From Pills to Pixels: Why Software as a Medical Device Will Redefine Pharmacovigilance

Pharmacovigilance was built around molecules. Its detection methods, reporting pathways, and regulatory architecture assume a static, biochemically active product whose harms can be characterised through individual case safety reports and disproportionality analysis. Healthcare is crossing a threshold at which clinical decisions are increasingly mediated by software — algorithms that triage stroke, predict deterioration, and individualise drug dosing. Regulators classify these tools as Software as a Medical Device (SaMD); the U.S. Food and Drug Administration alone has authorised over 1,250 AI-enabled medical devices for clinical use. SaMD does not produce molecular adverse effects; it produces decision-mediated harm — delayed alerts, false negatives, and predictions that degrade outside training populations. These events rarely surface in classical pharmacovigilance systems, yet directly affect morbidity and mortality. Recent guidance from the FDA, EMA, and India's CDSCO (2024–2025) acknowledges that traditional frameworks were not designed for adaptive AI, but the operational pharmacovigilance response has not kept pace. This Commentary identifies five structural shifts required for SaMD-era safety oversight: continuous real-world performance monitoring, algorithm change control with mandatory safety reassessment, outcome-based signal detection, extension of MedDRA terminology for algorithm-mediated harm, and unified drug-device-software governance. It addresses the central attribution problem — assigning harm when a clinician sits between the algorithm and the patient — and argues that safety physicians fluent in both clinical medicine and algorithmic logic are a prerequisite for safe digital medicine, and that this professional pathway must be deliberately built. As software increasingly shapes care, patient safety cannot remain analog.