Introduction

College campuses feature environments that are high risk for SARS-CoV-2 transmission, characterized by indoor locations with close contact (e.g., dorms, classrooms) and academic and social mixing that results in high contact rates [1]. Although student populations are at lower risk for severe COVID-19 disease themselves, increased infections among younger populations are often associated with corresponding increases among older, higher-risk populations [2]. Understanding the motivations and drivers of practicing SARS-CoV-2 protective behaviors among university student populations has consequences for both campus infection prevention and control as well as the health of the broader community.

Many universities that pursued in-person education in 2020, and had the resources to do so, implemented SARS-CoV-2 preventive measures including mask mandates, isolation and quarantine procedures, reduced classroom density, online learning, regular testing regimes of active infections using RT-PCR and/or antigen tests, and vaccine mandates [3–5]. The main goals of regular testing were to control disease spread through quarantine of close contacts of individuals who tested positive and to track the prevalence of infection in the campus community. However, students may have understood a positive SARS-CoV-2 test to mean that they could relax preventive behaviors, assuming that a prior infection conferred immunologic protection. SARS-CoV-2 antibody tests, though less frequently used than tests of active infections, are another source of information about prior infection status that may influence risk perceptions.

Learning the results of a SARS-CoV-2 test may influence individual behavior through several plausible mechanisms. Theoretically, risk compensation postulates that individuals have some amount of risk that they are willing to assume and will change their behaviors to match that level of risk [6]. Relatedly, behavioral disinhibition theory posits that feelings of protection against one health concern may cause people to engage in behaviors that put them at risk for other health issues. A necessary condition for these behavioral pathways to operate with SARS-CoV-2 tests is that people must have working perceptions that a relationship between prior infection and protection against future infections exists. These perceptions are likely related to overall COVID-19 and immunological knowledge [7–9].

A number of public health interventions have been questioned for potentially causing risk compensation including wearing a bicycle helmet to prevent head injury [10, 11], seat belt mandates to prevent traffic fatalities [12], pre-exposure prophylaxis [13] and male circumcision [14] to prevent HIV infection, and HPV vaccination to prevent cervical cancer [15, 16]. However, rigorous randomized studies have shown that though an individual may make riskier choices, the interventions in question still have an overall beneficial effect on the population outcomes they are designed to prevent. Given the many influences on SARS-CoV-2 protective behaviors that may also correlate with seeking information about prior infection status, a randomized trial is the most appropriate design to avoid confounding by these factors. Understanding the impact of information about past SARS-CoV-2 infection on preventive behavior is essential to managing viral control and for learning more about expected behavior post natural infection and vaccination. The SARS-CoV-2 vaccine has an overwhelmingly beneficial effect on lowering risk of serious COVID-19 disease and mortality [17]. However, understanding how much, if any, risk compensation might occur after natural infection or vaccination has important potential consequences for disease control.

In this study, we aimed to answer the research question, ‘Does learning the results of an antibody test change SARS-CoV-2 protective behaviors?’, with a randomized control trial (RCT) in a sample of undergraduate students during the Fall 2020 semester. Aligned with our hypothesis that behavior change would be differential by antibody status (i.e., those with a positive antibody test may change their behavior in a way that would be different than those with a negative antibody test), we assessed results separately for antibody-positive and antibody-negative participants.

Methods

Results

Between September 14, 2020 and September 30, 2020, 1,397 participants (response rate of 34.4%) consented to participate in this study from a sample of 4,069 randomly selected, eligible IU-B undergraduate students. We report here the results from 1,076 (77% of those who consented) who completed the baseline survey and baseline antibody test (Fig 1 and Table 1). The median age of participants was 20 years (IQR 19–21) and the ages of study participants largely aligned with traditional undergraduate student ages of 18–21 (90.6%). The majority of study participants identified as women (64%). The study sample was also majority white (79%), with lower representation of Asian (8%), Black (1%), and multi-racial (8%) students. Participants were fairly evenly distributed across the four years of traditional class standing (freshmen through seniors). About one-third (32%) of participants lived on-campus and about one-quarter (24%) were affiliated with Greek student organizations. As expected, none of these demographic variables differed significantly by randomized delayed versus immediate treatment condition.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Consort participant flow diagram.

