics using methods tailored to accommodate the diverse ways in which data have been collected and reported. We investigated how these variables are affected by disease severity, assay type and targeted antigen, and how patterns differ between IgG and IgM. We also assessed how antibody level kinetics relate to the probability of detecting viral RNA in various biological samples. We estimated mean values as well as observed variation of all variables in order to provide the complete picture required to interpret serological and RNA testing data, inform mitigation strategies and parameterize mathematical models of pathogen transmission while accounting for variability. This formal integration approach enabled us to leverage 3214 data points from 516 individuals with symptoms ranging from asymptomatic to critical, published in 21 studies, resulting in a quantitative synthesis of diverse data on anti-SARS-CoV-2 antibody patterns and RNA detection during the early phase of infection. Results Data extraction We extracted data from 21 preprints and peer-reviewed articles reporting data on SARS-CoV-2 RNA or IgG, IgM or neutralizing antibodies against the virus in humans (see Materials and methods). When available, disease severity information was classified into three groups: asymptomatic/subclinical (n = 11 individuals), mild/moderate (n = 166), and severe/critical (n = 58). Unfortunately, the sample size for the asymptomatic group was too low for quantitative analyses. For 359 individuals, insufficient data were available for disease severity categorization, and these individuals were therefore excluded from analyses of the impact of disease severity. Published results were variously reported as exact days, intervals up to 22 days, or mean times for multiple individuals, while test results were reported as values for one individual or mean values for multiple individuals. Data after 30 days post-symptom onset were particularly underrepresented, but included because in aggregate they provide key insights. When reporting enzyme-linked immunosorbent assay (ELISA) results in the main text, IgG results are shown for assays targeting the nucleoprotein (NP) antigen (ELISA-NP), and IgM results are shown for assays targeting the Spike antigen (ELISA-Spike; whole or subunit), as these assays are most often used for the two antibody types (To et al., 2020; Sethuraman et al., 2020). Results for other assays and antigens are shown in Figure 1—figure supplements 1 and 2. The distribution of seroconversion times Stepwise bootstrapping was used to estimate seroconversion times, using 270 data points from 99 individuals for IgG and 240 data points from 71 individuals for IgM. Mean IgG seroconversion time is 13.3 days post-symptom onset when using ELISA-NP and 12.6 for IgM using ELISA-Spike (Figure 1a). These results do not differ significantly (t = 0.22, df = 7.7, p=0.84) and are similar for magnetic chemiluminescence enzyme immunoassay (MCLIA; Figure 1b). Variation in seroconversion times is substantial regardless of assay, for both IgG (sd = 5.7) and IgM (sd = 5.8). Figure 1 with 5 supplements see all Download asset Open asset Seroconversion time distributions for IgG and IgM. (A) IgG and IgM detected using ELISA. (B) IgG and IgM detected using MCLIA. (C) IgM and (D) IgG seroconversion related to disease severity. IgG and IgM ELISA results are shown for the NP and Spike antigens, respectively, because these had the largest sample sizes. Lines indicate fitted normal distributions. Figure 1—source data 1 Fitted normal distribution parameters for seroconversion time using different assays. SD: standard deviation; ‘N too low’ indicates a sample size too small to compute a mean and SD. https://cdn.elifesciences.org/articles/60122/elife-60122-fig1-data1-v2.docx Download elife-60122-fig1-data1-v2.docx Disease severity does not significantly affect seroconversion time, for IgM or for IgG (Figure 1c–d). Mean IgM seroconversion time for mild/moderate cases is 12.3 days post-symptom onset vs 13.2 for severe/critical cases (t = −0.2, df = 23.5, p=0.83). Mean IgG seroconversion time for mild/moderate cases is 12.9 days post-symptom onset, vs 15.5 for severe/clinical cases (t = −0.96, df = 14.8, p=0.35). A detailed overview of seroconversion time results including means and standard deviations is provided in Figure 1—figure supplements 3–5 and Figure 1—source data 1. Antibody detection probability While estimates of seroconversion time provide information about the first moment at which antibodies can be detected, changes in detection probability over time provide useful information about the proportion of individuals that has detectable antibodies, and hence the expected test sensitivity at the population scale. Sample sizes for these analyses (see Materials and methods) are 8053 data