Since the SARS-CoV-2 virus emerged in late 2019, multiple pandemic waves have resulted in widespread social and economic disruption and > 6.5 million confirmed deaths worldwide.¹ A large excess-mortality study estimated that the true worldwide death toll attributable to COVID-19 is 2.9 to 3.3 times higher than the number of confirmed deaths.² Although death is the most impactful outcome of the pandemic, estimating the total number of SARS-CoV-2 infections also has public health utility. Throughout the pandemic, studies of the incidence of SARS-CoV-2 infection have informed public health mitigation measures, vaccination policy and distribution goals, and forecasts of future pandemic states.^3–7 Studies of the incidence of SARS-CoV-2 and its variants have ongoing importance for estimating levels and patterns of hybrid immunity in the population, informing the response to future pandemics, and providing a historical record of the impact of the COVID-19 pandemic.

Study designs for determining SARS-CoV-2 incidence vary, but all must address limitations of the diagnostic methods used to identify infections. Cross-sectional serosurveys have been used to identify the proportion of the population who have SARS-CoV-2 antibodies indicative of past infection.^8–12 Before vaccines were available, infection could be inferred by the presence of antibodies targeting the spike or nucleocapsid proteins; however, vaccine-and infection-induced spike antibodies cannot be distinguished. As a result, seroprevalence studies carried out in vaccinated populations have typically used nucleocapsid antibody to identify infection. These studies may fail to identify prior infections, because of antibody waning, and their utility for detecting reinfection is unclear.^13,14 Furthermore, serosurveys do not provide information on the timing of infection and can be subject to issues of selection bias and generalizability, depending on the sampling scheme.

Other incidence studies have focused on SARS-CoV-2 cases identified during acute infection by reverse-transcriptase–polymerase chain reaction (RT-PCR) or rapid antigen tests performed by health care providers or public health professionals, which are required to be reported to health departments in the United States. Because the availability and likelihood of testing have changed throughout the pandemic, these studies must account for time-varying under-ascertainment, often by incorporating seroprevalence data.^15–17 However, these studies typically do not follow individuals longitudinally, which limits their ability to estimate the incidence of re-infection that increased substantially after the emergence of the Omicron variant.^18,19

In this study, we followed a cohort with broad age representation in rural, central Wisconsin with the aims of estimating the incidence and seroprevalence of SARS-CoV-2 infection. We also describe the serologic response to infection and vaccination, and the persistence of antibodies after SARS-CoV-2 infection.

A total of 1520 people were enrolled in the cohort, which included broad representation across all age groups; 57% participants were female, 46% reported ≥ 1 chronic health condition, and 9% received ≥ 1 COVID-19 vaccine doses prior to enrollment (Table 1). By the end of follow-up, 25% of the cohort remained unvaccinated against COVID-19, whereas 49% had received ≥ 3 vaccine doses; 98% of all vaccine doses received in the cohort were mRNA vaccines. Older age, female sex, non-Hispanic White race, and presence of chronic health conditions were all significantly associated with increased vaccination (Table 1). Of the 1520 participants, 361 withdrew from the study before the end of the analysis period on July 31, 2022. The majority of loss to follow-up occurred at the end participants’ initial 1 year enrollment (November 2021-March 2022; Figure S3). Those lost to follow-up were more likely to be 18–29 years of age and to be unvaccinated and less likely to be non-Hispanic White and have ≥ 1 chronic conditions (Table S2). However, on average, characteristics of participants who participated for the entire period remained comparable to all enrolled participants with the exception that they were somewhat less likely to be unvaccinated (21% vs 25%; P = .04).

Overall, 992 of 1520 individuals (65%) had 1196 SARS-CoV-2 infections identified by study RT-PCR, EHR documentation, self-report, or serology (Table 2). Daily age- and vaccine dose–specific weighting to the MESA source population corresponded to an estimated 40 516 first infections among the 53 079 MESA residents (76%) and an incidence of approximately 12 first infections per 10 000 person-days from local start of the pandemic, March 1, 2020-July 31, 2022. Incidence of first infection was lowest among adults ≥60 years of age (7 per 10 000 person-days), and ranged from 10 to 13 per 10 000 person-days in younger age groups. Incidence also varied by period of variant circulation (Table 2). Incidence of first infection was 6.5 (95% CI, 5.9–7.1), 9.1 (95% CI, 7.8–10.7), and 34.2 (95% CI, 31.0–37.7) per 10 000 person-days in local periods of pre-Delta, Delta, and Omicron circulation, respectively. In all periods, incidence of re-infection was substantially lower than that of first infection. However, incidence of second infections during the Omicron period was higher than the incidence of first infection in any prior period, and third infections were only observed during the Omicron period.

