Josephine van Dongen

4 Rotavirus vaccine effectiveness 97 Statistical analyses plan For the primary analysis of VE, we used Cox regression modeling of the time to first episode of severe rotavirus AGE as a function of vaccination status with age (between two and 18 months) modeled as time-variable. Infants were censored at their first severe episode or when lost to follow-up. Rotavirus vaccination was modeled as a time-dependent covariate thereby accounting for weeks of observation in vaccinated infants pre dose one and the initial 14 days post dose one, when protection from vaccination is considered to be still minimal. VE was defined as 1 minus the Hazard Rate (HR) ×100%. The propor tional hazard assumption was tested using Schoenfelds residuals. The primary analysis was adjusted for rotavirus seasonality (pre-specified). Weighted rotavirus epidemic intensity was calculated using the weekly data on rotavirus positive tests from the national virological surveillance. This surveillance includes aggregated test results from 20 sentinel laboratories in the Netherlands serving primary and secondary care. 17 Weight per week was derived by Rpackage timeseries accounting for each years’ rotavirus season and amount of repor ting labs. We considered the following additional variables as covariates: breastfeeding, young siblings (five years of age or less) in the household, daycare attendance, gestational age, household socio economic status (defined as the highest obtained parental education) and, parental origin. 18 We used likelihood ratio test to select the final model with covariates and present 95% confidence interval (CI) of the HR. The impact of the HRV selective vaccination program was estimated from the propor tion of infants with rotavirus related hospitalization in the post-implementation versus the pre- implementation period, expressed as 1 minus the risk ratio (RR).We used a mixed model with a binomial distribution and RR were obtained from odds ratio using the method described by Knol et al. 19 To account for potential cluster effects within hospitals, type of study site (academic versus general hospital) was added as a random intercept to the model. For the secondary analyses of a complete HRV series and subgroup analyses we used the cox model as described above. ForVE against rotavirus AGE episodes of any severity, accounting for multiple episodes within one individual, we used negative binomial regression withVE calculated as 1 minus the incidence rate ratios (IRR) ×100%. Person-time of observation was included as offset in the model. All-cause AGE analyses were performed using the same methods as described above. To calculate posterior probability of VE in our study-population, we used Bayesian analysis, with uninformative priors, of the Cox model for the primary outcome of VE against severe rotavirus AGE. All analyses were performed according to protocol (ATP), where all pre-implementation infants whose parents indicated willingness to receive HRV were compared to all post-implementation infants who received ≥ 1 dose of HRV. Only for the impact of HRV on hospitalization, we performed an intention to treat (ITT) analysis where all infants in the pre-implementation period were compared to all infants in the post-implementation period irrespective of HRV