Comparison of methodological approaches in COVID-19 vaccine effectiveness estimation using observational data

Comparison of methodological approaches in COVID-19 vaccine effectiveness estimation using observational data | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Comparison of methodological approaches in COVID-19 vaccine effectiveness estimation using observational data Anne J Huiberts, Bente Smagge, Henri van Werkhoven, Brechje de Gier, and 7 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8831085/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Background Target trial emulation (TTE) is a framework to systematically address potential biases in causal inference when using observational data. We estimated vaccine effectiveness (VE) of the Omicron XBB.1.5 booster vaccination against SARS-CoV-2 infection between 2 October 2023 and 2 April 2024 using four TTE approaches. Methods A hypothetical target trial was designed where eligible participants would be randomly assigned to receive booster vaccination or not. Four approaches were used to emulate this hypothetical trial using data of an ongoing prospective cohort study in the Netherlands. The first and second approach defined time zero as the start of the booster vaccination rollout and considered vaccination as a time-varying variable. The first approach adjusted for confounders by regression adjustment, while the second used inverse probability weighting. The third and fourth approach used multiple time zeros. In the third approach, all persons who received a booster vaccination on a specific day were matched 1:1 to persons who were not (yet) vaccinated on that day. In the fourth approach, confounders were adjusted for using inverse probability weighting. Results Overall VE was 30% (95%CI:26–34), 28% (24–32), 28% (24–32) and 27% (24–31) in the first, second, third and fourth approach, respectively. VE decreased as time since vaccination increased, but this was somewhat less pronounced in the third approach. Conclusions Estimated VE was similar across the four approaches. The choice of approach should be based on the model assumptions, estimand of interest, and feasibility. Cohort studies Coronavirus Disease 2019 (COVID-19) Causal Inference Vaccination Target Trial Emulation Figures Figure 1 Figure 2 BACKGROUND In epidemiology, causal questions are ideally answered using an appropriately designed and conducted randomized trial. However, these trials are often not feasible, not ethical, or time-consuming. Furthermore, external validity is often limited. For COVID-19 vaccines, randomized trials were unable to address important questions regarding vaccine effectiveness (VE). In particular, VE in specific subpopulations, VE against newly emerging variants, duration of protection, and the benefit of booster vaccinations could not be assessed in trials ( 1 ). For these questions, public health policy relies on observational studies. However, the nature of observational data poses several challenges for causal inference, including confounding bias, selection bias, and immortal time bias ( 2 – 4 ). To avoid making fundamental errors that can result in biased estimates and erroneous causal conclusions, one can use “target trial emulation” (TTE) ( 5 ). TTE is a framework for designing and analyzing observational studies that aim to estimate a causal effect. The systematic approach of TTE consists of two steps: 1) specify a hypothetical randomized trial that would answer the causal question, and 2) specify how each element of this trial was emulated using observational data ( 6 ). The protocol must include certain key elements that define the causal estimand, which is the description of the exact treatment effect under study, and the data analysis plan. The key elements are the eligibility criteria, treatment strategies, treatment assignment, the start and end of follow-up, outcomes, causal contrasts. To ensure that the observational study protocol properly emulates the design of a randomized trial that would answer the causal question, three components should be aligned at time zero (baseline): eligibility criteria are met, treatment strategies are assigned, and follow-up is started. For this, a key challenge is non-unique time zero, i.e. a single person may meet the eligibility criteria at multiple times, for instance throughout a vaccination campaign ( 7 ). To ensure alignment at time zero, one could choose a single time zero, for example the first eligible time, or use every eligible time as time zero. The latter requires emulating multiple sequential target ‘trials’, each of them with a different start of follow-up ( 7 ). TTE has become increasingly popular in post-marketing vaccine evaluation in recent years, particularly following the introduction of COVID-19 vaccines ( 8 , 9 ). As TTE does not stipulate which specific analytical and statistical methods should be used, the VE studies that applied the TTE framework featured various analytical approaches and statistical methods ( 8 , 9 ). Nonetheless, single and multiple time zero(s) approaches can be distinguished and several commonly used methods can be identified, although variation remains between studies that use each of these. Most common for VE studies is sequential target trial emulation with matching of unvaccinated to vaccinated participants ( 8 ). Inverse probability weighting (IPW) is also frequently applied, both in single and multiple time zero(s) approaches ( 1 , 10 , 11 ). In the Netherlands, a COVID-19 booster vaccination campaign started on October 2, 2023. In this campaign, persons at increased risk for severe COVID-19 were eligible for vaccination: persons ≥ 60 years, health care workers, and medical risk groups. A monovalent mRNA vaccine targeting the SARS-CoV-2 Omicron XBB.1.5 subvariant (Comirnaty; BioNTech-Pfizer, Germany/United States) was used almost exclusively. VE of the Omicron XBB.1.5 booster in the Netherlands has been studied ( 12 ), but without explicitly emulating a randomized trial as specified by the TTE framework. The aim of this study was to assess real-world VE of the Omicron XBB.1.5 booster vaccination using the TTE framework and using different design and analysis approaches, explain their differences, and assess variability in the obtained estimates. METHODS Study design The VAccine Study COvid-19 (VASCO) is a population-based prospective cohort study that was initiated when COVID-19 vaccination was introduced in the Netherlands ( 13 ). The primary study objective is to estimate COVID-19 VE against SARS-CoV-2 infection. Study details are described elsewhere ( 13 ). Briefly, over 45,000 community-dwelling adults aged 18–85 years were included in VASCO between May and December 2021. Participants are followed for five years after enrollment and data is collected using online questionnaires and self-collected fingerprick blood samples for SARS-CoV-2 serology, both at scheduled intervals. Questionnaires included, among others, questions on sociodemographic factors, COVID-19 vaccination, and SARS-CoV-2 testing results. Participants could also notify COVID-19 vaccinations and positive SARS-CoV-2 tests at any time during follow-up in the study application. SARS-CoV-2 self-tests are provided to participants to facilitate testing in case of COVID-19-like symptoms. Written informed consent was obtained from all participants prior to enrolment into the study. The VASCO study is conducted in accordance with the principles of the Declaration of Helsinki and the study protocol was approved by the Medical Ethics Committee of the Stichting Beoordeling Ethiek Biomedisch Onderzoek , Assen, the Netherlands. Study population, exposure, outcome and covariates The study population included participants of the VASCO study who were eligible for the COVID-19 XBB.1.5 booster vaccination, specifically: persons aged over 60 years, healthcare workers and those belonging to a medical risk group. COVID-19 vaccination was based on vaccinations registered in the Dutch national COVID-19 vaccination Information and Monitoring System, supplemented with self-reported vaccinations. How these data sources were combined is described elsewhere ( 13 ). Booster vaccination ( the exposure ) was defined as receipt of any SARS-CoV-2 vaccination between 2 October 2023 and 2 April 2024, the study period. SARS-CoV-2 infection ( the outcome ) was defined as a self-reported positive SARS-CoV-2 test, or positive serology (seroconversion or a 4-fold increase in antibodies against the nucleoprotein, N-antibodies, detected in two consecutive serology samples). In case of a serology-based infection (i.e. no corresponding infection reported), the infection date was imputed between the before and after infection detection sampling dates using SARS-CoV-2 incidence based on reported positive SARS-CoV-2 tests in the VASCO population per age group per day, with higher probability for dates with higher incidence. Covariates were age group (18–39, 40–59, 60–69, 70–85), sex (male, female), educational level, presence of a medical risk condition and time since prior infection. Educational level was classified as low (no or primary education), intermediate (secondary school or vocational training) or high (bachelor’s degree or higher), based on baseline data. Presence of a medical risk condition was defined as having one or more of following conditions: diabetes mellitus, lung disease or asthma, asplenia, cardiovascular disease, immune deficiency, cancer, liver disease, neurological disease, renal disease, organ or bone marrow transplantation. Time since prior infection was defined as time since most recent prior infection before 2 October 2023 (start study period), was based on self-reported positive SARS-CoV-2 tests and N-antibodies detected in serology samples (as described above), and was classified as no prior infection, a prior infection ≥ 1 year ago or a prior infection < 1 year ago. The covariates age, presence of a medical risk condition and time since prior infection were time-varying. During follow-up, age increased and presence of a medical risk condition could switch from no to yes if a medical condition was reported in a follow-up questionnaire. Time since prior infection could switch from prior infection < 1 year ago to prior infection ≥ 1 year ago. Participants could not switch from no prior infection to prior infection < 1 year ago, as infections during follow-up were considered outcomes and such participants were no longer eligible for vaccination within the vaccination campaign period. Methodological approaches and statistical analyses A hypothetical trial was formulated to answer our research question. Table 1 describes each element of the hypothetical trial. Four approaches were used to emulate this hypothetical trial (Table 1 ; Fig. 1 ). In all approaches, loss to follow-up date was defined as the completion date of the participant’s most recent scheduled questionnaire, as information on vaccinations and infection might be incomplete thereafter. A sensitivity analysis was performed, in which loss to follow-up date was defined as the first date a scheduled questionnaire was not completed, i.e. date of protocol deviation. All analyses were performed with R software version 4.4.3, using packages ipw (version 1.2.1) ( 14 ), MatchIt (version 4.5.5), survival (version 3.7.0), and boot (version 1.3.31). Table 1 Target trial protocol for estimating vaccine effectiveness of XBB.1.5 booster vaccination against SARS-CoV-2 infection Protocol component Hypothetical target trial Emulated target trial (single time zero with covariate adjustment) Emulated target trial (single time zero with inverse probability weighting) Emulated target trial (multiple time zeros with covariate matching) Emulated target trial (multiple time zeros with inverse probability weighting) Eligibility criteria - Age 18–85 years - Eligible for XBB.1.5 booster vaccination at baseline (HCW, 60+, medical risk; no infection in 90 days prior) - Understanding Dutch - Age 18–85 years at time zero - Eligible for XBB.1.5 booster vaccination (HCW, 60+, medical risk; no infection in 90 days prior) - Understanding Dutch - Age 18–85 years at time zero - Eligible for XBB.1.5 booster vaccination (HCW, 60+, medical risk; no infection in 90 days prior) - Understanding Dutch - Age 18–85 years at time zero - Eligible for XBB.1.5 booster vaccination (HCW, 60+, medical risk, no COVID-19 vaccine dose received since 2 October 2023, no infection in 90 days prior) - Understanding Dutch - Age 18–85 years at time zero - Eligible for XBB.1.5 booster vaccination (HCW, 60+, medical risk, no COVID-19 vaccine dose received since 2 October 2023, no infection in 90 days prior) - Understanding Dutch Treatment strategies 1. Administer XBB.1.5 booster vaccine 2. Do not administer XBB.1.5 booster vaccine 1. Administer XBB.1.5 booster vaccine at time zero 2. Do not ever administer XBB.1.5 booster vaccine 1. Administer XBB.1.5 booster vaccine at time zero 2. Do not ever administer XBB.1.5 booster vaccine 1. Administer XBB.1.5 booster vaccine 2. Do not administer XBB.1.5 booster vaccine 1. Administer XBB.1.5 booster vaccine 2. Do not administer XBB.1.5 booster vaccine Treatment assignment Randomization Adjustment for potential time-fixed and time-varying confounders by regression analysis. Potential confounders were age, sex, medical risk condition, educational level and time since prior infection. Adjustment for potential time-fixed and time-varying confounders by inverse probability weighting 1:1 matching on calendar time (sequential trial) and other potential confounders at time zero. Potential confounders were age, sex, medical risk condition, educational level and time since prior infection. Matching on calendar time (sequential trial) + adjustment for potential confounders by inverse probability weighting at time zero. Follow-up Time zero is when eligibility criteria for inclusion are met, treatment is assigned, and follow-up begins. Follow-up is from time zero to outcome, loss-to-follow up, death, or administrative end of follow-up, whichever occurred first. Time zero is the start of the vaccination roll-out on 2 October 2023 for the unvaccinated group and time of vaccination in the vaccinated group. Follow-up is from time zero to outcome, loss to follow-up or administrative end of follow-up, whichever occurred first. Time zero is the start of the vaccination roll-out on 2 October 2023 for the unvaccinated group and time of vaccination in the vaccinated group. Follow-up is from time zero to outcome, loss to follow-up or administrative end of follow-up, whichever occurred first. Time zero is date of vaccination or equivalent index date of the matched control who did not receive vaccination on or before that date. Follow-up is from time zero to outcome, loss to follow-up, receipt of a vaccine-dose by an individual in the unvaccinated group (with concurrent censoring of the vaccinated member of the pair), or administrative end of follow-up, whichever occurred first. Time zero is date of vaccination or equivalent index week of any control who did not receive vaccination on or before that date. Follow-up is from time zero to outcome, loss to follow-up, receipt of a vaccine-dose by an individual in the unvaccinated group, or administrative end of follow-up, whichever occurred first. Outcome SARS-CoV-2 infection between treatment allocation and end of study SARS-CoV-2 