Determinants of pre-vaccination antibody responses to SARS-CoV-2: a population-based longitudinal study (COVIDENCE UK)

The COVID-19 pandemic has caused more than 220 million recorded infections and over 4.5 million recorded deaths [1], with these figures representing only a portion of the true burden [2]. Large population-based studies have identified various risk factors for SARS-CoV-2 infection, including non-White ethnicity and lower educational attainment [3‐5]. However, the vast majority of studies have been based on routine real-time reverse transcription PCR (RT-PCR) testing in healthcare settings or in the community; consequently, they are potentially open to collider bias, as the probability of being tested for infection can itself depend on the risk factors under investigation [6]. Access to testing has also changed across the course of the pandemic [7], meaning earlier studies were more likely to focus on people with symptomatic disease or a history of travel, or on specific populations such as healthcare workers.

Serological population-based studies offer a different approach by testing members of a population uniformly, including people who might not be captured by routine testing. This approach not only reduces the risk of collider bias but also can uncover previously undetected asymptomatic infections. Inclusion of asymptomatic SARS-CoV-2 infections in the analysis of risk factors is crucial, as asymptomatic individuals have been found to be as infectious as those with symptoms [8]. Serology studies also offer the opportunity to identify determinants of anti-SARS-CoV-2 antibody titres, which are a recognised correlate of protection against future infection [9, 10].

The largest population-based serology studies done to date have explored several sociodemographic and clinical risk factors, but have not considered risk factors related to lifestyle, diet, or levels of physical activity [3‐5, 11, 12]. These studies have focused on IgG antibodies alone [11, 12] or relied on immunoassays with low sensitivity [12], potentially missing infections. They have also tended to be cross-sectional in design, so that reverse causality could potentially explain associations between symptomatic seropositivity and modifiable risk factors. Additionally, studies investigating determinants of antibody titres have focused on specific populations such as healthcare workers [13, 14], limiting the generalisability of their findings.

We therefore undertook a prospective population-based study to uncover determinants of SARS-CoV-2 seropositivity and antibody titres, combining high statistical power with detailed assessment of sociodemographic, clinical and behavioural risk factors, and supported by an assay with proven sensitivity for detection of SARS-CoV-2 antibodies in non-hospitalised adults with mild or moderate COVID-19 [15].

Serology data were available for 12,294 of the 15,853 participants who consented to participate in the antibody study. We excluded data from 1074 participants who had been vaccinated against SARS-CoV-2 before providing their dried blood spot sample (Fig. 1). Of the 11,220 participants included, 1774 (15.8%) tested positive for SARS-CoV-2 antibodies. For the analysis of determinants of seropositivity, we excluded 90 (0.8%) participants who reported a positive RT-PCR or lateral flow test result for SARS-CoV-2 infection before enrolment, leaving a sample size of 11,130 participants with 1696 seropositive cases (Fig. 1). Selected baseline characteristics of included participants are shown in Table 1. 70.1% of participants were female, and 95.7% identified their ethnicity as White, with median age of 62.3 years (IQR 52.9–68.7; Table 1).

After adjustment for age, sex and duration of follow-up, 25 factors were independently associated with risk of SARS-CoV-2 seropositivity with p < 0.10 (Table 2). Additional file 1: Table S3 shows factors with no evidence of association. When the former factors were included together in a fully adjusted model, we observed that South Asian ethnicity (vs. White), working as a frontline worker in a health or care setting (vs. not working as a frontline worker), recent travel to a place of work or study, number of public transport journeys, visits to shops and other indoor public places, travel outside of the UK, high levels of alcohol consumption (≥ 15 units per week), high body-mass index (BMI ≥ 25 kg/m²), sex hormone therapy (i.e. hormone replacement therapy and hormonal contraception) and use of vitamin D supplements were independently associated with higher risk of SARS-CoV-2 infection as indicated by antibody seropositivity (Table 2). By contrast, postgraduate education (vs. primary or secondary), passive smoking, high levels of light physical exercise (walking ≥10 h per week) and prescribed paracetamol use were independently associated with lower risk of SARS-CoV-2 infection. In the fully adjusted model, the associations originally observed in minimally adjusted models for generational composition of households, living with a working-age adult, and lower impact physical activity no longer achieved conventional significance (Table 2). Excluding the 796 participants with symptom-defined probable COVID-19, who did not have a positive PCR or lateral flow test result before enrolment, had little effect on our findings and associations with only 5 items were substantially attenuated in the minimally adjusted model, including environmental tobacco smoke exposure, public transport journeys, people per bedroom, dairy products intake and use of sex hormone therapy (Additional file 1: Table S4).

