Introduction
Background
The COVID-19 pandemic, caused by SARS-CoV-2, remains a significant global health concern. Antibodies play a crucial role in immune protection against viral infections by neutralising the virus and providing immunity. However, SARS-CoV-2 antibody levels wane over time, as it is inefficient for the immune system to maintain a constant, high-level defence against a specific virus. Characterising the timing of SARS-CoV-2 antibody waning is important to better understand and predict the duration of immunity against future infections.
Prior work reveals a decline in COVID-19 antibodies over time among those previously infected with SARS-CoV-2. For instance, a study conducted by Guo et al in China revealed that neutralising antibodies against SARS-CoV-2 were still present at 12 months but with a waning pattern after the initial infection.1 In Denmark, Hansen et al demonstrated a decline in IgM and IgG levels over the 6 months between the baseline and the follow-up.2 Their study also found that individuals who were infected before receiving vaccines had higher antibody levels compared with those who were vaccinated without any prior infection. Additionally, Chansaenroj et al estimated that the half-life of neutralising antibodies, the time for antibody level to decrease to half of its peak value, was approximately 100.7 days.3 This study also found that the decline of serum antibody levels over time depended on symptom severity, and individuals with higher IgG antibody titres experienced a significantly longer persistence of SARS-CoV-2-specific antibody responses. These findings indicate that individuals can experience a strong initial immune response to SARS-CoV-2; however, antibody levels naturally decline over time, influenced by various factors, including vaccine status and the severity of symptoms experienced.
Although antibody trajectories in adults have been characterised, there is still a significant lack of data on young individuals, particularly college students. Only one study has characterised longitudinal antibody levels in college students and found a decline in IgG antibodies after natural infection to undetectable levels within 5 months.4 Many studies primarily report SARS-CoV-2 antibody seroprevalence (at a single time point), not dynamics. For example, a study involving 344 college students in a rural Southern state found a seroprevalence of 18.2% in September 2020, which decreased to 13.1% 3 months later.5 Seroprevalence studies have also been used to measure transmission, as in a study at Pennsylvania State University before and after the Fall 2020 term, which found a significant increase in SARS-CoV-2 seroprevalence among students.6
Understanding transmission dynamics among college students is important, as college campuses often feature densely populated residential settings, frequent social gatherings and in-person group activities. This unique environment creates a high risk for transmission and outbreaks among young adults, especially those aged 18–24.7–10 This demographic may seed infections within surrounding communities. While young adults generally experienced mild severity of illness with SARS-CoV-2 infection, transmission in this group may have serious implications for more vulnerable populations, such as the elderly.11 Effective and targeted pandemic interventions among college students require a thorough understanding of postinfection protection. Central to this is the need to characterise antibody dynamics over extended durations.
Objectives
This study aims to investigate the dynamics of SARS-CoV-2 seropositivity among college students from March to November 2020. We focus on the persistence of SARS-CoV-2 antibodies in students with a prior positive reverse transcription-polymerase chain reaction (RT-PCR) test and further evaluate how the presence of symptoms influenced the persistence of antibodies. We used the Strengthening the Reporting of Observational Studies in Epidemiology guidelines for cohort, case-control and cross-sectional studies (combined) checklist.
Materials and methods
Study design and settings
The current study used a mixed retrospective and prospective longitudinal design. It is part of the Indiana University (IU) COVID-19 Antibody Study, a randomised trial conducted over 10 weeks, from 8 September to 18 November 2020. This research initiative, identified as Protocol #2008293852, received ethical approval from the IU Human Subjects and Institutional Review Board.12
A total of 1397 IU—Bloomington undergraduate students, aged 18 or older and residing in Monroe County, Indiana, participated in the parent study. Participation required written informed consent and completion of a baseline survey. The baseline survey was administered online using the research electronic data capture (REDCap) tools hosted at IU.13 14 Eligible participants scheduled their baseline antibody testing visit between 8 and 30 September 2020 and completed their endline antibody test appointment during the week of 8–14 November 2020 (see parent study design flowchart in online supplemental S1 figure). The baseline survey collected data on demographics, social behaviours, risk and protective behaviours and history of SARS-CoV-2 tests (RT-PCR and antibody). Follow-up data were collected biweekly after the baseline antibody tests and included updates on SARS-CoV-2 testing, antibody test results received outside the study and risk behaviours.