infection (self-reported or serology-based) between 2 October 2023 and 2 April 2024 SARS-CoV-2 infection (self-reported or serology-based) between 2 October 2023 and 2 April 2024 SARS-CoV-2 infection (self-reported or serology-based) between index date and 2 April 2024 SARS-CoV-2 infection (self-reported or serology-based) between index week and 2 April 2024 Causal contrast of interest Per-protocol, individuals in the control arm were censored upon receiving vaccination Per-protocol Per-protocol Per-protocol Per-protocol Statistical analysis Cumulative incidence (risk) curves of SARS-COV-2 infection using the Kaplan-Meier estimator; vaccine effectiveness = 100% * (1 – risk ratio (RR)) Cox regression model with calendar time as underlying time scale and time-varying exposure, with regression adjustment for (time-varying) covariates; vaccine effectiveness = 100% * (1 – hazard ratio (HR)) Cox regression model with time-varying exposure, weighted by time-varying IPTWs and IPCWs; vaccine effectiveness = 100% * (1 – hazard ratio (HR)) and robust standard error estimate Cumulative incidence (risk) curves using the Kaplan-Meier estimator; vaccine effectiveness = 100% *(1 – risk ratio (RR)) and bootstrapping to compute 95%CI Cox regression model stratified by trial, adjusting for covariates using IPTW and IPCW; vaccine effectiveness = 100% * (1 – hazard ratio (HR)) and bootstrapping to compute 95%CI HCW = healthcare worker; IPTW = inverse probability treatment weight(ing); IPCW = inverse probability censoring weight(ing); RR = risk ratio; HR = hazard ratio 1. Single time zero with regression adjustment This first approach is a time-dependent Cox model, which is widely used for estimating VE in cohort studies with time-to-event data, but not usually considered part of the TTE canon ( 15 , 16 ). Probably this approach is hardly used within the TTE framework because it is known to produce biased estimates in the presence of time-varying confounders that are themselves affected by the treatment ( 17 – 19 ). However, treatment-confounder feedback by the measured time-varying confounders in our model seems unlikely. The first eligible time was chosen as time zero. Participants were eligible for XBB.1.5 booster vaccination at the start of the vaccination roll-out on 2 October 2023 or three months after their most recent prior SARS-CoV-2 infection, whichever occurred last. Follow-up started at time zero. Treatment assignment was not aligned at time zero, because participants could receive a vaccination at any time between time zero and the end of the vaccination campaign. To account for this, vaccination status was coded as a time-varying variable that transitioned from ‘unvaccinated’ to ‘vaccinated’ seven days after administration of the XBB.1.5 booster vaccination and remained ‘vaccinated’ thereafter. The seven days after vaccination administration were excluded from the analysis. Follow-up ended on the date of the first SARS-CoV-2 infection, loss to follow-up or the end of the study period on 2 April 2024, whichever came first. A Cox proportional hazards model with calendar time as the underlying time scale was used to estimate VE. The estimate was adjusted for potential confounding by adding age group, sex, medical risk condition, educational level, and time since prior infection as covariables to the Cox regression model. Age group, medical risk condition and time since prior infection were time-varying covariates. VE was calculated as 100% × (1 − hazard ratio). To evaluate VE by time since vaccination, the exposure was categorized into six-week periods with ‘unvaccinated’ as the reference category. 2. Single time zero with inverse probability weighting Robins’ generalized methods (G-methods) are a class of methods that, unlike stratification or standard regression (e.g. the time-dependent Cox model used in our first approach), appropriately adjusts for confounding when treatment-confounder feedback exists( 19 ). Among available G-methods, models using time-varying inverse probability weights (IPWs), also known as marginal structural models, are most commonly used for our type of data and causal question ( 18 ). Eligibility criteria, follow-up time and the exposure were defined in the same manner as in the first approach. However, we used a Cox proportional hazards model with inverse probability of treatment weighting (IPTW) to adjust for confounding and inverse probability of censoring weighting (IPCW) to adjust for informative censoring due to non-random loss to follow-up ( 14 , 20 , 21 ). The final weights were obtained by multiplying the stabilized time-varying IPTWs and IPCWs. The weighting procedures are described in detail in the Supplementary file, Statistical analyses . Finally, a Cox proportional hazards model, weighted by the time-varying stabilized weights and additionally adjusted for the time-fixed confounders, was fit. This adjustment was required because stabilized weights do not adjust for time-fixed covariates ( 1 , 20 ). To account for person-level clustering induced by weighting, a robust standard error estimate was obtained. VE by time since vaccination was evaluated by categorizing the exposure into six-week periods with ‘unvaccinated’ as the reference category. 3. Multiple time zeros with covariate matching In the third approach, we used every eligible time as time zero, i.e. multiple time zeros. For each calendar day of the vaccination period, persons who met the eligibility criteria on that day, and who had not been vaccinated before that day, were identified and classified as either having or not having received XBB.1.5 booster vaccination on that day. This way, a sequence of ‘trials’ was created, with a new trial starting on each day. Each person who received a vaccination was matched exactly to a randomly selected control (with replacement) on age group, sex, medical risk condition, educational level, and time since prior infection. Eligible individuals could be selected as unvaccinated controls repeatedly up to the day before they were vaccinated. For each matched pair, follow-up started on the day of vaccine administration of the vaccinated member of the pair and ended at the date of the first reported SARS-CoV-2 infection, the vaccination date of the control, loss to follow-up or the end of the study period, whichever came first. Cumulative incidence curves of self-reported positive SARS-CoV-2 test were constructed using the Kaplan-Meier estimator. Both individuals of a matched pair had to be still at risk by day 8 of follow-up. The risk of SARS-CoV-2 infection more than 7 days after vaccination was compared with the risk among their matched controls. VE was calculated as 100% × (1 – risk ratio). 95% confidence intervals (CIs) were computed using bootstrapping (1000 iterations) of the matched dataset, to account for sequential trials not being independent. We estimated VE for each six-week period since vaccination by including only pairs that were not censored before the start of the period and censoring them at the end of the period. 4. Multiple time zeros with inverse probability weighting Similar to the third approach, in the fourth approach all eligible times were used as time zero. When selecting trials per day, estimation of the IPCWs resulted in non-positivity in some of the groups due to the absence of censoring. Therefore, we created trials per week, rather than per day. In each week of the study period, we identified persons who had not been vaccinated before that week (index week), who met the eligibility criteria, and who still met the eligibility criteria 7 days later at the start of follow-up. All participants who did not receive XBB.1.5 booster vaccination in that week were classified as unvaccinated. All participants who did receive XBB.1.5 booster vaccination sometime during that week were classified as vaccinated. Eligible individuals were enrolled as unvaccinated in each trial up to the week before their vaccination, but once they received vaccination they could not enroll in subsequent trials. We excluded weeks in which < 20 individuals received a XBB.1.5 booster vaccination or < 20 individuals remained unvaccinated to avoid model separation when estimating the probability of censoring. Follow-up started one week after each index week (i.e. 7 to 13 days after an individual's vaccination date) and was continued until the week of the first reported SARS-CoV-2 infection, loss to follow-up or the end of the study period, whichever came first. Unvaccinated controls were censored at the end of the week of their vaccination. IPW was used to adjust for confounding (IPTW) and informative censoring due to loss to follow-up (IPCW LTFU ) and due to protocol deviation, i.e. vaccination of unvaccinated controls during follow-up (IPCW PD ) ( 11 ). The computation of the IPWs is described in the Supplementary file, Statistical analyses . The final weights were obtained by multiplying the stabilized IPTWs, IPTWs PD and IPCWs LTFU . Finally, weighted Cox models stratified by trial were fit. VE was calculated as 100% × (1 − hazard ratio). The weighted model was adjusted for the time-fixed confounders that were included in the numerator model of the IPCWs. The 95% CIs were computed using bootstrapping (1000 iterations). To evaluate VE by time since vaccination, the exposure was categorized into six-week periods with ‘unvaccinated’ as the reference category. RESULTS 1. Single time zero with regression adjustment In total, 24,288 VASCO participants were eligible for XBB.1.5 booster vaccination and were included in the analysis. Of those, 15,300 (63%) received vaccination during the study period (Table 2 ). Participants who received vaccination were older, more often male, more often had a high education level, more often had a medical risk condition, and less often had experienced a prior infection. During 265,660 vaccinated and 202,961 unvaccinated person-weeks, 2,900 and 3,318 infections occurred, respectively. Risk of infection was higher among unvaccinated participants ( Supplementary file, Figure S1 ). Furthermore, incidence was highest among those without prior infection and lowest among those with a prior infection in the past year ( Supplementary file, Figure S2 ). Overall VE against infection was 30% (95% CI 26–34) (Fig. 2 ). VE decreased as time since vaccination increased, being 35% (30–39), 24% (18–30) and 8% (-9-23) at 1–6, 7–12 and 13–18 weeks post-vaccination, respectively (Fig. 2 ; Supplementary file, Table S1 ). Table 2 Characteristics of participants included in single time zero with regression adjustment analysis and multiple time zeros with covariate matching analysis, the Netherlands, 2 October 2023–2 April 2024 Single time zero with regression adjustment a Multiple time zeros with covariate matching Total No XBB.1.5 booster vaccination Yes XBB.1.5 booster vaccination Total No XBB.1.5 booster vaccination Yes XBB.1.5 booster vaccination n = 24288 n = 8988 n = 15300 n = 25834 n = 12917 n = 12917 Age group 18–39 1083 (4.5%) 822 (9.1%) 261 (1.7%) 494 (1.9%) 247 (1.9%) 247 (1.9%) 40–59 4094 (16.9%) 2888 (32.1%) 1206 (7.9%) 2280 (8.8%) 1140 (8.8%) 1140 (8.8%) 60–69 13457 (55.4%) 4137 (46%) 9320 (60.9%) 16284 (63%) 8142 (63%) 8142 (63%) 70–85 5654 (23.3%) 1141 (12.7%) 4513 (29.5%) 6776 (26.2%) 3388 (26.2%) 3388 (26.2%) Sex Female 15008 (61.8%) 6293 (70%) 8715 (57%) 14974 (58%) 7487 (58%) 7487 (58%) Male 9276 (38.2%) 2693 (30%) 6583 (43%) 10860 (42%) 5430 (42%) 5430 (42%) Other 4 (0%) 2 (0%) 2 (0%) 0 (0%) 0 (0%) 0 (0%) Education level High 13418 (55.2%) 4677 (52%) 8741 (57.1%) 14542 (56.3%) 7271 (56.3%) 7271 (56.3%) Intermediate 6835 (28.1%) 2841 (31.6%) 3994 (26.1%) 6902 (26.7%) 3451 (26.7%) 3451 (26.7%) Low 3878 (16%) 1415 (15.7%) 2463 (16.1%) 4230 (16.4%) 2115 (16.4%) 2115 (16.4%) Other 157 (0.6%) 55 (0.6%) 102 (0.7%) 160 (0.6%) 80 (0.6%) 80 (0.6%) Medical risk condition Yes 11384 (46.9%) 4099 (45.6%) 7285 (47.6%) 11952 (46.3%) 5976 (46.3%) 5976 (46.3%) Time since prior infection 1 year ago 17857 (73.5%) 6394 (71.1%) 11463 (74.9%) 12884 (49.9%) 6442 (49.9%) 6442 (49.9%) No prior infection 3398 (14%) 1041 (11.6%) 2357 (15.4%) 3834 (14.8%) 1917 (14.8%) 1917 (14.8%) a For the purpose of this table, vaccination status and all covariates were determined at the end of follow-up IPW = inverse probability weighting 2. Single time zero with inverse probability weighting In the second approach, the same participants were included as in approach 1. None of the estimated stabilized IPWs was > 10, thus truncation to 10 for extreme IPWs was not applied. Assessment of the balance of the time-varying variables between the vaccinated and unvaccinated population is shown in the Supplementary file, Figure S3) . Overall VE against infection was 28% (24–32) (Fig. 2 ). VE decreased as time since vaccination increased, with 33% (28–38), 22% ( 15 – 28 ) and 2% (-18-20) at 1–6, 7–12 and 13–18 weeks post-vaccination, respectively (Fig. 2 ; Supplementary file, Table S1 ). 3. Multiple time zeros with covariate matching On 95 dates at least one person was vaccinated and could be matched to a (yet) unvaccinated individual. The number of matched pairs ranged between 1 and 348 per trial ( Supplementary file, Figure S4 ). In total, 12,917 vaccinated participants (out of 15,300) were matched to unvaccinated participants (Table 2 ), in whom 1,384 and 1,872 infections occurred, respectively ( Supplementary file, Figure S5 ). Most characteristics of included participants were comparable to characteristics of vaccinated individuals in the entire study population (as included in approach 1). Matched vaccinated individuals were less often in the oldest age group (≥ 70 years), possibly because vaccination coverage in this age group was high and few unvaccinated were available for matching. Overall VE against infection was 28% (24–32) (Fig. 2 ). VE decreased as time since vaccination increased with 32% (26–37), 33% (28–38) and 14% (-6-30) at 1–6, 7–12 and 13–18 weeks post-vaccination, respectively (Fig. 2 ; Supplementary file, Table S1 ). 