Post-marketing drug safety assessment begins in an evidential environment that is exceptionally valuable for detecting unexpected harm but poorly suited to automatic causal interpretation. Spontaneous reports, electronic health records, claims databases, registries, and other real-world sources can reveal safety patterns that were not detectable before approval, yet each source embeds distinct reporting processes, treatment-selection mechanisms, temporal structures, measurement limitations, and clinical uncertainties. Causal inference can reshape this landscape by forcing safety questions to be specified more precisely. Target-trial thinking clarifies eligibility, treatment strategies, time zero, follow-up, outcomes, and estimands; active and clinically credible comparators improve treatment-group comparability; longitudinal methods address treatment switching and changing risk; negative controls can expose systematic error; and quantitative bias analysis makes assumptions about unmeasured bias explicit. None of these methods, however, converts observational safety evidence into causal certainty by technique alone. Post-marketing decisions also depend on event phenotype, seriousness, reversibility, alternative explanations, individual case chronology, biological plausibility, signal maturity, and regulatory context. This Current Opinion argues that causal inference should therefore function as a disciplined layer of safety assessment rather than as a replacement for clinical judgment. The most defensible architecture is one in which statistical signals generate causal questions, causal methods test increasingly explicit counterfactual contrasts, and clinical reasoning remains independently responsible for dimensions that population-level estimation cannot fully encode.
IJPRAS