At the antibody test visits, clinical staff used a fingerstick to collect a small blood sample from each participant and applied it to the Beijing Genomics Institute (BGI) SARS-CoV-2 IgM/IgG rapid assay kit, which enables simultaneous detection of IgM and IgG antibodies using colloidal gold lateral flow. Trained field staff read the antibody test results, took a high-quality photo of the test kit with the study ID, uploaded it to a secured cloud drive and entered the test results into the REDCap system. Full details of the data collection procedures have been previously published.15 Quality control measures were implemented to ensure accurate antibody test readings. A trained research assistant independently interpreted the results using the photos taken by field staff. Any discordant results were adjudicated by five members of the research team.
Validation of the antibody test kit was conducted at the IU Health Pathology Laboratory in Indianapolis to assess measurement error. The Beckman Coulter Coronavirus SARS-CoV-2 IgG Ab with chemiluminescent enzyme immunoassay test served as the gold standard. For validation, 100 blood samples from patients who had tested positive using this test method during routine medical care were selected as positive cases. A negative control group was included, consisting of 100 age-matched samples from patients before the pandemic. Frozen blood samples from both groups were thawed and tested using both the Beckman Coulter test and the BGI SARS-CoV-2 IgM/IgG rapid assay. In this validation study, the BGI test rapid kits demonstrated a sensitivity of 66% and a specificity of 100% (online supplemental S1 table). The calculated positive and negative predictive values (PPV and NPV) were 100% and 74.6%, respectively. This suggests a potential misclassification rate of 25.4%, where truly positive participants may have been classified as negative in our study.
Patient and public involvement
Patient and public involvement was not sought in the design, conduct, reporting or dissemination plans of our research. Likewise, participants were not consulted on the burden of participation or the format and timing of result dissemination. The parent study was a randomised controlled trial designed to evaluate how receiving SARS-CoV-2 antibody test results affected participants’ adherence to COVID-19 protective behaviours. The lack of in-person operations and the complexity of launching a study during a time when there were substantial restrictions to on-campus events made it especially difficult to engage patients or members of the public in this time-sensitive research. We acknowledge the value of public involvement and have presented study findings to the university community.
Study population
Duration since infection and seropositivity
To investigate the duration of SARS-CoV-2 seropositivity among college students, we analysed data from 179 antibody tests with corresponding RT-PCR-confirmed infections (figure 1). This included participants with infections reported prior to enrolment who completed the baseline antibody visit, as well as those who reported infections during follow-up and completed the endline visit.
Flowchart for study population to assess time since RT-PCR confirmed infection and seropositivity at baseline and endline visits. *This cut point for exclusion was compared to a less stringent cut point of 2 weeks in sensitivity analyses. RT-PCR, reverse transcription-PCR.
A subset of participants (n=61) contributed both a baseline and an endline antibody test, having had a prior infection and attended both visits (baseline and endline). The remaining participants contributed only one test: either a baseline test (n=10) or an endline test (n=47) (refer to the flow diagram figure 1). In total, 71 participants contributed a baseline antibody test, and 108 contributed an endline antibody test, all conducted at least 3 weeks after their RT-PCR diagnosis. Participants were identified as having a history of prior infection at baseline if they answered yes to these two survey questions: ‘Have you ever been tested for SARS-CoV-2 (COVID-19) before?’ and ‘Have you ever tested positive for a SARS-CoV-2 (COVID-19) infection?’ In the follow-up surveys, participants could contribute to the analysis at the endline if they responded yes to both of these questions: ‘Have you been tested for an active SARS-CoV-2 infection (nasal swab or saliva test) since you filled out the baseline survey?’ and ‘If yes, have you tested positive for an active SARS-CoV-2 infection since you filled out the baseline survey?’ Participants were asked to provide the date of their positive RT-PCR tests. The earliest date of a retrospectively reported positive RT-PCR test was 20 March 2020.
Dates were calculated in weeks from the date of the positive RT-PCR test to the date of the antibody test. Seven cases required imputation of the time since the RT-PCR test: three were missing dates and four had implausible dates, where the reported positive RT-PCR test date in the baseline survey was after the baseline visit. To address these discrepancies, we used a simple population median imputation method, replacing these values with the median time interval observed in the entire cohort.
Moreover, we excluded data from participants whose antibody tests were within 3 weeks of their RT-PCR diagnosis (n=35). Law et al demonstrated that antibody tests typically reach peak sensitivity at approximately 2.6 weeks (95% CI 2.1 to 3.0 weeks).16 A 3-week cut-off maximises specificity by allowing sufficient time for a detectable increase in antibody titres. Additionally, we conducted a sensitivity analysis excluding only participants who were tested within 2 weeks to evaluate the impact of the selected cut-off on our findings (n=8).