4. Multiple time zeros with inverse probability weighting In 12 weeks, a trial started. Eight weeks were excluded because of < 20 individuals in the vaccinated and/or unvaccinated group; the week of Christmas and seven weeks after the official end of the vaccination campaign. The number of participants ranged between 5,722 and 20,871 per trial. In total, 141,638 non-unique individuals were included. After weighting, the vaccinated and unvaccinated population were comparable ( Supplementary file, Figure S6 ). Truncation of large IPWs was only required for < 0.01% of the estimated IPWs. In total, 2,744 and 19,886 infections occurred in vaccinated and unvaccinated participants ( Supplementary file, Figure S7 ). Overall VE against infection was 27% (24–31) (Fig. 2 ). VE decreased as time since vaccination increased, with 33% (29–38), 24% ( 18 – 29 ) and 5% (-13-22) at 1–6, 7–12 and 13–18 weeks post-vaccination, respectively (Fig. 2 ; Supplementary file, Table S1 ). The sensitivity analysis in which loss to follow-up date was defined as the first day that a scheduled questionnaire was not completed resulted in negligible differences in the estimates of the four approaches ( Supplementary file, Figure S8 ). DISCUSSION In this study we used four TTE approaches to estimate VE of the XBB.1.5 booster vaccination against SARS-CoV-2 infection. The four approaches yielded very similar overall VE estimates ranging between 27 and 30%, with considerably overlapping 95% confidence intervals. Somewhat larger deviations were observed between estimates of VE by time since vaccination. In particular, using the approach with multiple time zeros and covariate matching, we found a somewhat smaller decrease in VE point estimates over time compared to the other approaches. A potential explanation for this is that period-specific estimates may be affected by differential depletion of susceptibles ( 7 , 22 ). Only participants who remained event-free (and did not drop out) until a certain time, will be included in the subsequent period for which the hazard ratio (HR) is estimated. Individuals who are at increased risk of infection, e.g. due to weaker immunity or greater exposure, are likely to get infected earlier. Therefore, those at increased risk are ‘removed’ from the at-risk population at a higher rate, leaving individuals who are on average less susceptible or have less exposure for the estimation of VE at a longer time since vaccination. As the unvaccinated group is not protected by vaccination, infections are likely to occur at a faster rate in this group, leading to more rapid depletion of susceptibles in the unvaccinated compared to the vaccinated group. Due to this differential depletion of susceptibles, VE based on period-specific HRs may be artificially low long after vaccination, as the vaccinated group is compared to a progressively less at-risk unvaccinated group. In our approach using multiple time zeros with covariate matching, both the vaccinated and unvaccinated member of each matched pair were required to stay event-free (and not drop out) until the start of the time interval since vaccination, thereby vaccinated and unvaccinated individuals may be more comparable with respect to susceptibility of infection given the measured confounders. Compared to the approaches featuring Cox regression, the multiple time zeros with covariate matching approach thereby may be less affected by spurious waning caused by differential depletion of susceptibles. It might therefore be a preferred approach for investigating the effect of time since vaccination on the VE. The multiple time zeros with covariate matching approach (approach 3) also differs from the other three approaches in two other ways. Firstly, approach 3 estimates the average treatment effect on the treated (ATT), or, strictly speaking, the average treatment effect in the overlap population (ATO), as opposed to the overall average treatment effect (ATE). The ATE estimates the effect of treatment across the entire population regardless of whether individuals actually received treatment. It is the difference between the expected outcome if all individuals were treated and the expected outcome if none were treated. The ATT focuses on the effect only within those individuals that actually received treatment. This is a consequence of the matching in approach 3. As some vaccinated individuals who could not be matched were removed, actually approach 3 estimates the ATO: the average treatment in the overlap/equipoise population, i.e. the population with comparable chances of being (un)treated based on the matching variables ( 23 ). Differences between the ATE and ATT or ATO arise when the treatment effect varies between population subgroups (i.e., interactions between the treatment and individual characteristics) and the treated and untreated differ in these characteristics. Our data indeed showed differences in characteristics of vaccinated (treated) and unvaccinated (untreated) individuals, therefore at least the second requirement holds true. Secondly, approach 3 estimates a marginal effect, whereas the other approaches estimate a conditional effect. Conditional effects control for the values of other variables, making them context-dependent. Marginal effects summarize the average effect of a variable across the entire distribution of other variables of the model ( 24 ). For collapsible measures of effect, like the risk ratio estimated in approach 3, the marginal and conditional effects are the same in the absence of interactions between the exposure and covariates. For non-collapsible measures of effect, like the hazard ratio estimated in approaches 1, 2 and 4, the conditional effect differs from the marginal effect if the covariates that are conditioned on are predictive of the outcome, even if there is no association with the exposure (i.e. confounding) or effect modification ( 24 ). Because the estimand of the multiple time zeros with covariate matching approach differs from the other approaches in these ways, the effect estimates are not directly comparable and, although this is not evident in our results, their numeric value may differ. The methods used have varying assumptions and limitations. First, Kaplan-Meier (approach 3) is a non-parametric method with no assumptions about the distribution of the unobserved failure times due to administrative censoring. In contrast, the semi-parametric Cox regression model (used in approaches 1, 2 and 4), while not assuming a particular distribution for failure times or hazards either, imposes a priori restrictions on the relation between the baseline hazard and the hazard under other combinations of covariate values ( 25 ). Second, the approach using a single time zero with Cox regression is biased in the presence of treatment-confounder feedback, i.e. when a time-varying covariate that is affected by vaccination status is adjusted for ( 17 – 19 ). Studies that measured such covariates and want to appropriately adjust for them should avoid this approach. Third, using multiple time zeros with covariate matching results in a smaller number of participants and outcomes due to matching in case not every vaccinated individual can be matched to an unvaccinated individual and due to censoring of the entire pair when the unvaccinated received vaccination. This may result in wider CIs, specifically with increasing time since vaccination. By matching, this method also eliminates the possibility of studying the effect of individual confounders in sensitivity analyses ( 26 ). Finally, the multiple time zeros with IPW approach is computationally intensive and requires a lot of data. In our study, we created trials per week rather than per day because of non-positivity, but we expect limited effect of this on our results. It was impossible to estimate the IPCW LTFU by trial because this type of censoring was too rare, so we estimated the IPCW LTFU in the dataset including all trials. This assumes that the associations between the covariates and the probability of censoring due to loss to follow-up are the same across trials. The TTE framework is useful to systematically think through every step of the design and analysis of an observational study aiming to answer a causal question. This way, it helps researchers to recognize, appropriately deal with, and communicate about methodological pitfalls and potential biases inherent to the use of observational data for causal inference ( 9 , 26 , 27 ). Nonetheless, the TTE framework has several caveats. First, the benefit of the framework depends on how rigorously it is applied and this needs to be fully reported for readers to understand and assess. One systematic review reported that only 58% of studies explicitly aiming to emulate a target trial completely report how the target trial was emulated ( 9 , 28 ). Second, applying the TTE framework can still lead to multiple ‘as correct as possible’ study designs that may yield comparable estimates, as was the case in our study. On the other hand, if the estimates do differ between approaches, it may be difficult to uncover which one is (most) correct. Third, studies designed using the TTE framework may still be flawed, because TTE does not remove the limitations of the data, such as use of self-reported data, or unmeasured confounding ( 26 ). For example, in this study, we had to assume non-interference ( 8 ) and could only assess the direct effect of vaccination. We also did not have sufficient participants to estimate VE by subgroup eligible for vaccination, i.e. among persons aged ≥ 60 and among younger clinically vulnerable individuals. Fourth, (too) strict adherence to the TTE framework can be contentious. Overzealous application/interpretation of the TTE framework would suggest that studies should have a prospective cohort design in order to mimic an RCT, but case-control studies are valuable too and can indeed also be designed within the TTE framework( 29 ). For COVID-19 VE specifically, test-negative case-control designs have been advantageous to reduce bias introduced by health-seeking behavior. It is also debatable whether the single time zero approaches can be considered properly emulated target trials within the TTE framework, because treatment assignment is not aligned at time zero. However, these approaches handle this by allowing vaccination status to be time-varying, so it can be argued that the optimal analytical approach does not necessarily need to align eligibility, treatment assignment and the start of follow-up to obtain an unbiased estimate. Indeed, the similarity of our estimates suggests that these approaches do not provide inherently less reliable effect estimates. In short, TTE is neither a condition for nor a guarantee of unbiased causal inference, but it helps to avoid methodological pitfalls and biases. CONCLUSIONS In this study, the hypothetical target trial to estimate the VE of the XBB.1.5 booster vaccination was emulated using four different design and analysis approaches. Although we cannot know the true VE, the VE estimates were similar across the four approaches. To the extent that the four approaches have different assumptions and estimands, we can conclude that the VE estimates in this study are not sensitive to these differences, except possibly when estimating VE over time since vaccination. This study highlights that observational data can be analyzed in multiple valid ways to answer a causal research question. The choice of approach should depend on the estimand of interest (e.g. ATT vs. ATE, marginal vs. conditional effect), the plausibility of the assumptions of various approaches (e.g. proportional hazards, no treatment-confounder feedback) and practical and technical feasibility (e.g. computational power, data quantity). Abbreviations ATE = average treatment effect ATO = average treatment effect in the overlap population ATT = average treatment effect on the treated COVID-19 = Coronavirus Disease 2019 G-methods = generalized methods HR = hazard ratio IPCW = inverse probability of censoring weight(ing) IPTW = inverse probability of treatment weight(ing) IPW = inverse probability weight(ing) LTFU = loss to follow-up PD = protocol deviation SARS-CoV-2 = Severe Acute Respiratory Syndrome Coronavirus 2 TTE = target trial emulation VASCO = VAccine Study COvid-19 VE = vaccine effectiveness Declarations Ethics approval and consent to participate Written informed consent was obtained from all participants prior to enrolment into the study. The VASCO study is conducted in accordance with the principles of the Declaration of Helsinki and the study protocol was approved by the Medical Ethics Committee of the Stichting Beoordeling Ethiek Biomedisch Onderzoek , Assen, the Netherlands. Consent for publication Not applicable. Availability of data and materials Anonymized data reported from this study can be obtained from the corresponding author upon request. The dataset may include individual data and a data dictionary will be provided. Data requests should include a proposal for the planned analyses. Data transfer will require a signed data sharing agreement. Code for data processing, statistical analysis, figures, and tables can be found at GitHub (https://github.com/b-smagge/TTE_COVID_VE). Competing interests The authors declare that they have no competing interests. Funding This work was funded by the Ministry of Health, Welfare and Sports of the Netherlands. Authors’ contributions All authors have read and approved the final manuscript. AH, BS, HvW, BdG and MK conceptualized and designed the study. AH and BS performed data cleaning and data analysis. AH and BS drafted the manuscript. HvW, BdG, SM, HdM, SH, DG, JvdW, SvdH and MK critically reviewed the paper. Acknowledgements We would like to thank Maarten Schipper for his valuable statistical insights, which greatly contributed to the interpretation of the results in this work. References Hulme WJ, Williamson E, Horne EMF, Green A, McDonald HI, Walker AJ, et al. Challenges in Estimating the Effectiveness of COVID-19 Vaccination Using Observational Data. Ann Intern Med. 2023;176(5):685–93. Kuehne F, Arvandi M, Hess LM, Faries DE, Matteucci Gothe R, Gothe H, et al. Causal analyses with target trial emulation for real-world evidence removed large self-inflicted biases: systematic bias assessment of ovarian cancer treatment effectiveness. J Clin Epidemiol. 2022;152:269–80. Maringe C, Benitez Majano S, Exarchakou A, Smith M, Rachet B, Belot A, et al. Reflection on modern methods: trial emulation in the presence of immortal-time bias. Assessing the benefit of major surgery for elderly lung cancer patients using observational data. Int J Epidemiol. 2020;49(5):1719–29. Hernán MA, Sauer BC, Hernández-Díaz S, Platt R, Shrier I. Specifying a target trial prevents immortal time bias and other self-inflicted injuries in observational analyses. J Clin Epidemiol. 2016;79:70–5. Hernán MA, Robins JM. Using Big Data to Emulate a Target Trial When a Randomized Trial Is Not Available. Am J Epidemiol. 2016;183(8):758–64. Hernán MA, Wang W, Leaf DE. Target Trial Emulation: A Framework for Causal Inference From Observational Data. JAMA. 2022;328(24):2446–7. Hernán MA, Robins JM. Causal Inference: What If: Boca Raton. Chapman & Hall/CRC.; 2020. Komura T, Watanabe M, Shioda K. Exploring the Application of Target Trial Emulation in Vaccine Evaluation. Scoping Rev medRxiv. 2024:2024.07.26.24311066. Zuo H, Yu L, Campbell SM, Yamamoto SS, Yuan Y. The implementation of target trial emulation for causal inference: a scoping review. J Clin Epidemiol. 2023;162:29–37. Gravenstein S, DeVone F, Oyebanji OA, Abul Y, Cao Y, Chan PA, et al. Durability of immunity and clinical protection in nursing home residents following bivalent SARS-CoV-2 vaccination. EBioMedicine. 2024;105:105180. McConeghy KW, Bardenheier B, Huang AW, White EM, Feifer RA, Blackman C, et al. Infections, Hospitalizations, and Deaths Among US Nursing Home Residents With vs Without a SARS-CoV-2 Vaccine Booster. JAMA Netw Open. 2022;5(12):e2245417–e. Huiberts AJ, Hoeve CE, de Gier B, Cremer J, van der Veer B, de Melker HE, et al. Effectiveness of Omicron XBB.1.5 vaccine against infection with SARS-CoV-2 Omicron XBB and JN.1 variants, prospective cohort study, the Netherlands, October 2023 to January 2024. Eurosurveillance. 2024;29(10):2400109. Huiberts AJ, Hoeve CE, Kooijman MN, de Melker HE, Hahné SJ, Grobbee DE, et al. Cohort profile: an observational population-based cohort study on COVID-19 vaccine effectiveness in the Netherlands - the VAccine Study COVID-19 (VASCO). BMJ Open. 2024;14(10):e085388. van der Wal WM, Geskus RB. ipw: An R Package for Inverse Probability Weighting. J Stat Softw. 