points for IgG and 7935 for IgM, with daily mean sample sizes of 224 and 220, respectively. The probability of detecting IgG (ELISA-NP) increases over time, reaching a maximum around 25–27 days post-symptom onset, at which point between 98% and 100% of individuals test positive (Figure 2a). Detection probability remains at this maximum level for the remainder of the days available in the studies existing at the time of writing (up to 60 days for ELISA-Spike, Figure 2—figure supplement 1). This pattern is consistent across assays (Figure 2—figure supplement 1). IgM (ELISA-Spike) detection probability is similar to that of IgG until its peak near 90% around 23–25 days post-symptom onset, after which it starts to decrease, reaching roughly 65% detection probability around 60 days post-symptom onset (Figure 2, Figure 2—figure supplement 2). Although data on neutralizing antibody presence were sparse, we observe that detection probability rapidly rises to near 100%, where it remains up to the last time available in the dataset (Figure 2—figure supplement 3; 29 days post-symptom onset). Patterns in detection probability do not differ significantly between mild/moderate and severe/clinical cases, aside from a slightly steeper rise for severe/critical cases (Figure 2—figure supplement 4). Figure 2—source datas 1–2 provide the estimated detection probabilities over time for IgG and IgM. Figure 2 with 4 supplements see all Download asset Open asset Detection probability of IgG, IgM and NT (neutralizing) antibody (A) and RNA in different sample types (B) over time since symptom onset. Points are mean values for each day. Bold lines are flexible smoothed splines fit to the data. Error bars indicate binomial exact 95% confidence intervals of the mean, based on daily sample size. Note that error bars after day 30 tend to be large, due to the limited available data. IgG and IgM values are those detected using any assay/antigen. After day 25, results are pooled into 3-day periods in order to improve estimates. Figure 2—source data 1 IgG (ELISA-NP) detection probability. N: sample size (including interpolated samples). https://cdn.elifesciences.org/articles/60122/elife-60122-fig2-data1-v2.docx Download elife-60122-fig2-data1-v2.docx Figure 2—source data 2 IgM (ELISA-Spike) detection probability. N: sample size (including interpolated samples). https://cdn.elifesciences.org/articles/60122/elife-60122-fig2-data2-v2.docx Download elife-60122-fig2-data2-v2.docx Figure 2—source data 3 RNA – upper respiratory tract detection probability. N: sample size (including interpolated samples). https://cdn.elifesciences.org/articles/60122/elife-60122-fig2-data3-v2.docx Download elife-60122-fig2-data3-v2.docx Figure 2—source data 4 RNA – lower respiratory tract detection probability. N: sample size (including interpolated samples). https://cdn.elifesciences.org/articles/60122/elife-60122-fig2-data4-v2.docx Download elife-60122-fig2-data4-v2.docx Figure 2—source data 5 RNA – feces detection probability. N: sample size (including interpolated samples). https://cdn.elifesciences.org/articles/60122/elife-60122-fig2-data5-v2.docx Download elife-60122-fig2-data5-v2.docx RNA detection probability Samples sizes for observed and interpolated data are 7443 and 1793 for upper and lower respiratory samples and 1179 for fecal samples, with mean daily sample sizes of 226, 72 and 39, respectively. The probability of detecting viral RNA in respiratory and fecal samples is high (80–100%) at symptom onset and is consistently highest for lower respiratory tract samples (Figure 2b). Detection probability decreases rapidly at rates dependent on sample type, and most rapidly for upper respiratory tract samples, but the proportion of positive samples approaches zero around 30 days post-symptom onset for each sample type. Raw RNA detection probability data are provided in Figure 2—source datas 3–5. Antibody level kinetics Antibody kinetics were analyzed by fitting a Gompertz growth rate function using Bayesian MCMC. While all subsets of the data were fit well by this model, we found some differences in antibody level kinetics depending on antibody, assay and antigen (Figure 3A). Full model fitting results for each assay can be found in Appendix 1. Figure 3 Download asset Open asset IgG and IgM antibody level kinetics for ELISA NP and Spike assays (A) and disease severity for IgM (B) and IgG (C). Measured using ELISA Spike and ELISA NP, respectively. Fitted functions use the posterior mean values for increase rate and start of the increase phase (displacement). Dotted lines show upper and lower 95% credible intervals. Note that the upper CI of IgM ELISA severe overlaps with the lower CI of mild cases, as do the upper CIs of IgG ELISA mild and severe. In order to allow the comparison of increase rate patterns, normalized