Persistence of seropositivity
To investigate the persistence of seropositivity from baseline to endline, we analysed data from participants who tested seropositive at both time points. Specifically, we assessed whether individuals who were seropositive at baseline remained seropositive at the endline visit. Of the 1076 participants who completed the baseline visit, 49 tested positive for antibodies. Among these, 42 completed the endline visit, while seven were lost to follow-up (refer to figure 2). Symptom data were available for 26 of the 42 participants.
Study flow diagram for persistence of seropositive by symptomatic status.
Variables and measures
Exposures
The time interval between RT-PCR test positivity and antibody test results was measured in weeks. While a few participants reported two RT-PCR tests, we included only the baseline RT-PCR date to focus on the primary immune response. Reinfections were rare during the study period and typically triggered faster, milder immune responses due to existing immune memory,17 18 which could bias estimates of antibody dynamics.
Symptom severity was self-reported and was collected as a 4-level categorical variable (none, mild, moderate and severe). For analysis, this was dichotomised into asymptomatic (none) and symptomatic (mild to severe).
Outcomes
Antibody positivity: Participants were considered seropositive if their antibody test indicated the presence of either IgM or IgG antibodies. Conversely, participants were categorised as seronegative if the test indicated the absence of both IgM and IgG antibodies. We assessed antibody seropositivity at both the baseline and endline visits.
Persistence of seropositivity: For participants who tested positive for antibodies at baseline, we examined their antibody positivity status at the endline visit. The term ‘persistence of positivity’ refers to participants who continued to test positive for antibodies at the endline visit.
Covariates
Baseline characteristics that could potentially impact SARS-CoV-2 antibody levels were included as covariates to standardise comparisons. Age has been associated with antibody waning; females often exhibit stronger antibody responses and racial/ethnic differences likely primarily reflect differences in social determinants of health.19–21 Stratified analyses were not feasible due to limited sample size, which would have led to small subgroup counts and unstable estimates. Thus, the covariates included in adjusted models of the relationship between duration since infection and seropositivity were age, sex at birth (female vs male) and race (initially, race/ethnicity was recorded as an 8-level categorical variable but was later simplified into white, Hispanic, Asian and other categories for analysis). Visit (baseline vs endline) was also included as a dichotomous variable.
Statistical methods
Duration since infection and seropositivity
To compare time intervals between students who tested seropositive and seronegative at both baseline and endline visits, we used the Wilcoxon rank-sum test to assess differences in distributions between groups, as the time interval variable was not normally distributed. To investigate the relationship between the time interval between RT-PCR and antibody tests and the likelihood of seropositivity at baseline and endline, we used generalised linear mixed models (GLMMs) with logit link function for binomial distributions. These models included random intercepts for each participant to account for the within-person correlation arising from repeated measures over time. We estimated both crude ORs corresponding to 95% CIs and adjusted ORs (aORs) by including covariates for age, sex assigned at birth, race/ethnicity and visit (baseline or endline). All models were fitted using the PROC GLIMMIX procedure in SAS.
Persistence of seropositivity
We explored the impact of symptomatic status on the persistence of seropositivity with 2×2 contingency tables. Among students who tested seropositive at baseline, we calculated relative risks (RRs) and corresponding 95% CIs to compare the likelihood of persistence (ie, remaining antibody positive at endline) between symptomatic and asymptomatic individuals. RRs were estimated using weighted frequency tables in SAS (PROC FREQ) with the RELRISK option, and statistical significance was assessed using the Pearson χ2 test. Fisher’s exact test was used when expected cell counts were sparse (≥20% of cells with expected counts <5). Given the limited number of participants with symptom data (n=26), we did not fit multivariable regression models, as these methods do not perform well with such sparse data.22–24
Additionally, to account for measurement error, we present both the original table for symptomatic status by antibody persistence and sensitivity analysis tables based on the NPV and PPV from the validation analysis. This reclassification involved categorising the likely false negatives within the symptomatic and asymptomatic subgroup, under the assumption that all baseline positives reflected the PPV and included no false positives.
SAS software (V.9.4) was employed for analyses,25 and the significant level was set at 0.05.
Results
Duration since infection and seropositivity
The study included 179 tests among 172 (12.3%) college students who self-reported a SARS-CoV-2 infection. The majority of participants were female (63.2%), and the average age was 20 years old (SD=3.7 years). The racial/ethnic distribution of participants was as follows: white (84.3%), Asian (4.6%), Hispanic/Latinx (6.4%) and other (4.6%) (table 1). As expected for the college population, 90% of the students were 18–21 years old.