A total of 6602 serum specimens were collected across 5 scheduled visits (average: 4.3 specimens per participant; SD 1.0). The timing of the collection of these specimens relative to infection and vaccination in the cohort are presented in Figure 1A. Seroprevalence of SARS-CoV-2 antibodies binding either the spike or nucleocapsid protein increased from 6% (95% CI, 3–9) at baseline in November 2020 to > 94% in all months from February 2022 onward. The time-series trajectory of seroprevalence was consistent with the cumulative incidence of infection and vaccination in the cohort. Throughout the follow-up period, more individuals were seropositive for spike-binding antibodies than were seropositive for nucleocapsid-binding antibody (Figure 1B). As expected, the trajectory of nucleocapsid antibody seroprevalence more closely followed SARS-CoV-2 cumulative incidence than spike seroprevalence. Overall seroprevalence of spike or nucleocapsid antibody was low, with little variation by age prior to the availability of COVID-19 vaccines (Figure 1C). Increases in overall seroprevalence over time were consistent with age-specific vaccine availability and uptake. At the end of follow-up in July 2022, all age groups had seroprevalence > 95% except for children < 5 years (60%) for whom vaccines had only recently been authorized. Among all seropositive participants in summer 2022, 6% were unvaccinated without reported or documented infection, 12% were unvaccinated with history of infection, 41% were vaccinated with history of infection, and 41% were vaccinated without history of infection (Figure 1D).

To assess the performance of our seropositivity and OD change thresholds for accurately identifying infections, we estimated the proportion of individuals with serologic evidence of infection after a study or EHR-documented or self-reported positive SARS-CoV-2 test (Table 3). In each variant period, individuals who were spike and nucleocapsid seronegative prior to infection became spike seropositive after infection more than 95% of the time. However, further spike ELISA OD increases were only observed among 5% of individuals infected despite spike seropositivity. Postinfection nucleocapsid seropositivity among individuals who were spike and nucleocapsid seronegative prior to infection ranged from 66% in the pre-Delta period to 89% in the Delta period. In contrast to spike, further nucleocapsid ELISA OD increases were common among individuals infected despite nucleocapsid seropositivity (70% overall among those seropositive for nucleocapsid only or both nucleocapsid and spike).

We examined the persistence of SARS-CoV-2 antibodies to determine the extent to which antibody waning might have affected our seroprevalence estimates. We found that spike antibodies declined relatively slowly after infection (rate of ELISA OD change per 100 days: −0.04; 95% CI, −0.02 to −0.05), and no participants were observed to revert to seronegative status after being spike seropositive (Figure 2A). In contrast, nucleocapsid antibodies exhibited notable waning in the first year after infection (Figure 2B); only 58% of infected individuals remained seropositive after 300 days (Figure 2C). The decline in ELISA OD per 100 days was significantly slower (−0.10; 95% CI, −0.09 to −0.11) when using all follow-up data compared with the primary analysis restricting to ≤ 350 days of observation (−0.17; 95% CI, −0.15 to −0.19; Figure S8). Consistent with this, waning appeared to slow after about 1 year following infection; 52% of infected individuals remained seropositive at 600 days. The rate of spike antibody waning did not change when using full follow-up data vs restricting to ≤ 350 days of observation (Figure S8).

This study provides unique insights into the epidemiology of SARS-CoV-2 in a rural population during a period of rapid virus evolution. We observed a high cumulative incidence over 2 years, with a markedly higher incidence during the period of Omicron variant circulation compared with the Delta and pre-Delta periods. COVID-19 vaccine uptake was also relatively high. As a result of this high incidence of infection and vaccination, 98% of the cohort was seropositive to either SARS-CoV-2 spike or nucleocapsid protein by July 2022. Although seropositivity was high, it is important to note that we measured antibodies that bind SARS-CoV-2 proteins in ELISA assays. Levels of binding antibodies may not correlate with protection from infection as well as neutralizing antibody, particularly for SARS-CoV-2 variants of concern.²⁵