https://doi.org/10.1371/journal.pone.0279347.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Characteristics of the n = 1076 Indiana University undergraduate students enrolled in the trial, Fall 2020.

https://doi.org/10.1371/journal.pone.0279347.t001

Overall, 49 participants tested positive for SARS-CoV-2 antibodies at baseline (4.6%) and we observed 42 new seroconversions over the course of the 8 weeks of follow-up. The distribution of the four behavioral outcomes was generally balanced at baseline, with higher engagement reported for certain behaviors. Wearing a face mask (98.3% always or very often) and ensuring physical distancing (89.6% always or very often)] were more prevalent than staying home from work/school (43.2%) and avoiding social events (57.0%). The distribution of engagement in each behavior at baseline, 2-weeks (time of primary behavioral endpoints), 4-weeks, 6-weeks, and 8-weeks (endline) are displayed in Fig 2, stratified by randomized immediate vs. delayed treatment condition and by serostatus.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Protective behaviors by baseline sero-status in delayed and immediate randomized arms.

https://doi.org/10.1371/journal.pone.0279347.g002

In the overall study sample, we found no significant differences between treatment conditions in any of the four behavioral outcomes (Table 1). Two weeks after antibody test results were reported to participants in the immediate results condition, chi-square tests indicated that participants in this condition did not report significantly higher or lower engagement in wearing face masks, staying home from work and school, avoiding social events, or ensuring physical distancing in public. Similarly, no significant differences were observed between study conditions for the seroconversion outcome, with 22 delayed condition and 20 immediate condition participants experiencing seroconversion (6% vs. 5% respectively).

Regression models stratified by baseline serostatus did not reveal any significant behavioral differences by study condition (Table 2) for participants receiving either positive or negative antibody test results. Taking face mask use as a representative example, for seronegative participants, receiving antibody test results was not associated with higher or lower face mask engagement [RR (95% CI): 1.01 (1.00, 1.03)]. Similar results were observed for our smaller sample of seropositive participants [RR (95% CI): 0.91 (0.80, 1.04)]. We also did not observe significant differences in seroconversion risk by timing of antibody test results among those who were seronegative at baseline. Participants in the immediate condition did not exhibit a higher or lower risk of seroconversion at 8 weeks, compared to the delayed condition [RR (95% CI): 0.94 (0.52, 1.71)]. Assumptions for all modified Poisson models were assessed and met. In the sensitivity analysis using the summed Likert responses as the outcome, we also did not find any significant differences between protective behaviors at 2-week follow-up between those who received test results immediately compared to those randomized to the delayed arm (S1 Table and S2 Fig).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Associations between treatment condition (immediate vs delayed antibody test results) and key behavioral and biological outcomes, stratified by baseline antibody status.

https://doi.org/10.1371/journal.pone.0279347.t002

Discussion

In this study, we found no evidence for risk compensation among undergraduate students after receiving SARS-CoV-2 antibody test results. Students who were randomized to receive their antibody test results immediately did not report engaging in different levels of COVID-19 risk behaviors compared to students who were randomized to receive their antibody test results later. This lack of association held across all four behaviors we examined (staying home from school/work, avoiding social events, wearing face masks, and physical distancing), and for the SARS-CoV-2 seroconversion outcome. Importantly, stratified analysis by baseline antibody serostatus did not reveal differences in findings by whether participants received a positive or negative antibody test result.

That we found that behavior remained largely unchanged after receipt of antibody test results aligns with findings from prior studies of risk compensation in other settings. Five interventions are often recognized for their potential to cause risk compensation: bicycle helmets, seatbelts, voluntary medical male circumcision and pre-exposure prophylaxis for HIV prevention, and HPV vaccination. Exhaustive reviews in each of these areas have consistently concluded that there is, in fact, little evidence for increased risk-taking after intervention exposure [10–13, 16, 28]. In spite of the weak record for risk compensation in other settings, many have raised concerns about the threat of risk compensation in the COVID-19 pandemic [29]. Vaccines and face mask mandates have been two areas of concern for risk compensation, but very few empirical studies have tested these associations. However, there is at least some suggestive evidence that risk compensation may, in fact, occur in settings with face mask mandates [30]. With the lack of risk compensation observed in our study, we provide additional empirical data to better shape our understanding of behavioral disinhibition in the COVID-19 pandemic.