peak antibody levels were set to one for all functions. Figure 3—source data 1 Peak antibody level time posterior means and 95% credible intervals (CrI). https://cdn.elifesciences.org/articles/60122/elife-60122-fig3-data1-v2.docx Download elife-60122-fig3-data1-v2.docx Figure 3—source data 2 Peak antibody level time pairwise posterior differences. Posterior differences between means were calculated by subtracting the posterior mean value for the antibody/assay in the second column from that of the first column, for each MCMC iteration. Differences were considered significant when zero was not included in the 95% credible interval (indicated in bold font). https://cdn.elifesciences.org/articles/60122/elife-60122-fig3-data2-v2.docx Download elife-60122-fig3-data2-v2.docx Figure 3—source data 3 Growth rate posterior means and 95% credible intervals (CrI). https://cdn.elifesciences.org/articles/60122/elife-60122-fig3-data3-v2.docx Download elife-60122-fig3-data3-v2.docx Figure 3—source data 4 Growth rate pairwise posterior differences. Posterior differences between means were calculated by subtracting the posterior mean value for the antibody/assay in the second column from that of the first column, for each MCMC iteration. Differences were considered significant when zero was not included in the 95% credible interval (indicated in bold font). https://cdn.elifesciences.org/articles/60122/elife-60122-fig3-data4-v2.docx Download elife-60122-fig3-data4-v2.docx Peak antibody level is reached around days 14–20 post-symptom onset, and the timing depends on antigen: both IgG and IgM peak levels are reached earlier when measured using ELISA NP than when using ELISA Spike (ELISA NP mean = 14.3 days, 95% CrI 12.0–16.1; ELISA Spike mean = 20.0 days, 95% CrI 17.6–22.4; 95% CrI for the difference = 2.7 to 9.2). The peak timing does not differ significantly between IgG and IgM when both are measured using ELISA Spike (IgG mean = 20.4 days, 95% CrI 16.8–24.1; IgM mean = 19.1 days, 95% CrI 15.6–22.4; 95% CrI for the difference = −6.4 to 3.5), nor when using ELISA NP (IgG mean = 15.2 days, 95% CrI 12.8–17.2, IgM mean = 12.2, 95% CrI 7.8–16.2; 95% CrI for the difference = −1.8 to 7.8). All estimates and pairwise statistics, including those for antibody levels measured using MCLIA, are shown in Figure 3—source datas 1–2. Antibody growth rates measured using ELISA NP tend to be higher than those measured using ELISA Spike (ELISA NP mean = 0.55/day, 95% CrI 0.48–0.64; ELISA Spike mean = 0.39/day, 95% CrI 0.34–0.44; 95% CrI for the difference = 0.07 to 0.26). The rate of increase for IgM does not differ significantly from that of IgG when both are measured using ELISA Spike (IgG mean = 0.39/day, 95%CrI 0.32–0.46; IgM mean = 0.41/day, 95% CrI 0.34–0.49; 95% CrI for the difference = −0.08 to 0.13), nor when measured ELISA NP (IgG mean = 0.53/day, 95%CrI 0.45–0.61; IgM mean = 0.68/day, 95% CrI 0.42–1.03; 95% CrI for the difference = −0.50 to 0.14). All estimates and pairwise statistics, including those for antibody levels measured using MCLIA, are shown in Figure 3—source datas 3–4. Disease severity does not significantly affect the time at which peak levels are reached for IgG (Figure 3B; mild mean = 14.0 days, 95% CrI 10.8–17.1; severe mean = 15.9 days, 95% CrI 10.7 to 20.6; 95% CrI for the difference = −8.0 to 4.2). However for IgM, peak antibody levels are reached approximately 7.0 days earlier for mild cases than severe cases (Figure 3C; mild mean = 15.6 days, 95% CrI 12.8–19.0; severe mean = 22.7 days, 95% CrI 18.5–26.6; 95% CrI for the difference = −12.2 to −1.8). Corresponding patterns are observed for antibody growth rate, which does not differ between mild and severe cases for IgG (mild mean = 0.58/day, 95% CrI 0.45–0.72; severe mean = 0.51/day, 95% CrI 0.36–0.69; 95% CrI for the difference = −0.16 to 0.28), but does for IgM, with levels increasing more rapidly for mild cases (mild mean = 0.51/day, 95% CrI 0.42–0.60; severe mean = 0.34/day, 95% CrI 0.28–0.42; 95% CrI for the difference = 0.05 to 0.28). Discussion By leveraging and integrating multiple data sources on key aspects of the antibody response against SARS-CoV-2, we were able to produce quantitative estimates of the mean and variation of seroconversion timing, antibody level kinetics, and the changes in antibody and RNA detection probabilities. These results provide critical reference information for serological surveys, assay sensitivity and risk of false-negative results, transmission models and herd immunity assessments. By combining data from 21 different studies using different assays, antigens, protocols and patient groups, we were able to quantify the means and, crucially, the extent of variation of important serologic and RNA detection parameters. Together, these antibody and RNA detection