Abstract
Diagnostic prediction models are commonly used in general practice to support clinical decision-making. Traditionally, these models have been developed using statistical methods such as logistic regression. While these approaches have proven useful, they often produce average risk estimates that may not fully account for the complexity of individual patients. In recent years, the use of machine learning (ML), a subfield of artificial intelligence (AI), has grown in healthcare. We examine the similarities and differences between traditional statistical methods and AI/ML approaches for diagnostic prediction in general practice. Using examples from daily practice, we explore how ML techniques can add value, particularly in handling large, complex datasets such as those derived from electronic health records. We also discuss key challenges that hinder the adoption of AI/ML in general practice, including interpretability, data quality, external validation, clinical relevance, implementation and legal issues, and practical usability. We provide recommendations to overcome these challenges. The potential of AI/ML can only be realised if tools are developed collaboratively with GPs, focused on real-world clinical problems, and rigorously validated in practice settings. GP associations, GPs, patients, and primary care scientists should take an active role in the development, validation, and implementation of AI/ML-based diagnostic prediction tools for general practice.