Data and Code for Preprint 1: Polypharmacy Burden and PIP Among Older Adults in Tatale, Ghana.

Version History

Version 2.0 (July 22, 2026)
This update addresses technical omissions in the initial release to ensure full reproducibility of the indication-based screening (STOPP criteria).

• Data Enrichment: Added the missing conditions.csv file to the longitudinal patient archive (csv_aligned.zip). This file is essential for matching medications to specific clinical diagnoses.

• Code Correction: Updated align_ghana_data.py to fix a data-type conflict (TypeError) during the Hemoglobin calibration process and ensured proper sub-cohort filtering.

• Dataset Synchronization: Re-generated the master analytical file (P1_master_analytic_aligned.csv) using the corrected script to ensure 100% mathematical consistency.

Abstract

**Background:** Geriatric polypharmacy and potentially inappropriate prescribing (PIP) represent severe iatrogenic safety challenges in sub-Saharan Africa. However, clinical quality auditing in under-resourced settings is routinely impeded by electronic health record (EHR) scarcity and data-privacy constraints.

**Objective:** To generate, validate, and clinically audit a privacy-preserving synthetic cohort representing older adults in Tatale, Ghana, comparing the diagnostic yield and statistical concordance of the 2023 American Geriatrics Society (AGS) Beers and STOPP/START version 3 (v3) screening criteria.

**Methods:** Utilizing the Synthea framework, a synthetic cohort of $N = 3,958$ geriatric patients ($\ge 60$ years) diagnosed with comorbid hypertension and Type 2 diabetes was generated. Demographic parameters were calibrated to the 2022 Ghana Demographic and Health Survey (GDHS; $N = 5,785$). Multidimensional validation was conducted using the first Wasserstein distance ($W_1$). Local health system vulnerabilities, including an 11.2% duplicate prescribing rate and a 27.0% stockout-driven drug substitution probability, were programmatically modeled. Automated scripts screened the cohort's medications against both criteria. Descriptive, ANOVA, and concordance (Cohen's Kappa) analyses were performed in JASP 0.97.0.

**Results:** The synthetic cohort exhibited high demographic alignment with the target population ($W_1 = 15.25$ for age; simulated: 47.55% female; GDHS: 52.15% female). The baseline polypharmacy rate was 58.29% ($n = 2,307$). Programmatic screening under the STOPP criteria identified a significantly higher potentially inappropriate medication (PIM) prevalence (76.00%; 95% CI: 74.6%–77.3%) compared to the Beers criteria (49.22%; 95% CI: 47.6%–50.8%; McNemar's $p < 0.001$). Inter-criteria agreement was moderate ($\kappa = 0.469$; 95% CI: 0.446–0.492). Male sex ($\chi^2 = 58.97, p < 0.001$) and advanced age (ANOVA $F = 47.93, p < 0.001$) were associated with significantly higher medication counts.

**Conclusions:** Rural older adults in secondary-care settings face severe polypharmacy and PIP exposure. Programmatic screening using STOPP/START v3 exhibits superior diagnostic sensitivity compared to the Beers criteria. Validated synthetic clinical data modeling represents a scalable, privacy-preserving, and ethically responsible methodology to audit clinical prescribing safety in data-scarce health networks.