probability patterns provide an essential evidence base for informing sampling designs (Table 1). Figure 4 provides an overview of the key patterns. Table 1 Examples of how improved knowledge of antibody and RNA detection patterns can inform sampling designs. QuestionWhat to test forOptimal timing to testCommentsImportanceHas an individual been exposed in the past?IgG25-60(+) days post symptom onsetIgG persistence: possibly 1–2 years based on other human coronaviruses (Chang et al., 2005).Transmission models (Weitz et al., 2020; Kucharski et al., 2020) Herd immunity (Lassaunière et al., 2020; Theel et al., 2020).Is an individual currently infected?Viral RNA<30 days post-symptom onsetPreferable: sequential tests because of extensive variation in detection (Wölfel et al., 2020). Detection probability highest for lower respiratory tract or fecal samples, but upper respiratory tract samples are necessary to assess transmission potential.Assess transmission risk to others; contact tracing Giordano et al., 2020; Parameterization of transmission models (Weitz et al., 2020; Kucharski et al., 2020).How recently was an individual exposed?IgM, IgG>25 days post-symptom onsetIgG indicates exposure, which is more likely to be recent if IgM is also present, and longer ago if IgM is absent.Recent exposure is more likely correlated with transmission risk, and is a useful measure for prioritizing contact tracing, notably for asymptomatic cases (Okba et al., 2020). Figure 4 Download asset Open asset Antibody and RNA detection patterns during the early phase of SARS-CoV-2 infection. (Top) Fitted splines for the detection probabilities of serum IgG and IgM (measured using any assay/antigen), and of RNA in upper respiratory tract samples. (Middle) Modeled IgG and IgM level kinetics with 95% credible intervals, with normalized peak antibody levels set at one to allow direct comparison of growth rates. (Bottom) Estimated distribution of observed IgG and IgM seroconversion times. Seroconversion time is highly variable between individuals, with a mean around 12–13 days post-symptom onset. We find that IgG and IgM can be detected as early as 0 days post-symptom onset in 10–20% of patients, which indicates that seroconversion can happen at, and likely before, the onset of detectable symptoms. To our knowledge, seroconversion prior to symptom onset has not been reported, which is likely due to the fact that such cases are typically not under investigation using serologic assays. By integrating a wide range of data sources, we detect greater variation in seroconversion timing than previously observed, and importantly, it was possible to quantify the distributions around the mean seroconversion times (Huang et al., 2020; Zhao et al., 2020; Haveri et al., 2020). Patterns of IgM and IgG detection align with immunological expectations, as IgM antibodies are typically present during the early phase of the immune response, while IgG antibodies remain detectable for much longer periods (Xiao et al., 2020). We detected IgG and IgM antibodies in nearly all (98–100%) individuals by days 22–23 post-symptom onset, consistent with recent findings (Kraay et al., 2020). While IgG detection remains at this level for at least the range of times in the dataset (60 days for ELISA-Spike), the proportion of IgM-positive samples decreases after roughly 28 days post symptom onset, reaching around 65% by day 60. In other words, a growing proportion of individuals loses detectable IgM from day 30 onwards. We also detect a robust effect of viral antigen, where responses against NP rise faster than those against Spike, for both IgM and IgG. The quantification of changes in detection probability over time is relevant for clinical testing and assay choice and will determine test sensitivity (Sethuraman et al., 2020). It has been postulated that disease severity and humoral immunity against SARS-CoV-2 are correlated, but results so far have been inconclusive (Okba et al., 2020). Here, we did not detect any significant effects of disease severity on antibody patterns, with the single exception that we estimated a lower rate of IgM increase in severe/critical cases relative to mild/moderate cases. Regarding seroconversion times, an earlier study analyzed 28 cases to find that IgM seroconversion times appeared to be the same for severe and non-severe cases, but their analysis of 45 cases showed that IgG seroconversion was earlier for severe cases (Tan et al., 2020). Similarly, earlier seroconversion in severe cases has been observed for SARS-CoV-1 (Lee et al., 2006), but this result was not consistent across studies (Chan et al., 2005). Our findings do not support the idea that severe cases seroconvert faster. Indeed, the only significant effect of severity in our analyses is that the inferred growth rate of IgM levels