Prediction tools to prioritise hospitalised adult patients at risk of drug related problems: An umbrella review.

Risk prediction tools assist pharmacists to identify and prioritise hospitalised patients at risk of drug-related problems (DRPs) requiring clinical review. This umbrella review identifies existing risk prediction tools, summarises their performance, and highlights factors important for prioritising hospitalised adults at risk of DRPs for clinical review. A systematic search of three bibliographic databases from January 2010 to March 2024, identified systematic reviews that qualitatively or quantitatively examined patient- and/or medication-related factors in risk prediction or prioritisation models or tools for adult inpatients at risk of DRPs. Extracted data included citation, review type, number and date range of primary studies, population, setting, outcomes, and key risk factors. Risk prediction tools were summarised by country, sample size, performance metrics (discrimination, sensitivity, calibration), number of risk factors, and external validation. All extracted data was checked by a second reviewer. The standardized Joanna Briggs Institute critical appraisal instruments were used to assess the quality of eligible studies. Twenty systematic reviews met our inclusion criteria, of which 18 were of sufficient quality for synthesis. Selected articles covered 171 unique primary studies and 32 risk prediction tools, of which 12 demonstrated acceptable or good discrimination of which four also had good calibration. Two tools showed the highest potential for future clinical use based on their performance metrics and external validation data but would benefit from further validation in diverse populations. Commonly reported risk factors for DRPs were psycholeptic medications (reported in 13 systematic reviews), age, comorbidities and reduced renal function (n = 12 reviews). Based on the reviews included, no tools appear to be yet suitable for routine clinical practice due to limited external validation, calibration and unclear risk factor definitions; however, some show promise for further development and testing. Data on existing risk prediction tools and risk factors provides a foundation for refining or automating current models, for instance via machine learning approaches that can iteratively incorporate risk factors to ease use and improve prediction of DRPs within specific clinical settings.