Attributable risk is often used to assess the risk–benefit balance of a vaccine, guide vaccine technical committee recommendations, and ultimately help providers communicate the risk to the public [10, 11]. Given these important implications, it is essential to provide an assessment of attributable risk that reflects the recommended vaccinee population. If attributable risk is calculated based only on data from an SCCS analysis, the attributable risk is impacted by the study-specific GBS control period rate, such that vaccines with a lower GBS control period rate have a lower attributable risk for a given IRR as evidenced in the example of RZV and the two RSV vaccines. Our findings emphasize the non-comparability of attributable risk estimates that are derived from an individual SCCS study where SCCS control period rates of GBS are based on a study population, which is both exposed and has the outcome of interest—in other words, a non-generalizable population. Prevalence and incidence rates of GBS in the US have been increasing over time [13]. This is likely contributing in part to the lower attributable risk seen for a much higher IRR for RZV, as the data were collected earlier. Specifically, the United States Prescribing Information (USPI) of RZV utilized an SCCS-based IRR of 4.96 (95% CI 1.43, 17.27) associated with an “estimated 6 excess cases of GBS per million doses administered to adults aged 65 years or older” (i.e., attributable risk = 0.52 per 100,000 doses) [8]. Similarly, the monovalent H1N1 influenza vaccine was reported to have an IRR of 2.41 (95% CI 1.14, 5.11), with a corresponding attributable risk of 2.84 (0.21, 5.48) per million [9]. These relatively lower attributable risks compared to that of the RSV vaccines are due to the lower control period GBS rates in the RZV and H1N1 influenza vaccinee populations from their respective SCCS vaccine studies. In this study, we quantified changes in attributable risks using different external background rates and provided insights into potential bias arising from background rates that are not representative of the population recommended for vaccination. We also demonstrated that using a common, representative background rate to calculate the standardized attributable risk is informative in highlighting that studies with higher IRRs correspond to higher attributable risks and those with lower IRRs correspond to lower attributable risks. These findings highlight the importance and impact of background event incidence rates on attributable risk calculation and the benefit of using a background rate from a population representative of those recommended for vaccination to standardize to a population-level attributable risk.

Warnings regarding potential increased risk of GBS are present in the prescribing information of several commonly used vaccines [14,15,16,17,18] based on the results from an SCCS study design. The risk of vaccine-associated GBS in these warnings is usually expressed as an excess of cases per million doses—for example, 1–2 additional cases per one million doses for influenza vaccine [19], and 3–6 additional cases per one million doses for RZV [20]. Recognizing the impact of deriving attributable risk estimates solely from SCCS data, the FDA has included a disclaimer in the recent US package insert update for subunit RSV vaccines: “the background risk of GBS in a study population influences the excess GBS case estimate and may differ between studies, precluding direct comparison to excess GBS case estimates from other vaccine studies or populations” [21, 22]. Their recent study manuscript also includes this caution [7]. However, such SCCS-derived attributable risks have been applied to hypothetical populations of a million future vaccinees and used to compare the risk–benefit balance between vaccines [10].

While calculation of attributable risk aims to help healthcare providers and the public contextualize and interpret risk, if not methodologically sound, it may also lead to confusion and inappropriate comparisons between different vaccine brands [10]. For example, an Advisory Committee on Immunization Practices (ACIP) slide presentation from October 2024 concluded that the potential risk of GBS after protein subunit RSV vaccination was possibly higher than any marketed adult vaccine [11]. However, the IRR for other marketed vaccines is higher than these vaccines. The attributable risk is higher due to higher background risk of GBS in the SCCS study population.

This population-level attributable risk analysis has limitations. First, geographic, demographic, and methodological variations in background rates are a challenge to accurately determining GBS background incidence rates in the population recommended for vaccination. Although we limited background rates to Medicare populations to match the SCCS study populations and avoid heterogeneity from non-Medicare sources, the two GBS background incidence rates derived from the US Medicare population may not represent the contemporary population recommended for a specific vaccine because they are from 5 (RZV: 2017–2020) to 15 years (H1N1: 2009–2010) ago. Although the most recent background rate (4.63 per 100,000 person-years) was much higher than the earlier rate (1.83 per 100,000 person-years), the magnitude of change in the recalculated attributable risks compared with the original estimates depended on how closely the external background rate aligned with the SCCS-derived control rate. Changes ranged from 20–1054% when using the most recent background rate to 55–335% when using the earlier rate (for the attributable risk per 100,000 doses). Importantly, the direction of change in attributable risks was consistent with both the IRR and the background rate. To assess the robustness of these results, we conducted supplemental sensitivity analyses by applying two hypothetical GBS background rates across a broader spectrum. Similar trends were observed, and Spearman’s rank correlation indicated a stronger association between the IRR and the standardized attributable risk (ρ = 0.88) after adjusting for the background rate. Second, the updated analysis did not consider the potential effect modification with regards to the original IRR from SCCS studies and did not address any potential unmeasured time-varying confounders. Effect modification, if present, may impact the validity of applying an SCCS IRR to an external background GBS rate if the effect modifier is not present in the same frequency in the population from which the external background rate is derived. However, at this time, effect modifiers of the vaccine GBS relationship have not been identified, and the SCCS-derived IRRs are overall estimates without stratification by any potential effect modifier, such as age, frailty, and seasonality. We made efforts to harmonize the population characteristics by deriving the background rates from the same database (i.e., Medicare) and from overall estimates without stratification by any potential effect modifier, such as age, sex, or comorbidity. Therefore, the SCCS-derived population-level attributable risk using the generalizable published background rate could still be impacted by these limitations. This study raises questions about the inclusion of attributable risk estimates derived exclusively from an SCCS analysis in product labels and direct comparisons of attributable risk across vaccine studies or populations in vaccine risk–benefit assessments. Inaccurate estimation of attributable risk can affect interpretations, clinical decision-making, and recommendations by public health and regulatory agencies.