Artificial Intelligence Sparse Identification of Drug-Drug Interaction Signals Under Variable Pharmacovigilance Reporting Latency: A Causal Discovery Evaluation of Healthcare Polypharmacy Safety

Sparse identification of drug-drug interaction signals from spontaneous pharmacovigilance reports is complicated by variable reporting latency, which induces non-random missingness in the case base and biases naive causal discovery. The design is evaluated across a grid of 8, 16, 24 and 32 clusters, effect sizes of 3, 5 and 8 percentage points, and 2000 replications per cell, under a declared variance structure with between-cluster SD of 6.0, residual SD of 4.5 and intraclass correlation of 0.64. Power curves, minimum detectable effect curves, coefficient estimates with confidence intervals, residual diagnostics and kernel density estimates are reported.54 clusters are required to detect a 5 percentage point interaction at 80% power, and that the intraclass correlation of 0.64 substantially inflates the cluster requirement relative to naive sample-size formulas. The framework provides a design-theoretic basis for sparse causal discovery in polypharmacy safety surveillance and identifies the boundary conditions under which reporting-latency-robust signal identification is feasible.