2011;43(13):1–23. European Centre for Disease Prevention and Control. Protocol for a COVID-19 vaccine effectiveness study using health data registries, v.2.0. Stockholm: ECDC; 2024. Hoffman KL, Schenck EJ, Satlin MJ, Whalen W, Pan D, Williams N, et al. Comparison of a Target Trial Emulation Framework vs Cox Regression to Estimate the Association of Corticosteroids With COVID-19 Mortality. JAMA Netw Open. 2022;5(10):e2234425. Hernán MÁ, Brumback B, Robins JM. Marginal Structural Models to Estimate the Causal Effect of Zidovudine on the Survival of HIV-Positive Men. Epidemiology. 2000;11(5):561–70. Naimi AI, Cole SR, Kennedy EH. An introduction to g methods. Int J Epidemiol. 2016;46(2):756–62. Mansournia MA, Etminan M, Danaei G, Kaufman JS, Collins G. Handling time varying confounding in observational research. BMJ. 2017;359:j4587. Cole SR, Hernán MA. Constructing inverse probability weights for marginal structural models. Am J Epidemiol. 2008;168(6):656–64. Grafféo N, Latouche A, Geskus RB, Chevret S. Modeling time-varying exposure using inverse probability of treatment weights. Biom J. 2018;60(2):323–32. Hernán MA. The hazards of hazard ratios. Epidemiology. 2010;21(1):13–5. Greifer N, Stuart EA, editors. Choosing the Causal Estimand for Propensity Score Analysis of Observational Studies2021. Phillippo DM, Remiro-Azócar A, Heath A, Baio G, Dias S, Ades AE, et al. Effect modification and non-collapsibility together may lead to conflicting treatment decisions: A review of marginal and conditional estimands and recommendations for decision-making. Res Synthesis Methods. 2025;16(2):323–49. Harrell FE. Cox Proportional Hazards Regression Model. In: Harrell FE, editor. Regression Modeling Strategies: With Applications to Linear Models, Logistic Regression, and Survival Analysis. New York, NY: Springer New York; 2001. pp. 465–507. Pearce N, Vandenbroucke JP. Are Target Trial Emulations the Gold Standard for Observational Studies? Epidemiology. 2023;34(5):614–8. Fu EL. Target Trial Emulation to Improve Causal Inference from Observational Data: What, Why, and How? J Am Soc Nephrol. 2023;34(8):1305–14. Hansford HJ, Cashin AG, Jones MD, Swanson SA, Islam N, Douglas SRG, et al. Reporting of Observational Studies Explicitly Aiming to Emulate Randomized Trials: A Systematic Review. JAMA Netw Open. 2023;6(9):e2336023–e. Dickerman BA, García-Albéniz X, Logan RW, Denaxas S, Hernán MA. Emulating a target trial in case-control designs: an application to statins and colorectal cancer. Int J Epidemiol. 2020;49(5):1637–46. Additional Declarations No competing interests reported. Supplementary Files Additionalfile1.docx Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8831085","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":591038217,"identity":"ff152d41-d336-4956-b165-acd8a239b4d5","order_by":0,"name":"Anne J Huiberts","email":"","orcid":"","institution":"National Institute for Public Health and Environment (RIVM)","correspondingAuthor":false,"prefix":"","firstName":"Anne","middleName":"J","lastName":"Huiberts","suffix":""},{"id":591038218,"identity":"06d12755-89fd-4e57-be2c-da7bec068894","order_by":1,"name":"Bente Smagge","email":"","orcid":"","institution":"National Institute for Public Health and Environment (RIVM)","correspondingAuthor":false,"prefix":"","firstName":"Bente","middleName":"","lastName":"Smagge","suffix":""},{"id":591038219,"identity":"33eb591d-ac76-4d32-856f-5ba356dce5cf","order_by":2,"name":"Henri van Werkhoven","email":"","orcid":"","institution":"University Medical Center Utrecht (UMCU)","correspondingAuthor":false,"prefix":"","firstName":"Henri","middleName":"van","lastName":"Werkhoven","suffix":""},{"id":591038220,"identity":"5fd6f490-f2cc-4a9d-99d8-cf28b26c1dc6","order_by":3,"name":"Brechje de Gier","email":"","orcid":"","institution":"National Institute for Public Health and Environment (RIVM)","correspondingAuthor":false,"prefix":"","firstName":"Brechje","middleName":"","lastName":"de Gier","suffix":""},{"id":591038221,"identity":"2e7a4974-87b7-4127-90a2-6413ab4ab6d3","order_by":4,"name":"Susana Monge","email":"","orcid":"","institution":"Institute of Health Carlos III","correspondingAuthor":false,"prefix":"","firstName":"Susana","middleName":"","lastName":"Monge","suffix":""},{"id":591038222,"identity":"ad4dafea-9ae3-4402-9797-f68376cc1460","order_by":5,"name":"Hester E de Melker","email":"","orcid":"","institution":"National Institute for Public Health and Environment (RIVM)","correspondingAuthor":false,"prefix":"","firstName":"Hester","middleName":"E","lastName":"de Melker","suffix":""},{"id":591038223,"identity":"38220129-6b82-4649-b36b-1964993bfbab","order_by":6,"name":"Susan JM Hahné","email":"","orcid":"","institution":"National Institute for Public Health and Environment (RIVM)","correspondingAuthor":false,"prefix":"","firstName":"Susan","middleName":"JM","lastName":"Hahné","suffix":""},{"id":591038224,"identity":"4bf8a20f-0997-4ed5-9497-4259d3df11ea","order_by":7,"name":"Diederick E Grobbee","email":"","orcid":"","institution":"University Medical Center Utrecht (UMCU)","correspondingAuthor":false,"prefix":"","firstName":"Diederick","middleName":"E","lastName":"Grobbee","suffix":""},{"id":591038225,"identity":"095729c4-8b28-4e9b-b894-c5acee7bd17a","order_by":8,"name":"Janneke HHM van de Wijgert","email":"","orcid":"","institution":"University Medical Center Utrecht (UMCU)","correspondingAuthor":false,"prefix":"","firstName":"Janneke","middleName":"HHM van","lastName":"de Wijgert","suffix":""},{"id":591038226,"identity":"e3fb7509-3579-4301-bbe4-e4fdba461496","order_by":9,"name":"Susan van den Hof","email":"","orcid":"","institution":"National Institute for Public Health and Environment (RIVM)","correspondingAuthor":false,"prefix":"","firstName":"Susan","middleName":"van den","lastName":"Hof","suffix":""},{"id":591038227,"identity":"c0ea941d-e651-4ddf-9fe0-162b7d7008d1","order_by":10,"name":"Mirjam J Knol","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAy0lEQVRIiWNgGAWjYDACZgYGAwYwZm58AGYyMDcQq4Wx2QCihZGAFigAaWmTgLAJaNFt531Q8DOHwdicvbGtuqLgcDSDdCN+LWaH2Q0Me7cxmFn2HGy7ecbgcG6DzEFCWtgYDHi3MdgY3Ehsu9lgkJbbIJFIWIvhX6iWQqK1GANtMQNpYWwwsCFSi+w2CWODMwebJUFa2ghqOX+MzfDtNhvDDcebD35s+COR2y+RfACvFiBgA8aJBBKXkHogYH5AhKJRMApGwSgYyQAAff1DgdlPK7kAAAAASUVORK5CYII=","orcid":"","institution":"National Institute for Public Health and Environment (RIVM)","correspondingAuthor":true,"prefix":"","firstName":"Mirjam","middleName":"J","lastName":"Knol","suffix":""}],"badges":[],"createdAt":"2026-02-09 13:39:29","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8831085/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8831085/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":102862004,"identity":"d343bb7d-0eb5-47e9-b695-e9d43667c60a","added_by":"auto","created_at":"2026-02-17 16:14:55","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":781098,"visible":true,"origin":"","legend":"\u003cp\u003eVisualization of the target trial emulation approaches\u003c/p\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eA)\u003c/strong\u003e Single time zero approaches (approach 1 and 2); \u003cstrong\u003eB) \u003c/strong\u003eMultiple time zeros with covariate matching (approach 3); \u003cstrong\u003eC)\u003c/strong\u003e Multiple time zeros with inverse probability weighting (approach 4)\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-8831085/v1/be3a51933b98c7f9f203a936.png"},{"id":102862005,"identity":"9415fd5b-6f48-4e66-afbc-99a32d3c6f84","added_by":"auto","created_at":"2026-02-17 16:14:55","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":162406,"visible":true,"origin":"","legend":"\u003cp\u003eVaccine effectiveness estimates overall and by time since vaccination, by emulation approach\u003c/p\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003eIPW = inverse probability weighting\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-8831085/v1/9e9ced622f2e491a52d869a0.png"},{"id":104808622,"identity":"e5de6533-28fd-4079-b282-88aab0fe8716","added_by":"auto","created_at":"2026-03-17 12:39:04","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2494340,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8831085/v1/157981c0-006c-484b-aed3-e5c934e9ae37.pdf"},{"id":102862006,"identity":"8eee398e-3538-44cf-85ac-f77208afd178","added_by":"auto","created_at":"2026-02-17 16:14:56","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":2055088,"visible":true,"origin":"","legend":"","description":"","filename":"Additionalfile1.docx","url":"https://assets-eu.researchsquare.com/files/rs-8831085/v1/e597a4b12a0de2ebed81245b.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Comparison of methodological approaches in COVID-19 vaccine effectiveness estimation using observational data","fulltext":[{"header":"BACKGROUND","content":"\u003cp\u003eIn epidemiology, causal questions are ideally answered using an appropriately designed and conducted randomized trial. However, these trials are often not feasible, not ethical, or time-consuming. Furthermore, external validity is often limited. For COVID-19 vaccines, randomized trials were unable to address important questions regarding vaccine effectiveness (VE). In particular, VE in specific subpopulations, VE against newly emerging variants, duration of protection, and the benefit of booster vaccinations could not be assessed in trials (\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e). For these questions, public health policy relies on observational studies. However, the nature of observational data poses several challenges for causal inference, including confounding bias, selection bias, and immortal time bias (\u003cspan additionalcitationids=\"CR3\" citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eTo avoid making fundamental errors that can result in biased estimates and erroneous causal conclusions, one can use \u0026ldquo;target trial emulation\u0026rdquo; (TTE) (\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e). TTE is a framework for designing and analyzing observational studies that aim to estimate a causal effect. The systematic approach of TTE consists of two steps: 1) specify a hypothetical randomized trial that would answer the causal question, and 2) specify how each element of this trial was emulated using observational data (\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e). The protocol must include certain key elements that define the causal estimand, which is the description of the exact treatment effect under study, and the data analysis plan. The key elements are the eligibility criteria, treatment strategies, treatment assignment, the start and end of follow-up, outcomes, causal contrasts.\u003c/p\u003e \u003cp\u003eTo ensure that the observational study protocol properly emulates the design of a randomized trial that would answer the causal question, three components should be aligned at time zero (baseline): eligibility criteria are met, treatment strategies are assigned, and follow-up is started. For this, a key challenge is non-unique time zero, i.e. a single person may meet the eligibility criteria at multiple times, for instance throughout a vaccination campaign (\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e). To ensure alignment at time zero, one could choose a single time zero, for example the first eligible time, or use every eligible time as time zero. The latter requires emulating multiple sequential target \u0026lsquo;trials\u0026rsquo;, each of them with a different start of follow-up (\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eTTE has become increasingly popular in post-marketing vaccine evaluation in recent years, particularly following the introduction of COVID-19 vaccines (\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e). As TTE does not stipulate which specific analytical and statistical methods should be used, the VE studies that applied the TTE framework featured various analytical approaches and statistical methods (\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e). Nonetheless, single and multiple time zero(s) approaches can be distinguished and several commonly used methods can be identified, although variation remains between studies that use each of these. Most common for VE studies is sequential target trial emulation with matching of unvaccinated to vaccinated participants (\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e). Inverse probability weighting (IPW) is also frequently applied, both in single and multiple time zero(s) approaches (\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eIn the Netherlands, a COVID-19 booster vaccination campaign started on October 2, 2023. In this campaign, persons at increased risk for severe COVID-19 were eligible for vaccination: persons\u0026thinsp;\u0026ge;\u0026thinsp;60 years, health care workers, and medical risk groups. A monovalent mRNA vaccine targeting the SARS-CoV-2 Omicron XBB.1.5 subvariant (Comirnaty; BioNTech-Pfizer, Germany/United States) was used almost exclusively. VE of the Omicron XBB.1.5 booster in the Netherlands has been studied (\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e), but without explicitly emulating a randomized trial as specified by the TTE framework.\u003c/p\u003e \u003cp\u003eThe aim of this study was to assess real-world VE of the Omicron XBB.1.5 booster vaccination using the TTE framework and using different design and analysis approaches, explain their differences, and assess variability in the obtained estimates.\u003c/p\u003e"},{"header":"METHODS","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eStudy design\u003c/h2\u003e \u003cp\u003eThe VAccine Study COvid-19 (VASCO) is a population-based prospective cohort study that was initiated when COVID-19 vaccination was introduced in the Netherlands (\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e). The primary study objective is to estimate COVID-19 VE against SARS-CoV-2 infection. Study details are described elsewhere (\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e). Briefly, over 45,000 community-dwelling adults aged 18\u0026ndash;85 years were included in VASCO between May and December 2021. Participants are followed for five years after enrollment and data is collected using online questionnaires and self-collected fingerprick blood samples for SARS-CoV-2 serology, both at scheduled intervals. Questionnaires included, among others, questions on sociodemographic factors, COVID-19 vaccination, and SARS-CoV-2 testing results. Participants could also notify COVID-19 vaccinations and positive SARS-CoV-2 tests at any time during follow-up in the study application. SARS-CoV-2 self-tests are provided to participants to facilitate testing in case of COVID-19-like symptoms.\u003c/p\u003e \u003cp\u003e Written informed consent was obtained from all participants prior to enrolment into the study. The VASCO study is conducted in accordance with the principles of the Declaration of Helsinki and the study protocol was approved by the Medical Ethics Committee of the \u003cem\u003eStichting Beoordeling Ethiek Biomedisch Onderzoek\u003c/em\u003e, Assen, the Netherlands.