Demographic characteristics for the IU sample population
Of the total 172, 126 (73.3 %) self-reported symptoms at baseline. Notably, 24.6% were asymptomatic, while only 2.4% experienced severe symptoms. 116 (82.6%) participants completed both antibody test visits, and 46 (39.7%) of them tested seropositive at either baseline or endline visits. Students who self-reported SARS-CoV-2 infections and attended either lab visit were majority female (60.0%), majority white (82.8%), and tended to report mild symptoms (52.4%).
In the primary analysis, excluding the antibody data within a 3-week post-RT-PCR timeframe, the median (IQR) time interval was 6.3 (4.0–9.9) weeks for baseline samples and 10.5 (8.6–14.1) weeks for endline samples (table 2 and online supplemental S2 table). Notably, although the Wilcoxon rank-sum test did not indicate a statistically significant difference between the two groups, the seronegative group exhibited a longer time interval, particularly at the baseline time point. These findings persisted in sensitivity analyses with 2-week time restrictions (online supplemental S2 table). The mean (SD) time interval for baseline and endline samples was 7.6 (4.9) weeks and 11.4 (5.3) weeks, respectively, with the seropositive group displaying a shorter time interval than the seronegative group. In sensitivity analyses, excluding tests conducted at least 2 weeks after the initial RT-PCR diagnosis resulted in a smaller difference of means compared with the primary analysis among baseline samples.
Statistics summary of time interval among students with infection history
Additionally, significant differences emerged between seropositive and seronegative students at baseline and endline (online supplemental S2 table). There was a higher proportion of seropositivity among females than males in the baseline sample (85.7% vs 50.7% seronegative), but no sex difference at endline (63.0% vs 59.1% seronegative). Both groups reported predominantly mild symptoms, yet a higher proportion of seropositive students reported moderate symptoms (20.7% at baseline) compared with seronegative students (16.4% at baseline). Additionally, a greater proportion of seronegative students reported no symptoms compared with their seropositive counterparts (26.9% vs 17.2%).
Furthermore, we observed no statistically significant association between the time interval and seropositivity (table 3). This pattern persisted in our sensitivity analysis, which employed a 2-week time restriction.
Mixed-effect logistic model results for time interval (weeks) and seropositivity
Persistence of seropositivity
In the current study, among 42 participants who were seropositive at baseline, only 15 (35.7%) remained seropositive at the endline visit, highlighting a decline in detectable SARS-CoV-2 antibodies over time. Among the 26 participants who reported symptom status, eight (30.8%) remained seropositive at the endline. We likely underestimated antibody positivity due to the limitations of the rapid antibody assay, as revealed by comparisons with the hospital-based assay. To assess the bounds of the impact of this measurement error, we presented extreme case corrections. Table 4 presented the results of χ2 analyses investigating the relationship between symptom status and the persistence of antibodies at the endline visit, based on both the observed and reclassified data.
Fisher’s exact test results for the persistence of endline antibody visit by symptom status
Based on the observed data (table 4), among 21 students who reported symptoms, 7 (33.3%) remained seropositive at the endline, compared with 1 (20%) out of 5 asymptomatic students. In the observed data, symptomatic individuals had a higher, though non-significant, risk of seropositivity at the endline (RR=1.67, 95% CI 0.26 to 10.65). After reclassifying all likely false negatives to the symptomatic group, the relative risk increased (RR=2.86, 95% CI 0.48 to 17.14), still without statistical significance (p=0.32). 12 (57.1%) of symptomatic students remained seropositive, while seropositivity among asymptomatic students remained at 20% (table 4). In contrast, reclassifying all likely false negatives to the asymptomatic group, the association reversed significantly (RR=0.33, 95% CI 0.18 to 0.61, p=0.01), showing that 7 (33.3%) symptomatic students remained seropositive vs 100% seropositivity among asymptomatic students (table 4).
Discussion
In this study, we examined the longitudinal patterns of SARS-CoV-2 seropositivity among college students over a 10-week period early in the COVID-19 pandemic (September–November 2020). Analysing data from two rounds of antibody testing, we found no statistically significant difference in time intervals between RT-PCR positive test and antibody tests among seropositive and seronegative groups; however, the seropositive group showed a shorter mean time interval than the seronegative group among baseline samples.