Our estimates of nucleocapsid seroprevalence (62% in February 2022; Figure 1B) are similar to those from a national seroprevalence study estimating that 67% of Wisconsin residents had antibody to the SARS-CoV-2 nucleocapsid protein, indicating prior infection, by February 2022.^11,26 However, our results also highlight limitations of studies estimating incidence based on seroprevalence alone. Reinfection was common in our study; 19% of individuals with ≥ 1 infection had a second infection, and 7% of reinfected individuals had a third infection. Our estimates of reinfection may even be underestimated because serology was only used to define primary infection in incidence analyses. Seroprevalence studies using a simple seropositivity endpoint will likely miss most reinfections.

The results of our analysis of serologic response to infection suggest that increases in the SARS-CoV-2 nucleocapsid, but not spike, protein ELISA OD value can be used to identify reinfection in seroprevalence studies with longitudinal sampling. We also found that about 20% of previously seronegative individuals with SARS-CoV-2 infection did not become nucleocapsid seropositive after infection (or revert to seronegative status prior to collection of their postinfection specimen). Among those who became seropositive, approximately 40% reverted to seronegative status in less than 1 year. This is consistent with findings of previous studies demonstrating waning of antibodies to the nucleocapsid protein but relatively stable spike protein antibody levels over time.¹³ Methods to correct for assay sensitivity and antibody waning should be used to improve estimates of SARS-CoV-2 incidence.^16,27

The first SARS-CoV-2 infections in Wisconsin were identified in March 2020, but cohort enrollment did not begin until November 2020. As a result, we may have underestimated the incidence of infections that occurred early in the pandemic. However, our findings that nearly all individuals with SARS-CoV-2 infection became spike-protein seropositive, and that this antibody response was long-lasting, suggest that the magnitude of this underestimation likely is small. The cohort also experienced loss to follow-up, particularly at the end of the participants’ initial 1 year enrollment, which generally coincided with the early period of Omicron circulation. Those lost to follow-up up were more likely to be unvaccinated than those who were not. As a result we may have underestimated Omicron incidence and overestimated overall seroprevalence. We did not ascertain asymptomatic infection in all participants and did not distinguish between symptomatic asymptomatic infections in our analyses. If asymptomatic participants are less likely to have an antibody response after infection, our incidence and seroprevalence estimates may have been underestimated.

Greater than 70% of PACC participants completed primary series vaccination, whereas the counties that include MESA zip codes ranged in primary series completion from 35% to 58%.²⁸ Consistent with this, we estimated that 11% more MESA residents had ≥ 1 SARS-CoV-2 infection during the study period compared with PACC participants (76% vs 65%). The differences in incidence estimates between PACC and MESA were largest during periods of Delta and Omicron circulation after COVID-19 vaccines became available. This finding highlights the importance of accounting for differences in vaccine uptake when extrapolating estimates of SARS-CoV-2 incidence beyond the population directly under study. These results also highlight heterogeneity in vaccine uptake in rural settings. Additional research is needed to identify those populations least likely to be vaccinated and strategies for increasing vaccine uptake. For example, local data indicate that children living on farms are less likely to be vaccinated than other children living in the same communities.²⁹

Increases in reinfection observed in this and other studies after the emergence of the Omicron variant suggest that antigenic drift will play a prominent role in ongoing SARS-CoV-2 transmission.³⁰ However, infections among those previously exposed, through vaccination or prior infection, are less likely to result in severe outcomes.^31,32 This changing epidemiology will necessitate ongoing epidemiologic and virologic surveillance and continued revision of vaccination policy and recommendations. Prospective longitudinal studies, such as PACC, are essential to informing these activities. As SARS-CoV-2 transitions to endemicity, immunity is still uneven, and some populations may continue to have increased burden of infection and disease for the foreseeable future. In the United States, vaccine uptake has remained low among young children, and there is considerable variation in vaccination coverage among adults, due to vaccine hesitancy and issues of access.^33,34 Globally, issues of vaccine equity have contributed to a substantial number of preventable deaths, and more than one-third of the global population is SARS-CoV-2 seronegative.^35,36 Solving these issues in the context of a transition to endemicity will require sustained public health and research efforts.

Supplementary material is available at the American Journal of Epidemiology online.