Challenges in Using Clinical Data for AI-Enabled Diagnostic Support

Secondary use of clinical data for research and algorithm development can present challenges. Using a multi-hospital, retrospective dataset, we examined laboratory values and corresponding recorded diagnoses across common tests. Unanticipated findings prompted further investigation, leading to the identification of key interpretive considerations when using routinely collected clinical data. We highlight three challenges: data quality and completeness; temporality and clinical context; and coding and governance constraints. These challenges have important implications for the development of AI-enabled tools in clinical settings.