When investigating associations with antibody titres, analysed as a continuous outcome in the subset of seropositive participants only, we found that 35 factors were independently associated with antibody titres with p < 0.10 after adjustment for age, sex and duration of follow-up (Table 3). The distribution of titres for three of these factors—ethnicity, frontline worker status, and COVID-19 severity—are shown in Fig. 2, with higher medians for non-White ethnicities, health or social care frontline workers, and participants who were hospitalised for treatment of COVID-19. Additional file 1: Table S5 shows factors with no evidence of association with antibody titre. When the 33 factors were included together in a fully adjusted model, we found that South Asian ethnicity (vs. White), having a mortgage (vs. owning own home), working as a frontline worker in a health or care setting, being an ex-smoker (vs. a never-smoker), visits to shops and other indoor public places, travel outside of the UK, taking multivitamin supplements, consuming at least two portions of dairy products or calcium-fortified alternatives (vs. 0–1 portions), and high BMI were associated with higher antibody titres, whereas high levels of fruit, vegetable, or salad consumption and reporting feeling anxious or depressed at baseline were associated with lower antibody titres (Table 3). P-for-trend analyses suggested higher antibody titres with increasing intake of dairy or calcium-fortified alternatives and increasing BMI and lower antibody titres with increasing fruit, vegetable or salad consumption (Table 3). The associations in minimally adjusted models with chronic obstructive pulmonary disease, poor self-reported general health and use of metformin or statins were attenuated in the fully adjusted model and were no longer statistically significant (Table 3).

The addition of COVID-19 severity to our model of determinants of antibody titres in seropositive participants attenuated the associations observed for BMI and smoking status, but significant associations remained for all other variables (Table 3). Inclusion of the severity variable also led to a weak inverse association between male sex and antibody titres (GMR 0.94 [95% CI 0.89–1.00]; p = 0.03). Also, use of beta blockers (1.12 [1.01–1.24]; p = 0.03) and anticholinergics (1.16 [1.00–1.33]; p = 0.04) were positively associated with antibody titres in this model (Table 3). P-for-trend analyses suggested lower antibody titres with increasing fruit, vegetable and salad intake and higher titres with increasing intake of dairy or calcium-fortified alternatives and increasing age (Table 3). The variance in antibody titre explained by the fully adjusted model was increased from 14.5% (R² = 0.1447) to 20.5% (R² = 0.2047) after including disease severity as a covariate.

In this large, prospective, population-based serological study, we explored determinants of SARS-CoV-2 seropositivity and antibody titres, evaluating more than 80 potential sociodemographic, clinical and behavioural risk factors. We found that five factors—South Asian ethnicity, frontline occupation in health or social care, number of visits to shops and other indoor public places, international travel, and high BMI—were strongly associated both with higher risk of SARS-CoV-2 seropositivity among all participants and with higher antibody titres in the subset of seropositive participants. Lower levels of educational attainment and light physical exercise, and higher levels of public transport use and alcohol consumption, were found to associate with higher risk of SARS-CoV-2 seropositivity. Lower intake of dairy or calcium-fortified alternatives and higher intake of fruit, vegetables and salad were associated with lower antibody titres, as was reporting anxiety or depression. Importantly, most factors associated with antibody titres in seropositive participants retained significance after adjusting for disease severity.

Our results support previous studies that have found increased risk of SARS-CoV-2 seropositivity for healthcare workers [3, 4, 12], people of Asian ethnicity [3, 5, 12] and people with lower educational attainment [3, 4]. Non-White race/ethnicity has previously been highlighted as a determinant of both SARS-CoV-2 seropositivity [22‐24] and antibody titres [25], but questions remained over residual confounding [22]. Despite including a wide range of potential confounders, point estimates for all non-White participants remained elevated in both our seropositivity and titre analyses, and significantly so for South Asian participants, emphasising the need to further investigate the underlying biological or social factors driving this disparity. While we did not confirm increased seropositivity for Black participants, we lacked statistical power, as they represented only 0.4% of the cohort.

We identified two novel modifiable lifestyle factors associated with SARS-CoV-2 seropositivity: alcohol consumption and light physical exercise. High levels of alcohol intake are known to negatively affect immune response through several mechanisms [26], which supports our finding of higher risk among participants consuming more than 15 units of alcohol a week. By contrast, we observed lower risk among participants doing more than 10 h of light physical exercise per week. It has been speculated that there is a J-shaped relationship between exercise load and susceptibility to infection, whereby moderate exercise can improve immune response, but prolonged, high-intensity exercise can increase susceptibility to infection [27]. This curve might explain why we did not see similar benefits for vigorous physical activity.

Our finding that use of vitamin D supplements was associated with higher risk of seropositivity contrasts with a previous study, which found lower risk in a univariable model and no association after adjusting for confounders [23]; randomised controlled trials are needed to resolve questions around potential effects of vitamin D supplements on susceptibility to SARS-CoV-2. We found no associations for frontline workers not based in health or social care, at odds with previous findings [11, 12]. We also did not observe associations between seropositivity and age or sex, unlike other studies [5, 23, 24, 28, 29]. This may reflect the fact that we adjusted for more potential confounders than most of these studies and included behaviours that reflect social mixing and influence exposure to infectious index cases.

