Abstract / Summary
Abstract Background Colorectal cancer is one of the most frequently missed cancer diagnoses in primary care. Artificial intelligence (AI)-based colorectal cancer risk prediction may support earlier detection, but its influence on general practitioners' (GPs) diagnostic reasoning and decision-making has not been thoroughly evaluated. This study aims to evaluate how GPs interpret and use AI-based colorectal cancer risk predictions in an emulated electronic health record environment. Methods This multimethod study is conducted in an emulated primary care electronic health record environment using historical electronic health record data. GPs assess 40–60 de-identified patient cases in a randomised patient–doctor design, with AI-based risk predictions displayed for a randomly allocated subset of assessments. Quantitative outcomes include suspicion of colorectal cancer, strength of suspicion, intended management, and ICD-10 coding. Semi-structured interviews explore GPs' perceptions of trust, usability, interpretability, and the influence of AI-based risk predictions on diagnostic decision-making. At the time of submission, all ten participating GPs had completed the assessments and interviews, and data analyses had commenced but were not yet complete. Discussion The study will provide evidence on how GPs interpret and use AI-based colorectal cancer risk predictions within an emulated electronic health record environment, supporting future prospective evaluation in routine primary care. The findings will guide refinement of the intervention and inform the development of a future prospective cluster-randomised trial evaluating AI-based clinical decision support in routine primary care. Trial registration Not applicable. This study involves an emulated electronic health record evaluation without patient intervention.