What works to address inequalities in primary care: Development of Living Evidence Maps using machine learning

Objectives: Health inequalities in primary care persist across many high-income countries, with populations experiencing disadvantage often receiving less care despite greater need. Traditional evidence synthesis methods struggle to keep pace with the growing volume of research to help policy makers and practitioners to take timely, evidence-informed action. Our aim was to develop a Living Evidence Map of interventions that address inequalities in primary care, using machine learning (ML) to enhance the efficiency of evidence identification, screening, and mapping. Study design: Evidence synthesis. Methods: We used EPPI-Reviewer software to train a machine learning classifier to identify relevant studies. This was complemented by citation searching, and records were manually screened in order of predicted relevance (priority screening). Included studies were coded by intervention type, disadvantaged group, and health or care outcome, which was then visualised using EPPI-Visualiser. Results: A total of 31,871 articles were screened, resulting in 577 primary studies, 481 systematic reviews, and 6 umbrella reviews being included in the Living Evidence Map. Most studies focused on ethnic minority groups, with common interventions including education, advice and counselling, and culturally tailored care. There was a paucity of studies targeting gender and sexual minorities, and structural interventions (e.g. funding, workforce). Conclusions: The Living Evidence Map offers a dynamic, policy-relevant tool for navigating the evidence base on health inequalities in primary care. It serves both researchers and decision-makers by making it easier to see what evidence exists, where it's concentrated, and where there are gaps. However, further work is needed to improve inclusion of grey literature evidence and interventions addressing intersectional disadvantage to support decision-making for end users.