The strongest association with antibody titre was for disease severity, which explained a further 6% of variance when added to our full model, including 35 factors explaining 14.5% of variance in antibody titre together. After adjustment for disease severity, we uncovered ten factors associated with higher titres (greater age, South Asian ethnicity, higher BMI, working as a frontline worker in a health or care setting, greater number of visits to shops or other indoor places, international travel, taking multivitamin supplements, increased consumption of dairy products or calcium-fortified alternatives, and use of beta blockers and anticholinergic medications) and three associated with lower titres (male sex, high levels of fruit, vegetable, or salad consumption; and reporting feeling anxious or depressed at baseline). Different mechanisms may explain these associations. High intensity and frequency of exposure could be a cause of elevated antibody titres in participants who visited indoor public places or travelled abroad more frequently and in frontline health and social care workers [30], supported by previous findings of higher titres in healthcare workers [14]. Alternatively, greater immune reactivity could be a cause of higher titres, as seen in female participants, who generally have stronger innate and adaptive immune responses than males [31]. Diet and nutrition are known to affect immune responses [32] and thus might explain the higher titres observed with use of multivitamin supplements and higher levels of dairy intake (potentially reflecting higher calcium intakes) [33]. However, little evidence is available for the effect of vitamin supplementation in suboptimal rather than micronutrient-deficient diets [32], and after adjustment, we found no associations between intake of any individual vitamin supplements and antibody titres. The negative association with fruit, vegetable, and salad consumption was observed for the highest level of intake only (≥ 6 portions a day); however, despite 40% of vegan or vegetarian participants being included in that category, neither diet type was found to be associated with antibody titres, suggesting it is not the result of a restricted diet. Our finding of a significant positive dose–response relationship between age and antibody titres after adjustment for disease severity supports findings from other studies [13, 34, 35].

This study has several strengths. Use of serology to measure SARS-CoV-2 infection reduces collider bias, as serology testing was offered to all participants enrolled in COVIDENCE UK, in contrast to results from external routine testing that had limited availability, particularly at the start of the pandemic. Serology testing has also allowed us to better quantify the risk of SARS-CoV-2 infection by capturing previous infections that were asymptomatic or unconfirmed. A further strength is the use of an assay with high sensitivity and specificity that targets three different types of antibody [36], increasing the probability of identifying a past infection. Additionally, we used dried blood spots for our sampling, which have been found to reduce processing failures compared with microtubes [37], which are currently used by large seroprevalence surveys [38]. The prospective nature of our study reduces the potential for reverse causation explaining our findings, and the granularity of our questionnaire allowed us to explore potential determinants and confounders that other studies have not investigated.

Our study also has some limitations. First, COVIDENCE UK is a self-selected cohort, and thus several groups—such as people younger than 30 years, people of lower socioeconomic status, and non-White ethnic groups—are under-represented. This particularly affected our power to investigate outcomes for Black participants, who have been found to be at higher risk of SARS-CoV-2 infection [4, 5, 12] and adverse outcomes [22] than White people. However, insufficient representativeness in a cohort does not preclude identification of causal associations, and self-selection may result in better response to follow-up [39]. Second, as we included asymptomatic infections in our titre analysis, we were not able to adjust for timing of infection onset, preventing us from capturing the effects of temporal changes in antibody responses. As 40% of our seropositive participants did not experience symptoms suggestive of COVID-19 or provide a symptom onset date, excluding them would have greatly reduced the power and generalisability of our analysis. Third, as with any observational study, we cannot exclude the possibility that some of the associations we report might be explained by residual or unmeasured confounding. For example, the finding that passive smoking but not active smoking was associated with a reduced risk of seropositivity compared with never-smokers should be treated with caution, unless a plausible protective mechanism can be found. However, we have minimised potential confounding by adjusting for a comprehensive list of putative risk factors. Finally, we explored numerous potential associations and cannot discount the possibility that some may have achieved statistical significance as a result of type 1 error. We therefore hope that future studies will test for the associations we report to determine whether they can be replicated in different populations.

This prospective serological study shows that people of South Asian ethnicity, frontline workers in health or social care, people with high BMI, and those who had more visits to indoor public places or who had travelled abroad were at higher risk of SARS-CoV-2 seropositivity, after robust adjustment for confounders. Moreover, among seropositive participants, all of these factors associated independently with higher antibody titres, regardless of disease severity. We additionally show that higher alcohol consumption and lower light physical exercise, both modifiable lifestyle factors, are associated with higher risk of seropositivity. Future research should focus on modifiable risk factors for seropositivity, as well as determinants of antibody titres and other correlates of protection after SARS-CoV-2 infection, to better understand which groups are most at risk of reinfection and what preventive measures might be taken.