(301) Methodological development of an enhanced safety surveillance for the Janssen COVID-19 vaccine (JCOVDEN) through rapid cycle analysis of US administrative claims data

Publication Number: 1291

Background: As a part of a comprehensive routine pharmacovigilance plan, Janssen developed enhanced safety surveillance of its COVID-19 vaccine (JCOVDEN) by applying sequential analyses in multiple claims databases.

Objectives: To develop and apply methods to rapidly assess potential associations between JCOVDEN vaccination and adverse events of special interest (AESI) through sequential analysis of observational data

Methods: Self-controlled case series (SCCS) and comparative cohort method (CM) designs were used. We evaluated data quality across three data sources and characteristics of phenotypes and applied a set of prespecified study diagnostics. Results from analyses that failed diagnostics were not unblinded. Power was assessed through minimal detectable relative risk. We summarized expected absolute systematic error using negative control estimates. SCCS analyses utilized a spline function to adjust for calendar time to assess time trends and reverse causality was assessed comparing outcome risks in the 30 days before and after exposure. CM analyses included assessing attrition reasons, evaluating equipoise as a linear transformation of the propensity score, and measuring covariate balance with standardized differences.

Empirical calibration based on negative control outcomes was applied to correct for bias due to unmeasured confounding. A multiple testing correction was applied to the primary risk window analysis to account for the multiple outcomes, methods, databases, and sequential analyses.

SCCS analyses estimated incidence rate ratios (IRR) across exposed and unexposed time. CM compared recipients of JCOVDEN and mRNA vaccines separately using a conditional Cox proportional hazards model to estimate hazard ratios (HR) after variable ratio matching using a propensity score model fit via large-scale regularized regression.

Results: We conducted 2 analyses between Jan2021 and Dec2022 using MarketScan (JCOVDEN n=401,339), HealthVerity Marketplace (JCOVDEN n=756,919), and Optum Clinformatics® (JCOVDEN n=206,499). Across 56 prespecified AESIs, 4 risk windows, 5 analytic approaches (3 SCCS and 2 CM), and 3 data sets and a meta-analysis, 2,851 analyses passed study diagnostics. The meta-analysis effect estimates for SCCS and CM were 4.03 (2.17-7.47) and 3.59 (1.62-7.96) for Guillain-Barre syndrome (GBS).

Conclusions: We established a rapid and robust process to conduct safety surveillance of JCOVDEN with real-world data. The process applied a set of predefined databases, phenotypes, and study specific diagnostics to prioritize the evaluation of risk estimation that was integrated into ongoing regulatory reporting. We observed an increased risk of GBS, which has been deemed an important identified risk.