The context of the study setting and study design have important implications for interpretation of our findings. First, changes in risk perception among those who tested positive for antibodies may not have translated into changed behaviors due to COVID preventive measures taken at the university and community level. Indiana University and Monroe County, where it is located, both had mask mandates and bans on large group social gatherings during the study period. Second, in some social circles, social norms reinforced protective behaviors as being the socially responsible course of action. Those external restrictions on behavior may exert a larger influence than an internal assessment of risk. Third, study participants were provided clear guidance to continue to practice COVID-19 protective behaviors regardless of their antibody test results (see S1 Fig), which may also account for our findings of no observable behavior change. Relatedly, it is possible that risk behaviors did not differ between study arms because participants did not otherwise have strong perceptions of a relationship between serostatus and protection against future infections. Finally, as results were disseminated by email, we cannot confirm that everyone in each treatment condition received their results on the expected schedule, or at all. However, operational and survey data we collected indicated that participants received their test results at similar rates in both arms, alleviating concerns of differential intervention uptake (see Fig 1).

Results from this study likely do not generalize to university settings with different COVID-19 policies in place. However, the results are likely to be generalizable to other young adult populations at other large, public universities similar to the one where this study was conducted. We used random sampling from the IU student population and enrolled a large sample into the randomized trial. Though we observed a relatively low response rate, it is comparable to other response rates in university settings [31]. Nevertheless, to assume that our study population stands in for our larger target population, we have to assume that the enrolled study population had similar risk compensation responses as those who did not enroll. Undergraduate women, in particular, were overrepresented in our study population relative to the student body. Other demographic variables, however, generally tracked with those observed in the student body. After enrollment, we maintained a relatively high follow-up rate for the 2-week behavioral outcomes (>90%), but had more limited completeness of follow-up for the secondary seroconversion outcome at 8-weeks (79%). Reasons for the higher attrition at this later timepoint may have involved the higher burden and stricter COVID protocols with the in-person study visit required for the serology testing, and the timing at the end of the in-person semester coinciding with more limited availability of our student population.

Inference from this study is strengthened by the rigorous randomized study design used. By randomly assigning half of our participants to receive their results immediately, we were able to isolate the effect of receiving the results and minimize the potential for confounding to bias our results. Using a randomized trial design is critical because seeking testing for SARS-CoV-2, either for acute infection [32] or antibody testing, requires both personal motivation and access to services which render comparisons between those who receive testing and those who do not subject to significant possible bias. These factors, and those related to them, would be major sources of confounding in an observational study comparing those who did and did not receive antibody tests. However, although we had a large sample overall (n = 1076), only 47 participants tested positive for SARS-CoV-2 antibodies at baseline, limiting the precision with which we measured the trial effects in that subgroup. The relatively low number of positive tests we observed could be accurately capturing the low seropositivity at this early stage of the pandemic (September-November 2020). Negative antibody tests would also be expected for participants who have experienced a SARS-CoV-2 infection but not yet developed antibodies, or whose antibodies have already waned below detectable levels. Of course, some of our study participants may have falsely tested negative, due to the moderately low test sensitivity (64%) or researcher error in administering or reading the tests. In both of these scenarios, the measurement error is unlikely to be differential by trial arm as the study team was blinded to trial arm allocation.

Our findings provide reassuring evidence that the receipt of SARS-CoV-2 antibody test results are unlikely to cause major changes in protective behaviors, with the assumption that the results generalize beyond a university student population. Although individual risk perceptions may be updated with information on antibody positivity or antibody negativity status, those updated risk perceptions do not appear to result in risk compensation. These findings may also be informative to other experiences or interventions in the context of the COVID-19 pandemic such as natural infection, vaccination, and mask mandates. Future work on COVID-19 risk compensation should focus on extending our findings with these COVID-relevant exposures, in different populations, and by covering updated time periods in the pandemic.

Supporting information