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eStudy population, exposure, outcome and covariates\u003c/h3\u003e\n\u003cp\u003eThe \u003cem\u003estudy population\u003c/em\u003e included participants of the VASCO study who were eligible for the COVID-19 XBB.1.5 booster vaccination, specifically: persons aged over 60 years, healthcare workers and those belonging to a medical risk group. COVID-19 vaccination was based on vaccinations registered in the Dutch national COVID-19 vaccination Information and Monitoring System, supplemented with self-reported vaccinations. How these data sources were combined is described elsewhere (\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e). Booster vaccination (\u003cem\u003ethe exposure\u003c/em\u003e) was defined as receipt of any SARS-CoV-2 vaccination between 2 October 2023 and 2 April 2024, the study period. SARS-CoV-2 infection (\u003cem\u003ethe outcome\u003c/em\u003e) was defined as a self-reported positive SARS-CoV-2 test, or positive serology (seroconversion or a 4-fold increase in antibodies against the nucleoprotein, N-antibodies, detected in two consecutive serology samples). In case of a serology-based infection (i.e. no corresponding infection reported), the infection date was imputed between the before and after infection detection sampling dates using SARS-CoV-2 incidence based on reported positive SARS-CoV-2 tests in the VASCO population per age group per day, with higher probability for dates with higher incidence. \u003cem\u003eCovariates\u003c/em\u003e were age group (18\u0026ndash;39, 40\u0026ndash;59, 60\u0026ndash;69, 70\u0026ndash;85), sex (male, female), educational level, presence of a medical risk condition and time since prior infection. Educational level was classified as low (no or primary education), intermediate (secondary school or vocational training) or high (bachelor\u0026rsquo;s degree or higher), based on baseline data. Presence of a medical risk condition was defined as having one or more of following conditions: diabetes mellitus, lung disease or asthma, asplenia, cardiovascular disease, immune deficiency, cancer, liver disease, neurological disease, renal disease, organ or bone marrow transplantation. Time since prior infection was defined as time since most recent prior infection before 2 October 2023 (start study period), was based on self-reported positive SARS-CoV-2 tests and N-antibodies detected in serology samples (as described above), and was classified as no prior infection, a prior infection\u0026thinsp;\u0026ge;\u0026thinsp;1 year ago or a prior infection\u0026thinsp;\u0026lt;\u0026thinsp;1 year ago. The covariates age, presence of a medical risk condition and time since prior infection were time-varying. During follow-up, age increased and presence of a medical risk condition could switch from no to yes if a medical condition was reported in a follow-up questionnaire. Time since prior infection could switch from prior infection\u0026thinsp;\u0026lt;\u0026thinsp;1 year ago to prior infection\u0026thinsp;\u0026ge;\u0026thinsp;1 year ago. Participants could not switch from no prior infection to prior infection\u0026thinsp;\u0026lt;\u0026thinsp;1 year ago, as infections during follow-up were considered outcomes and such participants were no longer eligible for vaccination within the vaccination campaign period.\u003c/p\u003e\n\u003ch3\u003eMethodological approaches and statistical analyses\u003c/h3\u003e\n\u003cp\u003eA hypothetical trial was formulated to answer our research question. Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e describes each element of the hypothetical trial. Four approaches were used to emulate this hypothetical trial (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e; Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). In all approaches, loss to follow-up date was defined as the completion date of the participant\u0026rsquo;s most recent scheduled questionnaire, as information on vaccinations and infection might be incomplete thereafter. A sensitivity analysis was performed, in which loss to follow-up date was defined as the first date a scheduled questionnaire was not completed, i.e. date of protocol deviation.\u003c/p\u003e \u003cp\u003eAll analyses were performed with R software version 4.4.3, using packages ipw (version 1.2.1) (\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e), MatchIt (version 4.5.5), survival (version 3.7.0), and boot (version 1.3.31).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eTarget trial protocol for estimating vaccine effectiveness of XBB.1.5 booster vaccination against SARS-CoV-2 infection\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eProtocol component\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHypothetical target trial\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eEmulated target trial (single time zero with covariate adjustment)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eEmulated target trial (single time zero with inverse probability weighting)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eEmulated target trial (multiple time zeros with covariate matching)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eEmulated target trial (multiple time zeros with inverse probability weighting)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eEligibility criteria\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e- Age 18\u0026ndash;85 years\u003c/p\u003e \u003cp\u003e- Eligible for XBB.1.5 booster vaccination at baseline (HCW, 60+, medical risk; no infection in 90 days prior)\u003c/p\u003e \u003cp\u003e- Understanding Dutch\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e- Age 18\u0026ndash;85 years at time zero\u003c/p\u003e \u003cp\u003e- Eligible for XBB.1.5 booster vaccination (HCW, 60+, medical risk; no infection in 90 days prior)\u003c/p\u003e \u003cp\u003e- Understanding Dutch\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e- Age 18\u0026ndash;85 years at time zero\u003c/p\u003e \u003cp\u003e- Eligible for XBB.1.5 booster vaccination (HCW, 60+, medical risk; no infection in 90 days prior)\u003c/p\u003e \u003cp\u003e- Understanding Dutch\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e- Age 18\u0026ndash;85 years at time zero\u003c/p\u003e \u003cp\u003e- Eligible for XBB.1.5 booster vaccination (HCW, 60+, medical risk, no COVID-19 vaccine dose received since 2 October 2023, no infection in 90 days prior)\u003c/p\u003e \u003cp\u003e- Understanding Dutch\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e- Age 18\u0026ndash;85 years at time zero\u003c/p\u003e \u003cp\u003e- Eligible for XBB.1.5 booster vaccination (HCW, 60+, medical risk, no COVID-19 vaccine dose received since 2 October 2023, no infection in 90 days prior)\u003c/p\u003e \u003cp\u003e- Understanding Dutch\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eTreatment strategies\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1. Administer XBB.1.5 booster vaccine\u003c/p\u003e \u003cp\u003e2. Do not administer XBB.1.5 booster vaccine\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1. Administer XBB.1.5 booster vaccine at time zero\u003c/p\u003e \u003cp\u003e2. Do not ever administer XBB.1.5 booster vaccine\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1. Administer XBB.1.5 booster vaccine at time zero\u003c/p\u003e \u003cp\u003e2. Do not ever administer XBB.1.5 booster vaccine\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1. Administer XBB.1.5 booster vaccine\u003c/p\u003e \u003cp\u003e2. Do not administer XBB.1.5 booster vaccine\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e1. Administer XBB.1.5 booster vaccine\u003c/p\u003e \u003cp\u003e2. Do not administer XBB.1.5 booster vaccine\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eTreatment assignment\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eRandomization\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAdjustment for potential time-fixed and time-varying confounders by regression analysis. Potential confounders were age, sex, medical risk condition, educational level and time since prior infection.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eAdjustment for potential time-fixed and time-varying confounders by inverse probability weighting\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1:1 matching on calendar time (sequential trial) and other potential confounders at time zero. Potential confounders were age, sex, medical risk condition, educational level and time since prior infection.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eMatching on calendar time (sequential trial) + adjustment for potential confounders by inverse probability weighting at time zero.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eFollow-up\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTime zero is when eligibility criteria for inclusion are met, treatment is assigned, and follow-up begins. Follow-up is from time zero to outcome, loss-to-follow up, death, or administrative end of follow-up, whichever occurred first.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eTime zero is the start of the vaccination roll-out on 2 October 2023 for the unvaccinated group and time of vaccination in the vaccinated group. Follow-up is from time zero to outcome, loss to follow-up or administrative end of follow-up, whichever occurred first.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eTime zero is the start of the vaccination roll-out on 2 October 2023 for the unvaccinated group and time of vaccination in the vaccinated group. Follow-up is from time zero to outcome, loss to follow-up or administrative end of follow-up, whichever occurred first.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eTime zero is date of vaccination or equivalent index date of the matched control who did not receive vaccination on or before that date. Follow-up is from time zero to outcome, loss to follow-up, receipt of a vaccine-dose by an individual in the unvaccinated group (with concurrent censoring of the vaccinated member of the pair), or administrative end of follow-up, whichever occurred first.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eTime zero is date of vaccination or equivalent index week of any control who did not receive vaccination on or before that date. Follow-up is from time zero to outcome, loss to follow-up, receipt of a vaccine-dose by an individual in the unvaccinated group, or administrative end of follow-up, whichever occurred first.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eOutcome\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSARS-CoV-2 infection between treatment allocation and end of study\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSARS-CoV-2 infection (self-reported or serology-based) between 2 October 2023 and 2 April 2024\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSARS-CoV-2 infection (self-reported or serology-based) between 2 October 2023 and 2 April 2024\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eSARS-CoV-2 infection (self-reported or serology-based) between index date and 2 April 2024\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eSARS-CoV-2 infection (self-reported or serology-based) between index week and 2 April 2024\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eCausal contrast of interest\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePer-protocol, individuals in the control arm were censored upon receiving vaccination\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePer-protocol\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003ePer-protocol\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePer-protocol\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003ePer-protocol\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eStatistical analysis\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCumulative incidence (risk) curves of SARS-COV-2 infection using the Kaplan-Meier estimator; vaccine effectiveness\u0026thinsp;=\u0026thinsp;100% * (1 \u0026ndash; risk ratio (RR))\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eCox regression model with calendar time as underlying time scale and time-varying exposure, with regression adjustment for (time-varying) covariates; vaccine effectiveness\u0026thinsp;=\u0026thinsp;100% * (1 \u0026ndash; hazard ratio (HR))\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eCox regression model with time-varying exposure, weighted by time-varying IPTWs and IPCWs; vaccine effectiveness\u0026thinsp;=\u0026thinsp;100% * (1 \u0026ndash; hazard ratio (HR)) and robust standard error estimate\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eCumulative incidence (risk) curves using the Kaplan-Meier estimator; vaccine effectiveness\u0026thinsp;=\u0026thinsp;100% *(1 \u0026ndash; risk ratio (RR)) and bootstrapping to compute 95%CI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eCox regression model stratified by trial, adjusting for covariates using IPTW and IPCW; vaccine effectiveness\u0026thinsp;=\u0026thinsp;100% * (1 \u0026ndash; hazard ratio (HR)) and bootstrapping to compute 95%CI\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"6\"\u003eHCW\u0026thinsp;=\u0026thinsp;healthcare worker; IPTW\u0026thinsp;=\u0026thinsp;inverse probability treatment weight(ing); IPCW\u0026thinsp;=\u0026thinsp;inverse probability censoring weight(ing); RR\u0026thinsp;=\u0026thinsp;risk ratio; HR\u0026thinsp;=\u0026thinsp;hazard ratio\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003e1. Single time zero with regression adjustment\u003c/span\u003e \u003c/p\u003e \u003cp\u003eThis first approach is a time-dependent Cox model, which is widely used for estimating VE in cohort studies with time-to-event data, but not usually considered part of the TTE canon (\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e). Probably this approach is hardly used within the TTE framework because it is known to produce biased estimates in the presence of time-varying confounders that are themselves affected by the treatment (\u003cspan additionalcitationids=\"CR18\" citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e). However, treatment-confounder feedback by the measured time-varying confounders in our model seems unlikely.\u003c/p\u003e \u003cp\u003eThe first eligible time was chosen as time zero. Participants were eligible for XBB.1.5 booster vaccination at the start of the vaccination roll-out on 2 October 2023 or three months after their most recent prior SARS-CoV-2 infection, whichever occurred last. Follow-up started at time zero. Treatment assignment was not aligned at time zero, because participants could receive a vaccination at any time between time zero and the end of the vaccination campaign. To account for this, vaccination status was coded as a time-varying variable that transitioned from \u0026lsquo;unvaccinated\u0026rsquo; to \u0026lsquo;vaccinated\u0026rsquo; seven days after administration of the XBB.1.5 booster vaccination and remained \u0026lsquo;vaccinated\u0026rsquo; thereafter. The seven days after vaccination administration were excluded from the analysis. Follow-up ended on the date of the first SARS-CoV-2 infection, loss to follow-up or the end of the study period on 2 April 2024, whichever came first.\u003c/p\u003e \u003cp\u003eA Cox proportional hazards model with calendar time as the underlying time scale was used to estimate VE. The estimate was adjusted for potential confounding by adding age group, sex, medical risk condition, educational level, and time since prior infection as covariables to the Cox regression model. Age group, medical risk condition and time since prior infection were time-varying covariates. VE was calculated as 100% \u0026times; (1\u0026thinsp;\u0026minus;\u0026thinsp;hazard ratio). To evaluate VE by time since vaccination, the exposure was categorized into six-week periods with \u0026lsquo;unvaccinated\u0026rsquo; as the reference category.