A Pharmacovigilance Decision-Analytic Framework for Adverse Drug Events

Thesis Abstract

Adverse drug events (ADEs) impose substantial morbidity, healthcare costs, and public health risk, challenging traditional pharmacovigilance practices that rely on spontaneous reporting and retrospective signal detection. This study develops and tests a decision-analytic framework to optimize the detection, assessment, and management of ADEs by integrating quantitative risk modeling with qualitative expert judgment, thereby improving timeliness and accuracy of pharmacovigilance decisions. The aim is to create a coherent framework that synthesizes evidence from multiple data streams to support decision-makers in prioritizing signals, allocating investigative resources, and informing risk communication. Specific objectives are to (i) identify core decision points in pharmacovigilance workflows where uncertainty materially affects ADE management, (ii) construct a multi-criteria decision analysis (MCDA) model coupled with Bayesian updating to quantify evidence strength and risk prioritization, (iii) integrate a decision-analytic simulation with a conceptual pharmacovigilance ecosystem to evaluate trade-offs between patient safety, resource use, and regulatory actions, (iv) validate the framework using retrospective ADE case studies across three therapeutic areas, and (v) assess robustness under varying data quality and reporting delays. The study adopts a mixed-methods design combining quantitative modeling with qualitative stakeholder input. A purposive sample of 12 ADE case datasets drawn from national pharmacovigilance databases, electronic health records, and spontaneous reporting systems will be analyzed, supplemented by semi-structured interviews with 18 pharmacovigilance experts, clinicians, and regulatory personnel to elicit criteria weights and contextual constraints. Quantitative analysis will implement a hierarchical Bayesian network to model causal relationships between patient factors, drugs, interactions, and outcomes, integrating with an MCDA framework (e.g., TOPSIS or ELECTRE) to rank signals under differing policy scenarios. Sensitivity analyses will include probabilistic Monte Carlo simulations and scenario analyses to test robustness to missing data and reporting lags. Model calibration will use historical ADE detection timelines, positive predictive values, and actionability metrics measured against established regulatory decisions. The theoretical underpinning draws on information economics and behavioral decision theory, incorporating Prospect Theory to account for risk preferences of regulators and the cumulative prospect framework to model uncertainty in signal validation. Expected findings include (i) a validated decision-analytic framework capable of producing probabilistic signal scores, priority rankings, and recommended actions with quantified uncertainties; (ii) identification of key drivers of decision quality—data completeness, timeliness, and prior evidence strength—and their differential impact across therapeutic areas; (iii) demonstrable improvements in prioritization efficiency, reducing investigation backlog by an estimated 20–30% in simulated audits, while maintaining or enhancing ADE detection sensitivity; and (iv) evidence that integrating Bayesian updating with MCDA reduces decision bias attributable to incomplete information. The study anticipates discovering context-specific thresholds—for example, higher urgency for signals involving high-risk medications or vulnerable populations—embedded within the framework’s rule set. Contributions to knowledge include a novel, integrative decision-analytic framework for pharmacovigilance that bridges quantitative risk modeling with qualitative decision criteria, enabling transparent, reproducible, and explicable ADE management decisions. The framework advances methodological practice by operationalizing MCDA-Bayesian hybrids in pharmacovigilance, providing a transferable blueprint for other health-safety surveillance systems. It also informs policy by offering a structured approach to resource allocation, signal validation standards, and risk communication protocols in light of uncertain and heterogeneous data. The study concludes that a formalized, model-driven pharmacovigilance framework enhances decision quality in ADE management, improves the efficiency of signal-to-action workflows, and supports adaptive regulatory strategies under data constraints. Recommendations include adopting the framework as a decision-support tool within national pharmacovigilance centers, integrating real-time data feeds from EHRs and other surveillance systems, and standardizing reporting formats to strengthen prior evidence incorporation. Potential future work involves extending the model to account for drug class-specific mechanisms, integrating patient-centered outcomes into the MCDA criteria, and exploring automation opportunities for routine sensitivity analyses.