Extended Data Fig. 1: Evaluation of residual confounding.

We quantified the risk of residual confounding in primary comparisons between several cohorts (individuals exposed to the live-attenuated zoster vaccine [ZVL], recombinant zoster vaccine [RZV], the 23-valent pneumococcal polysaccharide vaccine [PPSV23], not exposed [Not exposed] to any herpes zoster [HZ] vaccine and individuals with only a single versus multiple HZ episodes) using a negative control via time shift (NCTS) analysis. Briefly, to conduct the NCTS analysis, we artificially shifted the exposure date of the exposed groups to 1 year prior to their actual recorded exposure while keeping other elements of the study design constant. Since this new exposure date is before any biological effect could have taken place, any differences in cohorts already present before exposure indicate residual confounding not accounted for via matching. The estimated risk of residual confounding is represented as a per-comparison group of bars for each of the primary comparisons with height reflecting an estimate of (1 – relative risk [RR]) at 3, 6, and 9 months after the shifted index date based on the Nelson-Aalen estimates of cumulative hazards along with error bars indicating a 95% confidence interval (CI). The respective sample sizes in terms of the number of matched pairs of individuals at the shifted index date are: 77,331 (RZV [2+ doses] vs ZVL), 276,987 (RVZ [1 dose] vs PPSV23), 191,073 (RZV [2+ doses] vs PPSV23), 344,159 (ZVL vs PPSV23), 85,650 (HZ [multiple] vs HZ [single]), 481,124 (RZV [2+ doses] vs Not exposed), 441,740 (ZVL vs Not exposed). The higher the y-axis value, the more residual confounding was present in a comparison. We found that comparisons between elective vaccines (ZVL, RZV, PPSV23) were generally at a lower risk of residual confounding than comparisons to Not exposed, a finding suggesting that using another elective vaccine (PPSV23) as control is effective at reducing residual confounding. Error bars represent 95% CIs.

Source data