Some features of statistical analysis of spontaneous adverse drug reporting data

Abstract

Introduction. Risk of adverse drug reactions (ADRs) is a serious issue in pharmacotherapy and a major public health concern. Safety signal detection during the post-marketing phase is one of the most important goals of drug safety surveillance. Spontaneous reporting systems (SRS) are still widely used to identify safety signals based on real-world data. Various data mining statistical methods have been developed for this purpose, and they are classified into frequentist and Bayesian approaches. Statistical methods can also be used for the analysis of patient-related risk factors (demographic characteristics, concomitant diseases or medications). Identification of patients at high ADR risk is important for personalized pharmacotherapy.Objective. To present and review issues and features of the statistical methods for SRS data, developed by other authors and published in the literature, this tool may be useful for appropriate statistical analysis and accurate interpretation of passive surveillance data.Methods. In this paper, we present the known and commonly used frequentist or classical methods for correct statistical analysis of spontaneous reports. These methods for signal detection and their modification for drug-host factor interaction analysis are relatively easy to understand, interpret, and compute based on the contingency 2x2 tables: reporting odds ratio (ROR), proportional reporting ratio (PRR), and normal approximation test. Different approaches to the multiple comparison problem in passive safety surveillance settings were also discussed.Results. As an example, the aforementioned methods were applied to analyze sex disparities in liver toxicity based on the spontaneous reports extracted from the Russian National Pharmacovigilance database. The tests identified drugs for which liver toxicity demonstrates significant disproportionality regarding sex compared with other AEs. The results of all statistical methods were similar.Conclusions. Although spontaneous report databases are subject to numerous potential sources of bias and well-known limitations, these large-scale databases remain a widely used, effective, and relatively inexpensive approach for post-marketed drug surveillance. With the use of correct statistical methods, spontaneous reporting databases can provide valuable information for hypothesis generation, which should be investigated further, as well as essential data on the evaluation of risk factors and risk populations.