\u003c/p\u003e \u003cp\u003e \u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003e2. Single time zero with inverse probability weighting\u003c/span\u003e \u003c/p\u003e \u003cp\u003eRobins\u0026rsquo; generalized methods (G-methods) are a class of methods that, unlike stratification or standard regression (e.g. the time-dependent Cox model used in our first approach), appropriately adjusts for confounding when treatment-confounder feedback exists(\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e). Among available G-methods, models using time-varying inverse probability weights (IPWs), also known as marginal structural models, are most commonly used for our type of data and causal question (\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eEligibility criteria, follow-up time and the exposure were defined in the same manner as in the first approach. However, we used a Cox proportional hazards model with inverse probability of treatment weighting (IPTW) to adjust for confounding and inverse probability of censoring weighting (IPCW) to adjust for informative censoring due to non-random loss to follow-up (\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e). The final weights were obtained by multiplying the stabilized time-varying IPTWs and IPCWs. The weighting procedures are described in detail in the \u003cb\u003eSupplementary file, Statistical analyses\u003c/b\u003e.\u003c/p\u003e \u003cp\u003eFinally, a Cox proportional hazards model, weighted by the time-varying stabilized weights and additionally adjusted for the time-fixed confounders, was fit. This adjustment was required because stabilized weights do not adjust for time-fixed covariates (\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e). To account for person-level clustering induced by weighting, a robust standard error estimate was obtained. VE by time since vaccination was evaluated by categorizing the exposure into six-week periods with \u0026lsquo;unvaccinated\u0026rsquo; as the reference category.\u003c/p\u003e \u003cp\u003e \u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003e3. Multiple time zeros with covariate matching\u003c/span\u003e \u003c/p\u003e \u003cp\u003eIn the third approach, we used every eligible time as time zero, i.e. multiple time zeros. For each calendar day of the vaccination period, persons who met the eligibility criteria on that day, and who had not been vaccinated before that day, were identified and classified as either having or not having received XBB.1.5 booster vaccination on that day. This way, a sequence of \u0026lsquo;trials\u0026rsquo; was created, with a new trial starting on each day. Each person who received a vaccination was matched exactly to a randomly selected control (with replacement) on age group, sex, medical risk condition, educational level, and time since prior infection. Eligible individuals could be selected as unvaccinated controls repeatedly up to the day before they were vaccinated. For each matched pair, follow-up started on the day of vaccine administration of the vaccinated member of the pair and ended at the date of the first reported SARS-CoV-2 infection, the vaccination date of the control, loss to follow-up or the end of the study period, whichever came first.\u003c/p\u003e \u003cp\u003eCumulative incidence curves of self-reported positive SARS-CoV-2 test were constructed using the Kaplan-Meier estimator. Both individuals of a matched pair had to be still at risk by day 8 of follow-up. The risk of SARS-CoV-2 infection more than 7 days after vaccination was compared with the risk among their matched controls. VE was calculated as 100% \u0026times; (1 \u0026ndash; risk ratio). 95% confidence intervals (CIs) were computed using bootstrapping (1000 iterations) of the matched dataset, to account for sequential trials not being independent. We estimated VE for each six-week period since vaccination by including only pairs that were not censored before the start of the period and censoring them at the end of the period.\u003c/p\u003e \u003cp\u003e \u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003e4. Multiple time zeros with inverse probability weighting\u003c/span\u003e \u003c/p\u003e \u003cp\u003eSimilar to the third approach, in the fourth approach all eligible times were used as time zero. When selecting trials per day, estimation of the IPCWs resulted in non-positivity in some of the groups due to the absence of censoring. Therefore, we created trials per week, rather than per day. In each week of the study period, we identified persons who had not been vaccinated before that week (index week), who met the eligibility criteria, and who still met the eligibility criteria 7 days later at the start of follow-up. All participants who did not receive XBB.1.5 booster vaccination in that week were classified as unvaccinated. All participants who did receive XBB.1.5 booster vaccination sometime during that week were classified as vaccinated. Eligible individuals were enrolled as unvaccinated in each trial up to the week before their vaccination, but once they received vaccination they could not enroll in subsequent trials. We excluded weeks in which\u0026thinsp;\u0026lt;\u0026thinsp;20 individuals received a XBB.1.5 booster vaccination or \u0026lt;\u0026thinsp;20 individuals remained unvaccinated to avoid model separation when estimating the probability of censoring. Follow-up started one week after each index week (i.e. 7 to 13 days after an individual's vaccination date) and was continued until the week of the first reported SARS-CoV-2 infection, loss to follow-up or the end of the study period, whichever came first. Unvaccinated controls were censored at the end of the week of their vaccination. IPW was used to adjust for confounding (IPTW) and informative censoring due to loss to follow-up (IPCW\u003csub\u003eLTFU\u003c/sub\u003e) and due to protocol deviation, i.e. vaccination of unvaccinated controls during follow-up (IPCW\u003csub\u003ePD\u003c/sub\u003e) (\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e). The computation of the IPWs is described in the \u003cb\u003eSupplementary file, Statistical analyses\u003c/b\u003e. The final weights were obtained by multiplying the stabilized IPTWs, IPTWs\u003csub\u003ePD\u003c/sub\u003e and IPCWs\u003csub\u003eLTFU\u003c/sub\u003e.\u003c/p\u003e \u003cp\u003eFinally, weighted Cox models stratified by trial were fit. VE was calculated as 100% \u0026times; (1\u0026thinsp;\u0026minus;\u0026thinsp;hazard ratio). The weighted model was adjusted for the time-fixed confounders that were included in the numerator model of the IPCWs. The 95% CIs were computed using bootstrapping (1000 iterations). To evaluate VE by time since vaccination, the exposure was categorized into six-week periods with \u0026lsquo;unvaccinated\u0026rsquo; as the reference category.\u003c/p\u003e"},{"header":"RESULTS","content":"\u003cp\u003e \u003cb\u003e1. Single time zero with regression adjustment\u003c/b\u003e \u003c/p\u003e \u003cp\u003eIn total, 24,288 VASCO participants were eligible for XBB.1.5 booster vaccination and were included in the analysis. Of those, 15,300 (63%) received vaccination during the study period (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). Participants who received vaccination were older, more often male, more often had a high education level, more often had a medical risk condition, and less often had experienced a prior infection. During 265,660 vaccinated and 202,961 unvaccinated person-weeks, 2,900 and 3,318 infections occurred, respectively. Risk of infection was higher among unvaccinated participants (\u003cb\u003eSupplementary file, Figure \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e\u003c/b\u003e). Furthermore, incidence was highest among those without prior infection and lowest among those with a prior infection in the past year (\u003cb\u003eSupplementary file, Figure S2\u003c/b\u003e). Overall VE against infection was 30% (95% CI 26\u0026ndash;34) (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). VE decreased as time since vaccination increased, being 35% (30\u0026ndash;39), 24% (18\u0026ndash;30) and 8% (-9-23) at 1\u0026ndash;6, 7\u0026ndash;12 and 13\u0026ndash;18 weeks post-vaccination, respectively (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e; \u003cb\u003eSupplementary file, Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e\u003c/b\u003e).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eCharacteristics of participants included in single time zero with regression adjustment analysis and multiple time zeros with covariate matching analysis, the Netherlands, 2 October 2023\u0026ndash;2 April 2024\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"7\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colspan=\"3\" nameend=\"c4\" namest=\"c2\"\u003e \u003cp\u003eSingle time zero with regression adjustment\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e \u003cp\u003eMultiple time zeros with covariate matching\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTotal\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNo XBB.1.5 booster vaccination\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eYes XBB.1.5 booster vaccination\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eTotal\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eNo XBB.1.5 booster vaccination\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eYes XBB.1.5 booster vaccination\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;24288\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;8988\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;15300\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;25834\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;12917\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;12917\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"7\" nameend=\"c7\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eAge group\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e18\u0026ndash;39\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1083 (4.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e822 (9.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e261 (1.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e494 (1.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e247 (1.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e247 (1.9%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e40\u0026ndash;59\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e4094 (16.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2888 (32.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1206 (7.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e2280 (8.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e1140 (8.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e1140 (8.8%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e60\u0026ndash;69\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e13457 (55.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e4137 (46%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e9320 (60.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e16284 (63%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e8142 (63%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e8142 (63%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e70\u0026ndash;85\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e5654 (23.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1141 (12.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e4513 (29.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e6776 (26.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e3388 (26.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e3388 (26.2%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"7\" nameend=\"c7\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eSex\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFemale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e15008 (61.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e6293 (70%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e8715 (57%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e14974 (58%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e7487 (58%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e7487 (58%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e9276 (38.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2693 (30%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e6583 (43%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e10860 (42%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e5430 (42%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e5430 (42%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOther\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e4 (0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2 (0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2 (0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0 (0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0 (0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0 (0%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"7\" nameend=\"c7\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eEducation level\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHigh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e13418 (55.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e4677 (52%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e8741 (57.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e14542 (56.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e7271 (56.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e7271 (56.3%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIntermediate\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e6835 (28.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2841 (31.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e3994 (26.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e6902 (26.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e3451 (26.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e3451 (26.7%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLow\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3878 (16%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1415 (15.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2463 (16.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e4230 (16.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e2115 (16.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e2115 (16.4%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOther\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e157 (0.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e55 (0.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e102 (0.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e160 (0.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e80 (0.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e80 (0.6%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"7\" nameend=\"c7\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eMedical risk condition\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e11384 (46.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e4099 (45.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e7285 (47.