Our study findings suggest that SARS-CoV-2 antibody levels among college students may decline relatively rapidly, as only 35% of participants who initially tested positive for antibodies at the baseline remained positive at the endline. This observation aligns with previous research indicating a decrease in SARS-CoV-2 antibody levels over time, particularly among young individuals. Serum antibody concentration has shown a positive association with participant age, with individuals under 30 displaying the lowest antibody levels, highlighting potential variations in immune responses across age groups.19
In addition to these findings, other studies have also reported a swift reduction in antibody levels during the initial months following infection, emphasising the general trend of declining antibody levels in younger populations. These studies include the work of Seow et al, which reported a rapid decline in antibody levels within 3 months after infection, with a median half-life of 36 days in a cohort with a median age of 55 (range: 23 to 95 years); Lumley et al, which focused on healthcare workers with a median age of 39 (range: 17–69 years) and Zhao et al, which examined hospitalised patients with a median age of 48 years (IQR: 35–61).20 26 27 All of these studies documented a similar decline within the first few months postinfection in various populations.
However, some studies have reported a more prolonged persistence of SARS-CoV-2 antibodies, especially among individuals who experienced more severe cases of COVID-19. Notable examples include Di Chiara et al and Dan et al, which noted that antibody responses persisted for up to 8–12 months after infection in paediatric and specific adult populations.28 29 These varying observations highlight the complexity of antibody persistence and its potential differences across demographic groups. Our findings underscore the need for ongoing research on antibody levels, particularly among young adults.
Our study found that those with evidence of infection but seronegative status were more likely to report experiencing no symptoms compared with those with evidence of infection but seropositive status. These findings are consistent with previous research indicating seronegative individuals were more likely to have experienced an asymptomatic, as opposed to symptomatic, infection.30 31 However, it is important to acknowledge a potential study limitation. Specifically, there is a possibility that some seropositive students, particularly those testing positive for SARS-CoV-2 during the follow-up period, may have been presymptomatic at the time of testing or experienced asymptomatic infections that went undetected or unreported. This acknowledgement underscores the need for cautious interpretation of the findings and recognition of potential variability in symptom reporting among seropositive individuals, particularly during the follow-up phase. However, our findings are consistent with the understanding that antibody testing may not always provide a comprehensive picture of infection, particularly when individuals have had no or mild symptoms.
Given our small sample size, the observed connection between symptomatic status and antibody persistence may have been influenced by measurement error. While the observed data suggest a potential association but lack statistical significance, reclassified results illustrate the potential bounds under the extremes of reclassifying the outcome. A larger sample size would have enhanced the statistical power of this study, thus contributing to more robust and reliable conclusions.
Our study has several limitations that should be taken into account when interpreting the findings. First, there is a potential for recall bias in self-reported RT-PCR history. Participants who received a positive RT-PCR result may be more likely to accurately recall and report their testing history due to the significance of the event. While this may enhance data reliability within the included sample, differential reporting across subgroups could still introduce selection bias. For example, if females are more likely to report a prior positive test and also differ in their likelihood of seropositivity, this could lead to biased estimates and misrepresent the true patterns of antibody persistence across demographic groups. Additionally, variations in symptom severity, perception and testing experiences can impact the direction and magnitude of recall bias in both groups. Another limitation is the possibility of non-response bias. Some students may not have reported their RT-PCR results, may have been reinfected after the reported dates or may have provided incorrect information. This introduces the possibility of non-differential measurement error, which could occur in both groups. Recognising these limitations is crucial for accurate interpretation and highlights the need for cautious generalisation of the results. Furthermore, measurement error associated with the antibody kits used represents another potential limitation. However, this limitation was addressed through sensitivity analyses using the NPV and PPV against a Chemiluminescent Enzyme Immunoassay test, considered a gold standard due to its high sensitivity and specificity. It is important to emphasise that reclassification of false negatives significantly impacts the observed associations and reflects potential underestimation of the true prevalence of SARS-CoV-2 antibodies in our study population.
These findings do not provide insights into the persistence of immunity after vaccination, which may differ significantly from the natural immunity acquired through infection. The study specifically focused on antibody levels resulting from natural SARS-CoV-2 infection and did not include a time period during which SARS-CoV-2 vaccines were available to the general public.
Despite these limitations, our study provides valuable insights into the prevalence and timing of SARS-CoV-2 antibody waning among university students. By 12 weeks postinfection, most students (75%) who were seropositive at baseline were no longer seropositive. This highlights the need for further research with larger and more diverse populations, as well as the use of more sensitive antibody tests, to better understand the true seroprevalence of COVID-19 and its implications. Additionally, the magnitude and duration of the antibody response to SARS-CoV-2 may vary based on several factors, including symptom severity. Understanding how these factors influence the persistence of SARS-CoV-2 antibodies over time is critical for guiding public health strategies and informing future vaccination efforts.
