We used the Doetinchem Cohort Study [18, 19], that started in 1987 with a population-based sample of men and women aged 20–59 years old who have been followed up every 5 years. The study collects data on lifestyle factors, biological measurements, physical and cognitive functioning, social aspects, comorbidities, and other background characteristics. From this cohort we invited all 3647 remaining participants to take part in the COVID-19 vaccination study. Participants were included in the study if they planned to receive COVID-19 vaccination or had completed the primary vaccination series within the last 28 days, as a month post second vaccination was the primary endpoint of the study.

The numbers of participants in the study are depicted in Fig. 1. In total 1457 DCS subjects were included in the vaccination study. As the study commenced after the start of the national vaccination campaign and vaccines were rolled-out per age group from old to young according to the national guidelines, some persons missed the pre-vaccination (T0) or even the T1 sampling. Thus, the number of individuals included in the study increased at subsequent timepoints. The median interval between the two vaccination doses was 35 days (interquartile range, IQR: 35–35) and did not differ between ages. At pre-vaccination (T0), 916 of the participants had a baseline antibody measurement taken, had complete cohort data, and were negative for COVID-19 infection. At 1 month after the first vaccination (T1) this applied to 1118 individuals and at a month after the second vaccination (T2) to 1257 individuals. Prior to receiving a vaccination 8.3% tested positive for COVID-19 and 1 month after completing the primary vaccination series this was 8.2%.

For further analysis, persons who had not yet been infected prior to vaccination or during our study (infection naive) were selected. One thousand twenty individuals were sampled at both T1 and T2. In these individuals the fold increase in antibody concentration between the two vaccinations was determined. Since the majority (78% at T1 and T2) of the participants was vaccinated with BNT162b2, the main analyses were done on this group. Persons of 60–65 years of age have mainly been vaccinated with AZD1222 (20% at T1 and T2). Therefore, the antibody response across the different timepoints has been evaluated in this subset of individuals.

Blood samples and questionnaires were taken prior to COVID-19 vaccination (T0 +7), 28 (-8 + 15) days after the first vaccination (T1), and 28 (-15 + 24 days) after the second vaccination (T2). The median interval between the two vaccination doses was 35 days (interquartile range, IQR: 35–35). Questionnaires covered demographic factors, COVID-19 vaccination information (type and date of vaccination), and SARS-CoV-2 testing information. Finger-prick blood samples were self-collected in microtubes and returned by mail. Serum was isolated from each sample by centrifugation and stored at -20°C until sample processing.

Immunoglobulin G (IgG) antibody concentrations against Spike S1 and Nucleoprotein (N) were measured simultaneously using a bead-based assay as previously described [26]. IgG concentrations were calibrated against the International Standard for human anti-SARS-CoV-2 immunoglobulin (20/136 NIBSC standard) and expressed as binding antibody units per milliliter (BAU/ml) [27]. The threshold for seropositivity was set at 10.1 BAU/ml for Spike S1 [28] and 14.3 BAU/ml for Nucleoprotein [29].

Participants had filled in questionnaires relating to quality of life and general health during each 5-year follow up phase of the DCS (Round 1 – 7) prior to the vaccination study. Further data was collected covering various topics such as demographic and lifestyle factors, and comorbidities, both self-reported and confirmed by physicians. Questionnaires were sent out via mail but participants could rely on assistance from professional healthcare workers in case they requested it. Questionnaires sent out prior to the vaccination study were also validated by a professional healthcare worker and the participant during the physical examination performed in each round of the Doetinchem Cohort Study. In addition, the physical examination included measurement of blood pressure, lung function, a cognitive test battery, physical functioning, as well as taking a blood sample for measurement of total- and HDL-cholesterol, and glucose. For CRP and glycA which had been measured in stored blood samples previously, the most recent measurements were used.

Using the collected data a frailty index was calculated. This frailty index is a measure consisting of 36 ‘deficits’ defined based on chronic conditions, cognitive, physical, and psychological functioning as described before [30]. The 36 deficits were selected based on previous inclusion in existing frailty indexes, a prevalence of greater than one percent in the entire DCS cohort, and if there was a known association with cognitive, physical, or psychological functioning. Health deficits were either dichotomized or trichotomized with 0 indicating total absence, 0.5 indicating partial/mild presence, and 1 indicating total presence of a given deficit. The sum of deficits was then divided by the number of deficits included resulting in an index ranging from 0 (completely non-frail) to 1 (completely frail). This measure of frailty has been linked to various inflammatory markers and clinically relevant health related outcomes before within the DCS [31].

IgG concentrations were log-transformed prior to all analyses resulting in approximately normally distributed values. In all analyses the IgG response at T1, T2, and the relative increase between these two timepoints were analyzed separately. All statistical analyses were performed using R version 4.2.0. Statistical significance was defined using a p-value not greater than 0.05.

To test whether frailty and age influenced both the absolute and relative vaccine induced IgG response a Pearson correlation analysis was performed. This was done to determine which specific variables, used to construct the frailty index, to include for further analysis. Linear regression models correcting for age and sex were constructed to highlight how the different frailty-related parameters as well as other comorbidities were associated with the IgG response independent of age and sex. For the resulting P values of these linear models a Benjamini-Hochberg correction was performed for the adjusted P values.

A multivariable linear regression model was constructed including several preselected variables commonly associated with clinically relevant health outcomes. These variables included age, sex, socioeconomic status, physical activity, waist circumference, smoking behavior, alcohol consumption, systolic blood pressure, (HDL) cholesterol, creatinine, glucose, glycA, and CRP concentrations, as well as kidney function, lung function, frailty index, and the number of comorbidities.

Following this, a multivariate linear regression model was constructed using all frailty-related parameters and comorbidities. First, multiple stepwise regression was performed on a subset of the samples without missing data to select which variables would be included in the regression model. This was done to identify which combination of variables led to the most parsimonious model that best explained the vaccine induced IgG. The variables selected out of these frailty-related parameters and other comorbidities were used to create a multivariate linear model using all samples available in order to estimate the effects of each of these selected variables.