Model Driven Method for exploring Individual and Confounding Effects in Spontaneous Adverse Event Reporting Databases

Abstract

Abstract Background Spontaneous Adverse Event Reporting (SAER) databases play a crucial part in post-marketed drug surveillance. The reported odds ratio (ROR) is commonly used to detect the significant signal of AE-drug combinations. However, the typical ROR may be biased by heterogeneity from individual factors such as gender and age. In addition, confounding biases render the typical ROR far from indicative of causality. Therefore, these issues urgently require solutions. Methods Model driven ROR (MD-ROR) was proposed as an alternative to typical ROR to explore individual and confounding effects in SAER databases. Unlike the traditional 2*2 cross table approach, our method employed Poisson regression with two-way interactions to estimate the MD-ROR, which was shown to be equivalent to typical ROR. Subsequently, we introduce the MD-ROR under three-way interaction to reveal the heterogeneity behind pooled crude ROR and to identify subgroup effects on the signals of AE-drugs. We also introduce adjusted MD-ROR to address confounding biases by flexibly defining confound effects in the model. To test our methods, the simulation data and FDA Adverse Event Reporting System (FAERS) database were both used. Result The simulated data suggested the subgroup effects estimated by MD-ROR were unbiased and efficient. Additionally, the adjusted MD-ROR was more robust against confounding biases than crude ROR. Application of our method to FAERS database showed differences in drug interaction and cardiac adverse events between males and females for Midazolam existed. In addition, the AE-drug combinations, Midazolam-septic shock and Midazolam-depression, were found overestimated potentially due to confounding biases from gender. Conclusion Our study highlighted that MD-ROR is a promising method for exploring individual and confounding effects in SAER databases. Our method provides a bridge between SAER databases and flexibly customized models.