is slower for severe/critical cases. It is not clear whether this reflects a relevant biological difference, considering that all other parameters do not differ among disease severity categories. The consensus patterns from our meta-analysis suggest that any interaction between disease severity and antibody response must be subtle and sensitive to other sources of variation, explaining the inconsistencies seen across studies. Note that the IgG seroconversion histogram for severe/critical cases (Figure 1d) appears bimodal, with fewer datapoints between 13 and 18 days post-symptom onset. This could either be a consequence of an underrepresentation of these times in the different studies or a signal of a true underlying pattern, but unfortunately the data to distinguish between these two hypotheses are not currently available. Given the finding that disease severity does not have major effects on early-phase antibody patterns, and assuming no cryptic relationship between severity and the factors governing protective immunity, then mild cases could be substantial contributors to the development of herd immunity development. This finding may also be important for vaccine efficacy; however, it is not yet known whether the presence of IgG or IgM correlates with protective immunity (Altmann et al., 2020), although we do observe a similar pattern for neutralizing antibody detection (Figure 2a). The extensive individual variation in antibody patterns, which is a common phenomenon across many viral infections (Pacis et al., 2014), may affect the accuracy of transmission models (Weitz et al., 2020). For example, if seroconversion times reflect the actual end of infectiousness and onset of immunity (i.e. the transition from Infectious to Removed in SEIR-type models Li et al., 2020), the observed range of 0 to 40 days post-symptom onset may need to be represented in the infectious period distribution. It is important to carefully consider how this variation may affect model conclusions, and whether it should be taken into account explicitly (Wearing et al., 2005), especially given the heavy reliance of policy-makers on COVID-19 transmission models (Kissler et al., 2020). We observed clear patterns of RNA detection that have several important implications, particularly for sampling designs. First, it is clear that the probability of detecting RNA is highly dependent on sample type, consistent with previous observations (Tan et al., 2020; Memish et al., 2014). Lower respiratory tract samples have the highest probability of testing positive for SARS-CoV-2 RNA, particularly after about 15 days post-symptom onset. During the first 8 days, 100% of lower respiratory tract samples tested positive for RNA. While detection probabilities for fecal and upper respiratory tract samples are nearly this high at symptom onset, they decrease much more rapidly, with the lowest average detection probabilities for upper respiratory samples. Nevertheless, it appears that by 30 days post-symptom onset detection probability approaches zero for all sample types, although it is important to note that the dataset did not include lower respiratory samples beyond day 29, which means that the true detection endpoint in lower respiratory samples could not be determined. These results match those from multiple studies (Tan et al., 2020; To et al., 2020; Sethuraman et al., 2020; Guo et al., 2020). When interpreting results on RNA detection, it is important to note that the presence of RNA does not necessarily imply the presence of live virus (Theel et al., 2020; Wölfel et al., 2020). One potential caveat for any analysis of data reported as time since symptom onset is that variation in the incubation period (time between infection and symptom onset) can affect the estimated timing of antibody kinetics and RNA detection. The mean incubation period is estimated to be around 7–8 days, with a standard deviation of 4.4 (Ma et al., 2020). The clear antibody and RNA detection patterns we observe here suggest that the effect of this variation does not obscure broad patterns, but relative results may be affected if the incubation period differs between certain groups of individuals. This could indeed be the case for disease severity, as mild cases are estimated to have a longer incubation period (8.3 days) than severe cases (6.5 days) (Ma et al., 2020). In summary, this study provides an up-to-date, comprehensive reference of key antibody and RNA detection parameters, including estimates of variation that can be used to inform serological surveys and transmission models (Table 1). As more data on SARS-CoV-2 become available, parameters can be updated through the use of the algorithms made available in the accompanying R code. Materials and methods Article selection Request a detailed protocol We considered