Answers are grounded in this paper and cite their sources.
Background The rapid advancement of information technology and machine learning (ML) has garnered significant interest from health care managers and policymakers. However, incorporating reliable and transparent predictions into health systems remains challenging, necessitating an investigation into gaps in digital health interventions within real-world settings characterized by resource constraints and reliability issues. Objective This study aimed to map and synthesize the available evidence on the application of ML algorithms to real-world data in health systems and identify their operational targets and clinical implications for health planning and public policy formulation. Methods Following the Joanna Briggs Institute (JBI) methodology for scoping reviews and PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses Extension for Scoping Reviews) guidelines, we searched the PubMed, Embase, Scopus, and LILACS databases for empirical studies published between 2016 and 2025 that evaluated the application of ML to health management, surgical queues, and public policy using real-world data. The study protocol was retrospectively registered with the OSF. Results Of the studies identified through the database searches (2016-2025), only 6 empirical studies met all inclusion criteria, revealing that the real-world deployment of ML in health management and public policy remains at an early stage. The included models addressed micro-operational tasks, such as surgical blood transfusion forecasting, early warning system mortality prediction, postsurgical discharge planning, laboratory error detection, patient-reported outcome measure (PROM) prediction, and emergency care classification. Although high discriminatory performance was reported (area under the receiver operating characteristic curve [AUROC] 0.76-0.99), the studies were limited by single-center retrospective designs, a lack of independent external validation, and inadequate reporting of model calibration. Conclusions Current evidence on ML for macrolevel health planning and public policy formulation is limited and micro-operationally focused. Translating these algorithms into public health governance (eg, waiting list regulation in universal health systems such as the Sistema Único de Saúde [SUS]) requires moving beyond “black-box” predictions toward explainable AI (XAI), natural language processing (NLP) to analyze unstructured data, transparent model calibration, and rigorous evaluations of health equity.
No takes yet. Share an insight, caveat, or question.
Rocha et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: