Study design and participants
COVIDENCE UK is a prospective, longitudinal, population-based observational study of COVID-19 in the UK population (www.qmul.ac.uk/covidence) [16]. Inclusion criteria were age 16 years or older and UK residence at enrolment, with no exclusion criteria. Participants were invited via a national media campaign to complete an online baseline questionnaire to capture information on potential symptoms of COVID-19 experienced since February 1, 2020, results of any COVID-19 tests, and details of a wide range of potential risk factors for COVID-19 (Additional file 1: Table S1). Online monthly follow-up questionnaires captured incident test-confirmed COVID-19 and symptoms of acute respiratory infection (Additional file 1: Table S2). The study was launched on May 1, 2020.
The antibody study described here was introduced as an approved protocol amendment (amendment 3; November, 2020). Participants enrolled before the amendment were invited via email to participate in the antibody study and to give additional consent. As part of the antibody study, participants were invited to participate in serology testing from November, 2020. For this analysis, we included all participants who enrolled in the study between May 1 and November 2, 2020, partaking in serology testing who were not vaccinated against COVID-19 or who provided their dried blood spot sample on or before the date of their first COVID-19 vaccination. This paper reports findings from analysis of data collected up to April 18, 2021.
COVIDENCE UK was sponsored by Queen Mary University of London and approved by Leicester South Research Ethics Committee (ref 20/EM/0117). It is registered with ClinicalTrials.gov (NCT04330599).
Procedures
Antibody study participants were sent a kit containing instructions, lancets, and blood spot collection cards, to be posted back to the study team. Once returned, the samples were logged by the study team and sent in batches to the Clinical Immunology Service at the Institute of Immunology and Immunotherapy of the University of Birmingham (Birmingham, UK). Up to two more test kits were offered to participants whose initial samples were found to be insufficient for testing. Blood spot samples were taken from November 6, 2020, to April 18, 2021.
Semi-quantitative determination of antibody titres in dried blood spot eluates was done using a commercially available ELISA that measures combined IgG, IgA, and IgM (IgGAM) responses to the SARS-CoV-2 trimeric spike glycoprotein (product code MK654; The Binding Site [TBS], Birmingham, UK). The SARS-CoV-2 spike used is a soluble, stabilised, trimeric glycoprotein truncated at the transmembrane region [17, 18]. This assay has been CE-marked with 98.3% (95% CI 96.4–99.4) specificity and 98.6% (92.6–100.0) sensitivity following RT-PCR-confirmed mild-to-moderate COVID-19 that did not result in hospitalisation [15]. A cut-off ratio relative to the TBS cut-off calibrators was determined by plotting 624 pre-2019 negatives in a frequency histogram. A cut-off coefficient was then established for IgGAM (1.31), with ratio values classed as positive (≥ 1) or negative (< 1). Dried blood spots were pre-diluted at a 1:40 dilution with 0.05% PBS-Tween using a Dynex Revelation automated absorbance microplate reader (Dynex Technologies; Chantilly, VA, USA). Plates were developed after 10 min using 3,3′,5,5′-tetramethylbenzidine core and orthophosphoric acid used as a stop solution (both TBS). Optical densities at 450 nm were measured using the Dynex Revelation. Results of ELISA for detection of anti-Spike antibodies in dried blood spot eluates have previously been shown to have almost perfect agreement with those performed on serum (Cohen’s kappa = 0.83) [15].
Outcomes
Study outcomes were presence versus absence of antibodies against SARS-CoV-2 (binary outcome assessed in all participants who did not report having tested positive for SARS-CoV-2 infection via RT-PCR or lateral flow test before enrolment) and antibody titres (continuous outcome measured in all seropositive participants).
Independent variables
Eighty-eight putative risk factors for SARS-CoV-2 infection were selected a priori, covering sociodemographic, occupational and lifestyle factors; longstanding medical conditions and prescribed medication use; Bacille Calmette Guérin and measles, mumps, and rubella vaccine status; and diet and supplemental micronutrient intake (Additional file 1: Tables S1, S2). These factors, which were obtained from the baseline questionnaire, were included as independent variables in our models. To produce patient-level covariates for each class of medications investigated, participant responses were mapped to drug classes listed in the British National Formulary or the DrugBank and Electronic Medicines Compendium databases if not explicitly listed in the British National Formulary, as previously described [16]. Index of Multiple Deprivation (IMD) 2019 scores were assigned according to participants’ postcodes, and categorised into quartiles. Duration of follow-up was defined as the number of days between the date of enrolment and the date of dried blood spot collection.
Statistical analysis
Using the Stata powerlog program, we estimated that a minimum sample size of 10,964 would be required to detect a difference of at least 2% in the proportion of exposed vs. unexposed participants experiencing a given binary outcome [equivalent to an odds ratio (OR) of 1.08], with 90% power, for a binary exposure with maximum variability (probability 0.50 changing to 0.52) and a moderate correlation (R2 = 0.4) with other variables in a logistic regression model, using a two-sided test and 5% significance. The antibody study was a pragmatic study including all participants meeting the inclusion criteria, with no sample size specified.
Logistic regression models were used to estimate ORs and 95% CIs for potential determinants of SARS-CoV-2 seropositivity. Linear regression models with robust standard errors were used to estimate geometric mean ratios (GMRs) and 95% CIs for potential determinants of log-transformed antibody titres in seropositive participants. We first estimated ORs and GMRs in minimally adjusted models, and carried forward factors independently associated with each outcome at the 10% significance level to fully adjusted models. Both the minimally adjusted and fully adjusted models were controlled for age (< 30 years, 30 to < 40 years, 40 to < 50 years, 50 to < 60 years, 60 to < 70 years, and ≥ 70 years), sex (male vs. female) and duration of follow-up (days). We calculated p for trend for ordinal variables by re-running the regressions treating each ordinal variable in turn as continuous. Analyses were done for all participants with available data; missing data were not imputed. Correction for multiple comparisons was not applied, on the grounds that we were testing a priori hypotheses for all risk factors investigated [19].
In a sensitivity analysis, we excluded participants from the seropositivity analysis who were classified as having had probable COVID-19 before enrolment on the basis of self-reported symptoms, using the symptom algorithm described and validated by Menni and colleagues [20].
As antibody titres have been found to be associated with disease severity [13, 21], we did an exploratory analysis to investigate the extent to which COVID-19 severity might explain associations between independent variables and antibody titres, by including this as an explanatory variable in the titre analysis. COVID-19 severity was classified into three groups: ‘asymptomatic’ (non-hospitalised seropositive participants, who either did not report any symptoms of acute respiratory infection or whose symptoms were classified as having < 50% probability of being due to COVID-19, using the symptom algorithm by Menni and colleagues [20]), ‘symptomatic non-hospitalised’ (non-hospitalised seropositive participants who reported symptoms of acute respiratory infection that were classified as having ≥ 50% probability of being due to COVID-19, using the symptom algorithm [20]) and ‘hospitalised’ (seropositive participants who were hospitalised for treatment of COVID-19).
We present descriptive statistics as n (%), mean (SD), or median (IQR). Statistical analyses were done using Stata (version 14.2; StataCorp, College Station, TX, USA).
Role of the funding source
The study funders had no role in the study design, data analysis, data interpretation, or writing of the report.