FAIR-PV 2.0: A Shapley-Based Framework for Patient-Level Causal Attribution in Pharmacovigilance with Monte Carlo Uncertainty Quantification

Background and Aims Pharmacovigilance (PV) uses signal detection, disproportionality analyses, and qualitative causality tools (WHO-UMC, Naranjo) that identify population-level associations but cannot quantify individual causal responsibility or integrate effect sizes, baseline risks, or multifactorial interactions. The FAIR framework (originally from occupational medicine) offers structured quantitative attribution. We adapted FAIR for PV as FAIR-PV 2.0, adding Shapley values and Monte Carlo uncertainty. Material and Methods FAIR-PV includes: (1) OHAT risk-of-bias evidence filtering; (2) relative effect extraction (RR/HR/OR); (3) personalized absolute risk modeling; (4) Shapley partitioning of excess risk across drug, comorbidities, interactions; and (5) medico-legal interpretation (>50% ”more likely than not”; ≥20% ”substantial”). Robustness tested via Monte Carlo simulation. Three clinical cases analyzed. Results Shapley allocations remained stable across risk scenarios. DOAC bleeding case (p 0 =0.006; DOAC RR=1.55, renal RR=2.2, NSAID RR=1.6): excess risk 0.0177 partitioned DOAC 46%, renal 32%, NSAID 22% (MC median DOAC 46%, 90%UI 32-58%). Anti-PD1 pneumonitis: drug 63% (52-75%). Drug contribution >20% probability stayed >90% for RR≥1.5. Conclusion FAIR-PV 2.0 delivers quantitative, reproducible individual causal attribution in PV, complementing existing tools for clinical, regulatory, and medico-legal applications.