JMIR Publications

ABSTRACT

Background:

Older adults with polypharmacy are at high risk of medication-related harm, and medicines optimisation is a recognised strategy to address this. Artificial intelligence (AI)-enabled clinical decision support systems (CDSS) and other digital medicines optimisation interventions have the potential to improve prescribing for older adults with polypharmacy. However, implementation remains inconsistent, and the contextual factors and mechanisms influencing successful adoption are not well understood.

Objective:

This realist synthesis aimed to explain how, why, for whom, and under what circumstances digital and AI-supported medicines optimisation interventions are successfully implemented in routine clinical practice for older adults with polypharmacy.

Methods:

A realist synthesis was conducted following RAMESES guidance. Eight electronic databases (MEDLINE via PubMed, Embase, CINAHL, PsycINFO, Scopus, Web of Science, ScienceDirect, and the Cochrane Library) were searched from inception to December 2025. Eligible studies evaluated AI- or digital-supported interventions for medicines optimisation in adults with polypharmacy. Data were synthesised using Context–Mechanism–Outcome (CMO) configurations, with programme theories iteratively refined through abductive, retroductive, and constant comparative analysis. Where mechanisms were not explicitly reported, they were transparently inferred using realist reasoning and identified in the supplementary analytic audit trail.

Results:

Of 1079 records identified across the eight databases, 358 were screened after removal of duplicates and non-retrievable records, and 23 studies met the inclusion criteria. Analysis of 384 extracted context-mechanism-outcome observations generated seven interrelated programme theories. Successful implementation was consistently associated with integration into routine clinical workflows, clinically relevant and actionable recommendations, multidisciplinary review, human oversight, patient-centred shared decision-making, organisational leadership and digital readiness, and transparent, explainable AI systems that fostered trust and professional legitimacy. Conversely, workflow disruption, alert fatigue, limited contextual reasoning, inadequate organisational support, and poor transparency reduced clinician engagement and implementation. These findings indicate that implementation success depends on interactions between technological, organisational, professional, and patient-level contexts rather than technology alone.

Conclusions:

This realist synthesis provides an explanatory framework describing the contextual conditions and mechanisms that facilitate implementation of digital and AI-supported medicines optimisation interventions in older adults with polypharmacy. The findings may inform the design, implementation, and evaluation of future digital prescribing systems and implementation strategies. However, the synthesis should be interpreted cautiously because it was restricted to English-language open-access publications, some mechanisms were inferred where not explicitly reported, and programme theory development involved interpretive judgements inherent to realist synthesis. Further prospective implementation research is needed to test and refine the proposed programme theories.