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e11952 (46.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e5976 (46.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e5976 (46.3%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"7\" nameend=\"c7\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eTime since prior infection\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;1 year ago\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3033 (12.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1553 (17.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1480 (9.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e9116 (35.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e4558 (35.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e4558 (35.3%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026gt;\u0026thinsp;1 year ago\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e17857 (73.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e6394 (71.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e11463 (74.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e12884 (49.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e6442 (49.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e6442 (49.9%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNo prior infection\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3398 (14%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1041 (11.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2357 (15.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e3834 (14.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e1917 (14.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e1917 (14.8%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"7\"\u003e\u003csup\u003ea\u003c/sup\u003e For the purpose of this table, vaccination status and all covariates were determined at the end of follow-up\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eIPW\u0026thinsp;=\u0026thinsp;inverse probability weighting\u003c/p\u003e \u003cp\u003e \u003cb\u003e2. Single time zero with inverse probability weighting\u003c/b\u003e \u003c/p\u003e \u003cp\u003eIn the second approach, the same participants were included as in approach 1. None of the estimated stabilized IPWs was \u0026gt;\u0026thinsp;10, thus truncation to 10 for extreme IPWs was not applied. Assessment of the balance of the time-varying variables between the vaccinated and unvaccinated population is shown in the \u003cb\u003eSupplementary file, Figure S3)\u003c/b\u003e. Overall VE against infection was 28% (24\u0026ndash;32) (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). VE decreased as time since vaccination increased, with 33% (28\u0026ndash;38), 22% (\u003cspan additionalcitationids=\"CR16 CR17 CR18 CR19 CR20 CR21 CR22 CR23 CR24 CR25 CR26 CR27\" citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e) and 2% (-18-20) at 1\u0026ndash;6, 7\u0026ndash;12 and 13\u0026ndash;18 weeks post-vaccination, respectively (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e; \u003cb\u003eSupplementary file, Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e\u003c/b\u003e).\u003c/p\u003e \u003cp\u003e \u003cb\u003e3. Multiple time zeros with covariate matching\u003c/b\u003e \u003c/p\u003e \u003cp\u003eOn 95 dates at least one person was vaccinated and could be matched to a (yet) unvaccinated individual. The number of matched pairs ranged between 1 and 348 per trial (\u003cb\u003eSupplementary file, Figure S4\u003c/b\u003e). In total, 12,917 vaccinated participants (out of 15,300) were matched to unvaccinated participants (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e), in whom 1,384 and 1,872 infections occurred, respectively (\u003cb\u003eSupplementary file, Figure S5\u003c/b\u003e). Most characteristics of included participants were comparable to characteristics of vaccinated individuals in the entire study population (as included in approach 1). Matched vaccinated individuals were less often in the oldest age group (\u0026ge;\u0026thinsp;70 years), possibly because vaccination coverage in this age group was high and few unvaccinated were available for matching. Overall VE against infection was 28% (24\u0026ndash;32) (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). VE decreased as time since vaccination increased with 32% (26\u0026ndash;37), 33% (28\u0026ndash;38) and 14% (-6-30) at 1\u0026ndash;6, 7\u0026ndash;12 and 13\u0026ndash;18 weeks post-vaccination, respectively (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e; \u003cb\u003eSupplementary file, Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e\u003c/b\u003e).\u003c/p\u003e \u003cp\u003e \u003cb\u003e4. Multiple time zeros with inverse probability weighting\u003c/b\u003e \u003c/p\u003e \u003cp\u003eIn 12 weeks, a trial started. Eight weeks were excluded because of \u0026lt;\u0026thinsp;20 individuals in the vaccinated and/or unvaccinated group; the week of Christmas and seven weeks after the official end of the vaccination campaign. The number of participants ranged between 5,722 and 20,871 per trial. In total, 141,638 non-unique individuals were included. After weighting, the vaccinated and unvaccinated population were comparable (\u003cb\u003eSupplementary file, Figure S6\u003c/b\u003e). Truncation of large IPWs was only required for \u0026lt;\u0026thinsp;0.01% of the estimated IPWs. In total, 2,744 and 19,886 infections occurred in vaccinated and unvaccinated participants (\u003cb\u003eSupplementary file, Figure S7\u003c/b\u003e). Overall VE against infection was 27% (24\u0026ndash;31) (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). VE decreased as time since vaccination increased, with 33% (29\u0026ndash;38), 24% (\u003cspan additionalcitationids=\"CR19 CR20 CR21 CR22 CR23 CR24 CR25 CR26 CR27 CR28\" citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e) and 5% (-13-22) at 1\u0026ndash;6, 7\u0026ndash;12 and 13\u0026ndash;18 weeks post-vaccination, respectively (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e; \u003cb\u003eSupplementary file, Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e\u003c/b\u003e).\u003c/p\u003e \u003cp\u003eThe sensitivity analysis in which loss to follow-up date was defined as the first day that a scheduled questionnaire was not completed resulted in negligible differences in the estimates of the four approaches (\u003cb\u003eSupplementary file, Figure S8\u003c/b\u003e).\u003c/p\u003e"},{"header":"DISCUSSION","content":"\u003cp\u003eIn this study we used four TTE approaches to estimate VE of the XBB.1.5 booster vaccination against SARS-CoV-2 infection. The four approaches yielded very similar overall VE estimates ranging between 27 and 30%, with considerably overlapping 95% confidence intervals. Somewhat larger deviations were observed between estimates of VE by time since vaccination. In particular, using the approach with multiple time zeros and covariate matching, we found a somewhat smaller decrease in VE point estimates over time compared to the other approaches.\u003c/p\u003e \u003cp\u003eA potential explanation for this is that period-specific estimates may be affected by differential depletion of susceptibles (\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e). Only participants who remained event-free (and did not drop out) until a certain time, will be included in the subsequent period for which the hazard ratio (HR) is estimated. Individuals who are at increased risk of infection, e.g. due to weaker immunity or greater exposure, are likely to get infected earlier. Therefore, those at increased risk are \u0026lsquo;removed\u0026rsquo; from the at-risk population at a higher rate, leaving individuals who are on average less susceptible or have less exposure for the estimation of VE at a longer time since vaccination. As the unvaccinated group is not protected by vaccination, infections are likely to occur at a faster rate in this group, leading to more rapid depletion of susceptibles in the unvaccinated compared to the vaccinated group. Due to this differential depletion of susceptibles, VE based on period-specific HRs may be artificially low long after vaccination, as the vaccinated group is compared to a progressively less at-risk unvaccinated group. In our approach using multiple time zeros with covariate matching, both the vaccinated and unvaccinated member of each matched pair were required to stay event-free (and not drop out) until the start of the time interval since vaccination, thereby vaccinated and unvaccinated individuals may be more comparable with respect to susceptibility of infection given the measured confounders. Compared to the approaches featuring Cox regression, the multiple time zeros with covariate matching approach thereby may be less affected by spurious waning caused by differential depletion of susceptibles. It might therefore be a preferred approach for investigating the effect of time since vaccination on the VE.\u003c/p\u003e \u003cp\u003eThe multiple time zeros with covariate matching approach (approach 3) also differs from the other three approaches in two other ways. Firstly, approach 3 estimates the average treatment effect on the treated (ATT), or, strictly speaking, the average treatment effect in the overlap population (ATO), as opposed to the overall average treatment effect (ATE). The ATE estimates the effect of treatment across the entire population regardless of whether individuals actually received treatment. It is the difference between the expected outcome if all individuals were treated and the expected outcome if none were treated. The ATT focuses on the effect only within those individuals that actually received treatment. This is a consequence of the matching in approach 3. As some vaccinated individuals who could not be matched were removed, actually approach 3 estimates the ATO: the average treatment in the overlap/equipoise population, i.e. the population with comparable chances of being (un)treated based on the matching variables (\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e). Differences between the ATE and ATT or ATO arise when the treatment effect varies between population subgroups (i.e., interactions between the treatment and individual characteristics) and the treated and untreated differ in these characteristics. Our data indeed showed differences in characteristics of vaccinated (treated) and unvaccinated (untreated) individuals, therefore at least the second requirement holds true. Secondly, approach 3 estimates a marginal effect, whereas the other approaches estimate a conditional effect. Conditional effects control for the values of other variables, making them context-dependent. Marginal effects summarize the average effect of a variable across the entire distribution of other variables of the model (\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e). For collapsible measures of effect, like the risk ratio estimated in approach 3, the marginal and conditional effects are the same in the absence of interactions between the exposure and covariates. For non-collapsible measures of effect, like the hazard ratio estimated in approaches 1, 2 and 4, the conditional effect differs from the marginal effect if the covariates that are conditioned on are predictive of the outcome, even if there is no association with the exposure (i.e. confounding) or effect modification (\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e). Because the estimand of the multiple time zeros with covariate matching approach differs from the other approaches in these ways, the effect estimates are not directly comparable and, although this is not evident in our results, their numeric value may differ.\u003c/p\u003e \u003cp\u003eThe methods used have varying assumptions and limitations. First, Kaplan-Meier (approach 3) is a non-parametric method with no assumptions about the distribution of the unobserved failure times due to administrative censoring. In contrast, the semi-parametric Cox regression model (used in approaches 1, 2 and 4), while not assuming a particular distribution for failure times or hazards either, imposes a priori restrictions on the relation between the baseline hazard and the hazard under other combinations of covariate values (\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e). Second, the approach using a single time zero with Cox regression is biased in the presence of treatment-confounder feedback, i.e. when a time-varying covariate that is affected by vaccination status is adjusted for (\u003cspan additionalcitationids=\"CR18\" citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e). Studies that measured such covariates and want to appropriately adjust for them should avoid this approach. Third, using multiple time zeros with covariate matching results in a smaller number of participants and outcomes due to matching in case not every vaccinated individual can be matched to an unvaccinated individual and due to censoring of the entire pair when the unvaccinated received vaccination. This may result in wider CIs, specifically with increasing time since vaccination. By matching, this method also eliminates the possibility of studying the effect of individual confounders in sensitivity analyses (\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e). Finally, the multiple time zeros with IPW approach is computationally intensive and requires a lot of data. In our study, we created trials per week rather than per day because of non-positivity, but we expect limited effect of this on our results. It was impossible to estimate the IPCW\u003csub\u003eLTFU\u003c/sub\u003e by trial because this type of censoring was too rare, so we estimated the IPCW\u003csub\u003eLTFU\u003c/sub\u003e in the dataset including all trials. This assumes that the associations between the covariates and the probability of censoring due to loss to follow-up are the same across trials.