Patterns of diagnostic errors using hierarchical cluster analysis: a single-center, cross-sectional study in general internal medicine

Objectives: Diagnostic errors are caused by disease complexity, variable settings in which care is provided, and multiple cognitive and system-level factors. However, whether outpatient diagnostic errors aggregate into specific failure modes that align with the clinical context remains unclear. Therefore, we aimed to explore these modes by examining the salient causes of unexpected readmission. Methods: We conducted a cross-sectional study to analyze unexpected readmissions among outpatients in general internal medicine who visited a clinic within 14 days of their initial visit. Two physicians independently assessed the diagnostic errors using the Revised Safer Dx Instrument and Diagnostic Error Evaluation and Research (DEER) taxonomy. We applied hierarchical clustering to the DEER taxonomy and conducted Fisher's exact tests to examine the associations with patient, physician, and system factors (presence of a referral letter, personal protective equipment [PPE] use, and late session). Results: Among the 146 patients, 50 (34.2%) experienced diagnostic errors. Cluster analysis revealed two main categories of failure modes: (1) diagnostic prioritization and urgency appraisal failure, and (2) information acquisition and synthesis failure. The latter category was further divided into three clinically relevant subtypes: data-gathering failure, where PPE use was more common; information synthesis and interpretation failure, in which referral letters were more common; and undertesting and safety-netting failure. Conclusions: We identified that diagnostic errors were divided into four diagnostic failure modes that mapped to distinct DEER failure patterns and other factors. Linking these failure modes to measurable contextual variables may provide targets for diagnostic safety programs.