A validated synthetic cohort of 2,307 older Ghanaian adults (≥60 years) with comorbid hypertension and type 2 diabetes revealed that 58.3% were on polypharmacy regimens, with potentially inappropriate medication (PIM) rates of 76.0% under STOPP/START v3 criteria versus 49.2% under AGS Beers 2023 — a statistically significant divergence (McNemar's p<0.001) with only moderate inter-criteria agreement (kappa=0.469). Male sex and advanced age independently predicted higher medication counts. The cohort was calibrated to Ghana's 2022 Demographic and Health Survey and modeled real-world system failures including an 11.2% duplicate prescribing rate and 27% stockout-driven drug substitution probability.
This preprint, not yet peer-reviewed, addresses a genuine methodological gap in global geriatric pharmacoepidemiology: how to audit prescribing safety where electronic health records are absent. The Synthea-based synthetic data approach offers a scalable alternative, though its central limitation is profound — synthetic patients are not real patients. The 76% PIM rate may reflect the modeled system vulnerabilities as much as actual clinical practice, making causal inference impossible. The finding that STOPP/START v3 detects substantially more PIMs than Beers criteria aligns with European studies showing STOPP's broader scope, but neither tool was designed for West African formularies, where drug availability, local disease burden, and prescribing culture differ substantially. For health systems researchers, this is an incremental but methodologically creative contribution demonstrating that privacy-preserving simulation could help low-resource health networks identify systemic prescribing risks without waiting for EHR infrastructure.