\u003c/p\u003e \u003cp\u003eThe TTE framework is useful to systematically think through every step of the design and analysis of an observational study aiming to answer a causal question. This way, it helps researchers to recognize, appropriately deal with, and communicate about methodological pitfalls and potential biases inherent to the use of observational data for causal inference (\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e). Nonetheless, the TTE framework has several caveats. First, the benefit of the framework depends on how rigorously it is applied and this needs to be fully reported for readers to understand and assess. One systematic review reported that only 58% of studies explicitly aiming to emulate a target trial completely report how the target trial was emulated (\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e). Second, applying the TTE framework can still lead to multiple \u0026lsquo;as correct as possible\u0026rsquo; study designs that may yield comparable estimates, as was the case in our study. On the other hand, if the estimates do differ between approaches, it may be difficult to uncover which one is (most) correct. Third, studies designed using the TTE framework may still be flawed, because TTE does not remove the limitations of the data, such as use of self-reported data, or unmeasured confounding (\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e). For example, in this study, we had to assume non-interference (\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e) and could only assess the direct effect of vaccination. We also did not have sufficient participants to estimate VE by subgroup eligible for vaccination, i.e. among persons aged\u0026thinsp;\u0026ge;\u0026thinsp;60 and among younger clinically vulnerable individuals. Fourth, (too) strict adherence to the TTE framework can be contentious. Overzealous application/interpretation of the TTE framework would suggest that studies should have a prospective cohort design in order to mimic an RCT, but case-control studies are valuable too and can indeed also be designed within the TTE framework(\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e). For COVID-19 VE specifically, test-negative case-control designs have been advantageous to reduce bias introduced by health-seeking behavior. It is also debatable whether the single time zero approaches can be considered properly emulated target trials within the TTE framework, because treatment assignment is not aligned at time zero. However, these approaches handle this by allowing vaccination status to be time-varying, so it can be argued that the optimal analytical approach does not necessarily need to align eligibility, treatment assignment and the start of follow-up to obtain an unbiased estimate. Indeed, the similarity of our estimates suggests that these approaches do not provide inherently less reliable effect estimates. In short, TTE is neither a condition for nor a guarantee of unbiased causal inference, but it helps to avoid methodological pitfalls and biases.\u003c/p\u003e"},{"header":"CONCLUSIONS","content":"\u003cp\u003eIn this study, the hypothetical target trial to estimate the VE of the XBB.1.5 booster vaccination was emulated using four different design and analysis approaches. Although we cannot know the true VE, the VE estimates were similar across the four approaches. To the extent that the four approaches have different assumptions and estimands, we can conclude that the VE estimates in this study are not sensitive to these differences, except possibly when estimating VE over time since vaccination. This study highlights that observational data can be analyzed in multiple valid ways to answer a causal research question. The choice of approach should depend on the estimand of interest (e.g. ATT vs. ATE, marginal vs. conditional effect), the plausibility of the assumptions of various approaches (e.g. proportional hazards, no treatment-confounder feedback) and practical and technical feasibility (e.g. computational power, data quantity).\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cp\u003eATE = average treatment effect\u003c/p\u003e\n\u003cp\u003eATO = average treatment effect in the overlap population\u003c/p\u003e\n\u003cp\u003eATT = average treatment effect on the treated\u003c/p\u003e\n\u003cp\u003eCOVID-19 = Coronavirus Disease 2019\u003c/p\u003e\n\u003cp\u003eG-methods = generalized methods\u003c/p\u003e\n\u003cp\u003eHR = hazard ratio\u003c/p\u003e\n\u003cp\u003eIPCW = inverse probability of censoring weight(ing)\u003c/p\u003e\n\u003cp\u003eIPTW = inverse probability of treatment weight(ing)\u003c/p\u003e\n\u003cp\u003eIPW = inverse probability weight(ing)\u003c/p\u003e\n\u003cp\u003eLTFU = loss to follow-up\u003c/p\u003e\n\u003cp\u003ePD = protocol deviation\u003c/p\u003e\n\u003cp\u003eSARS-CoV-2 = Severe Acute Respiratory Syndrome Coronavirus 2\u003c/p\u003e\n\u003cp\u003eTTE = target trial emulation\u003c/p\u003e\n\u003cp\u003eVASCO = VAccine Study COvid-19\u003c/p\u003e\n\u003cp\u003eVE = vaccine effectiveness\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWritten informed consent was obtained from all participants prior to enrolment into the study. The VASCO study is conducted in accordance with the principles of the Declaration of Helsinki and the study protocol was approved by the Medical Ethics Committee of the \u003cem\u003eStichting Beoordeling Ethiek Biomedisch Onderzoek\u003c/em\u003e, Assen, the Netherlands.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and materials\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAnonymized data reported from this study can be obtained from the corresponding author upon request. The dataset may include individual data and a data dictionary will be provided. Data requests should include a proposal for the planned analyses. Data transfer will require a signed data sharing agreement. Code for data processing, statistical analysis, figures, and tables can be found at GitHub (https://github.com/b-smagge/TTE_COVID_VE).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare that they have no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis work was funded by the Ministry of Health, Welfare and Sports of the Netherlands.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthors\u0026rsquo; contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll authors have read and approved the final manuscript. AH, BS, HvW, BdG and MK conceptualized and designed the study. AH and BS performed data cleaning and data analysis. AH and BS drafted the manuscript. HvW, BdG, SM, HdM, SH, DG, JvdW, SvdH and MK critically reviewed the paper.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe would like to thank Maarten Schipper for his valuable statistical insights, which greatly contributed to the interpretation of the results in this work.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eHulme WJ, Williamson E, Horne EMF, Green A, McDonald HI, Walker AJ, et al. Challenges in Estimating the Effectiveness of COVID-19 Vaccination Using Observational Data. Ann Intern Med. 2023;176(5):685\u0026ndash;93.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKuehne F, Arvandi M, Hess LM, Faries DE, Matteucci Gothe R, Gothe H, et al. Causal analyses with target trial emulation for real-world evidence removed large self-inflicted biases: systematic bias assessment of ovarian cancer treatment effectiveness. J Clin Epidemiol. 2022;152:269\u0026ndash;80.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMaringe C, Benitez Majano S, Exarchakou A, Smith M, Rachet B, Belot A, et al. Reflection on modern methods: trial emulation in the presence of immortal-time bias. Assessing the benefit of major surgery for elderly lung cancer patients using observational data. Int J Epidemiol. 2020;49(5):1719\u0026ndash;29.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHern\u0026aacute;n MA, Sauer BC, Hern\u0026aacute;ndez-D\u0026iacute;az S, Platt R, Shrier I. Specifying a target trial prevents immortal time bias and other self-inflicted injuries in observational analyses. J Clin Epidemiol. 2016;79:70\u0026ndash;5.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHern\u0026aacute;n MA, Robins JM. Using Big Data to Emulate a Target Trial When a Randomized Trial Is Not Available. Am J Epidemiol. 2016;183(8):758\u0026ndash;64.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHern\u0026aacute;n MA, Wang W, Leaf DE. Target Trial Emulation: A Framework for Causal Inference From Observational Data. JAMA. 2022;328(24):2446\u0026ndash;7.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHern\u0026aacute;n MA, Robins JM. Causal Inference: What If: Boca Raton. Chapman \u0026amp; Hall/CRC.; 2020.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKomura T, Watanabe M, Shioda K. Exploring the Application of Target Trial Emulation in Vaccine Evaluation. Scoping Rev medRxiv. 2024:2024.07.26.24311066.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZuo H, Yu L, Campbell SM, Yamamoto SS, Yuan Y. The implementation of target trial emulation for causal inference: a scoping review. J Clin Epidemiol. 2023;162:29\u0026ndash;37.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGravenstein S, DeVone F, Oyebanji OA, Abul Y, Cao Y, Chan PA, et al. Durability of immunity and clinical protection in nursing home residents following bivalent SARS-CoV-2 vaccination. EBioMedicine. 2024;105:105180.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMcConeghy KW, Bardenheier B, Huang AW, White EM, Feifer RA, Blackman C, et al. Infections, Hospitalizations, and Deaths Among US Nursing Home Residents With vs Without a SARS-CoV-2 Vaccine Booster. JAMA Netw Open. 2022;5(12):e2245417\u0026ndash;e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuiberts AJ, Hoeve CE, de Gier B, Cremer J, van der Veer B, de Melker HE, et al. Effectiveness of Omicron XBB.1.5 vaccine against infection with SARS-CoV-2 Omicron XBB and JN.1 variants, prospective cohort study, the Netherlands, October 2023 to January 2024. Eurosurveillance. 2024;29(10):2400109.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuiberts AJ, Hoeve CE, Kooijman MN, de Melker HE, Hahn\u0026eacute; SJ, Grobbee DE, et al. Cohort profile: an observational population-based cohort study on COVID-19 vaccine effectiveness in the Netherlands - the VAccine Study COVID-19 (VASCO). BMJ Open. 2024;14(10):e085388.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003evan der Wal WM, Geskus RB. ipw: An R Package for Inverse Probability Weighting. J Stat Softw. 2011;43(13):1\u0026ndash;23.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEuropean Centre for Disease Prevention and Control. Protocol for a COVID-19 vaccine effectiveness study using health data registries, v.2.0. Stockholm: ECDC; 2024.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHoffman KL, Schenck EJ, Satlin MJ, Whalen W, Pan D, Williams N, et al. Comparison of a Target Trial Emulation Framework vs Cox Regression to Estimate the Association of Corticosteroids With COVID-19 Mortality. JAMA Netw Open. 2022;5(10):e2234425.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHern\u0026aacute;n M\u0026Aacute;, Brumback B, Robins JM. Marginal Structural Models to Estimate the Causal Effect of Zidovudine on the Survival of HIV-Positive Men. Epidemiology. 2000;11(5):561\u0026ndash;70.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNaimi AI, Cole SR, Kennedy EH. An introduction to g methods. Int J Epidemiol. 2016;46(2):756\u0026ndash;62.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMansournia MA, Etminan M, Danaei G, Kaufman JS, Collins G. Handling time varying confounding in observational research. BMJ. 2017;359:j4587.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCole SR, Hern\u0026aacute;n MA. Constructing inverse probability weights for marginal structural models. Am J Epidemiol. 2008;168(6):656\u0026ndash;64.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGraff\u0026eacute;o N, Latouche A, Geskus RB, Chevret S. Modeling time-varying exposure using inverse probability of treatment weights. Biom J. 2018;60(2):323\u0026ndash;32.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHern\u0026aacute;n MA. The hazards of hazard ratios. Epidemiology. 2010;21(1):13\u0026ndash;5.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGreifer N, Stuart EA, editors. Choosing the Causal Estimand for Propensity Score Analysis of Observational Studies2021.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePhillippo DM, Remiro-Az\u0026oacute;car A, Heath A, Baio G, Dias S, Ades AE, et al. Effect modification and non-collapsibility together may lead to conflicting treatment decisions: A review of marginal and conditional estimands and recommendations for decision-making. Res Synthesis Methods. 2025;16(2):323\u0026ndash;49.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHarrell FE. Cox Proportional Hazards Regression Model. In: Harrell FE, editor. Regression Modeling Strategies: With Applications to Linear Models, Logistic Regression, and Survival Analysis. New York, NY: Springer New York; 2001. pp. 465\u0026ndash;507.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePearce N, Vandenbroucke JP. Are Target Trial Emulations the Gold Standard for Observational Studies? Epidemiology. 2023;34(5):614\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFu EL. Target Trial Emulation to Improve Causal Inference from Observational Data: What, Why, and How? J Am Soc Nephrol. 2023;34(8):1305\u0026ndash;14.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHansford HJ, Cashin AG, Jones MD, Swanson SA, Islam N, Douglas SRG, et al. Reporting of Observational Studies Explicitly Aiming to Emulate Randomized Trials: A Systematic Review. JAMA Netw Open. 2023;6(9):e2336023\u0026ndash;e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDickerman BA, Garc\u0026iacute;a-Alb\u0026eacute;niz X, Logan RW, Denaxas S, Hern\u0026aacute;n MA. Emulating a target trial in case-control designs: an application to statins and colorectal cancer. Int J Epidemiol. 2020;49(5):1637\u0026ndash;46.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Cohort studies, Coronavirus Disease 2019 (COVID-19), Causal Inference, Vaccination, Target Trial Emulation","lastPublishedDoi":"10.21203/rs.3.rs-8831085/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8831085/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003eTarget trial emulation (TTE) is a framework to systematically address potential biases in causal inference when using observational data. We estimated vaccine effectiveness (VE) of the Omicron XBB.1.5 booster vaccination against SARS-CoV-2 infection between 2 October 2023 and 2 April 2024 using four TTE approaches.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eA hypothetical target trial was designed where eligible participants would be randomly assigned to receive booster vaccination or not. Four approaches were used to emulate this hypothetical trial using data of an ongoing prospective cohort study in the Netherlands. The first and second approach defined time zero as the start of the booster vaccination rollout and considered vaccination as a time-varying variable. The first approach adjusted for confounders by regression adjustment, while the second used inverse probability weighting. The third and fourth approach used multiple time zeros. In the third approach, all persons who received a booster vaccination on a specific day were matched 1:1 to persons who were not (yet) vaccinated on that day. In the fourth approach, confounders were adjusted for using inverse probability weighting.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eOverall VE was 30% (95%CI:26\u0026ndash;34), 28% (24\u0026ndash;32), 28% (24\u0026ndash;32) and 27% (24\u0026ndash;31) in the first, second, third and fourth approach, respectively. VE decreased as time since vaccination increased, but this was somewhat less pronounced in the third approach.\u003c/p\u003e\u003ch2\u003eConclusions\u003c/h2\u003e \u003cp\u003eEstimated VE was similar across the four approaches. The choice of approach should be based on the model assumptions, estimand of interest, and feasibility.\u003c/p\u003e","manuscriptTitle":"Comparison of methodological approaches in COVID-19 vaccine effectiveness estimation using observational data","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-02-17 16:14:50","doi":"10.21203/rs.3.rs-8831085/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"ad77554e-f3f4-4933-bb5d-8c8f7af09e17","owner":[],"postedDate":"February 17th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2026-03-16T08:12:40+00:00","versionOfRecord":[],"versionCreatedAt":"2026-02-17 16:14:50","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8831085","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8831085","identity":"rs-8831085","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}