Observed Versus Expected Analysis-How Does It Fit in the Pharmacovigilance Toolkit?

Abstract

Observed versus expected (O/E) analyses have been used in an unprecedented scale for the safety monitoring of the COVID-19 mass vaccination. The extent of their usage changed its nature, which consisted of a mixture of medical expertise and epidemiology, into something more algorithmic and automated. By doing so, the observed versus expected analysis became closer to disproportionality analysis (DPA), which is also a type of observed versus expected analysis that differs in the way the expected is calculated. A qualitative assessment of the strengths and limitations of both methods concludes that the algorithmic O/E is more likely to underestimate under-reporting, is more likely to be sensitive to asymmetrical differences in the definition of the condition of interest, and is more dependent on a greater variety of data sources or medical knowledge that might not be accurate for emerging safety issues (exposure, background incidence rate, and risk window). Provided some adjustment (stratification and/or subgrouping) of the routine disproportionality into a targeted disproportionality occurs, which would account for the epidemiological specifics of the vaccine and event-of-interest, the targeted DPA has the potential to be promoted from a signal detection method into a signal evaluation method that could advantageously replace the algorithmic O/E analysis. Research on the setup of a sensitivity analysis framework integrating several standardized choices of disproportionality settings, along with measures (qualitative or quantitative) of the biases for each choice, could be more beneficial for the pharmacovigilance field than studies designed to estimate the background incidence rates of adverse events of special interest for the sole purpose of being used in O/E analyses.