AI-driven e-prescription and pharmacovigilance platform for personalized medicine in community pharmacy

Thesis Abstract

The rapid digitization of community pharmacy services and rising concerns about safe and effective medication management necessitate an AI-powered integrated platform that automates e-prescribing workflows while enhancing real-time pharmacovigilance and personalized medicine delivery. This study aims to design, implement, and evaluate a pilot AI-driven e-prescription and pharmacovigilance platform that supports clinicians, pharmacists, and patients in optimizing therapeutic outcomes. Specific objectives include (1) developing a modular system that generates personalized drug recommendations and interaction alerts using patient-level data (demographics, comorbidities, genomics where available, prior adverse reactions), (2) integrating real-time pharmacovigilance subsystems that detect signals from prescription data, patient-reported outcomes, and spontaneous ADR reports, (3) assessing usability and acceptance among community pharmacists (n = 60) and prescribers (n = 30) across 12 urban and suburban pharmacies, and (4) evaluating impact on prescription accuracy, time-to-dispense, and incidence of adverse drug events over a 6-month deployment period. The methodology adopts an explanatory sequential mixed-methods design. The quantitative phase employs a quasi-experimental pre-post design with matched controls, collecting data from 1,200 patient prescriptions and 400 pharmacist–patient interactions, analyzed using multivariate regression to quantify changes in prescription accuracy (error rate per 100 prescriptions), time-to-dispense, and ADR reporting rates, alongside interrupted time series analysis to assess temporal effects. The qualitative phase uses thematic analysis of semi-structured interviews with pharmacists and prescribers (n = 40) to elucidate usability, perceived safety improvements, and integration challenges, guided by the Technology Acceptance Model (TAM) and the Unified Theory of Acceptance and Use of Technology (UTAUT). The research will employ a Bayesian hierarchical model to synthesize heterogeneous data sources and quantify uncertainty in outcome estimates. Data sources include electronic health records, e-prescription logs, pharmacovigilance records, and patient-reported outcome measures collected via secure mobile surveys. Instrument validity will be ensured through expert-validated item pools, pilot testing (n = 20 pharmacists), and reliability checks (Cronbach’s alpha > 0.7 for scales). Expected findings include statistically significant reductions in prescription error rates and time-to-dispense, enhanced detection and timely reporting of ADRs through automated signal generation, and increased user acceptance among pharmacy staff and prescribers due to perceived usefulness and trust in AI-driven recommendations. The study anticipates that personalized recommendations, when augmented with real-time pharmacovigilance signals, will improve therapeutic appropriateness and patient safety, particularly for polypharmacy and high-risk populations. The anticipated contribution to knowledge comprises (i) an empirically validated, scalable framework for AI-enabled e-prescribing integrated with pharmacovigilance in community pharmacies, (ii) methodological insights into implementing privacy-preserving, interoperable AI solutions in primary care, and (iii) evidence on the behavioral determinants of technology adoption among pharmacists and prescribers in real-world settings, contextualized by theories of technology acceptance (TAM, UTAUT) and socio-technical systems theory. The study concludes that an AI-driven platform can meaningfully improve prescribing accuracy, ADR detection, and patient outcomes when aligned with clinician workflows and robust governance for data privacy and safety. Practical recommendations include adopting modular architecture for incremental deployment, establishing governance mechanisms for data stewardship and pharmacovigilance signal validation, and providing targeted training to enhance trust and utilization. Limitations include potential selection bias due to urban site concentration, constraints on genomics data availability, and the need for longer-term follow-up to capture rare adverse events; future research should explore integration with comprehensive clinical decision support, cross-sector data sharing, and assessment across diverse health systems.