Machine Learning to Monitor Diagnostic Safety Risks in Emergency Departments: A Study Protocol (Preprint)
BACKGROUND Diagnostic decision-making, especially in emergency departments (EDs), is a highly complex cognitive process involving uncertainty and susceptibility to error. A combination of parameters including patient factors (e.g. history, behaviors, complexity, and comorbidity), provider/care-team factors (e.g. cognitive load, information gathering, and synthesis), and system factors (e.g. health information technology, crowding, shift-based work, and interruptions) may contribute to diagnostic errors. Records with potential diagnostic errors have been identified using electronic triggers that flag certain patterns of care (i.e., triggers), such as the escalation of care or death after ED discharge. Sophisticated data analytics and machine learning techniques that can be applied to existing electronic health record (EHR) datasets could shed light on potential risk factors influencing diagnostic decision-making. OBJECTIVE To identify variables contributing to potential diagnostic errors in the ED using large scale EHR data. METHODS We will apply trigger algorithms to EHR data repositories to generate a large dataset of trigger-positive and trigger-negative encounters. Samples from both sets will be validated using medical record reviews where we expect to find a higher number of diagnostic safety problems in the trigger positive subset. Advanced data mining and machine learning techniques will be used to evaluate relationships between certain patient, provider/care-team, and system risk factors and diagnostic safety signals in the statistically matched groups of trigger-positive and trigger-negative charts. RESULTS This study received funding in February 2019, and is approved by the Institutional Review Board at two health systems. Trigger queries are being developed at both organizations and sample cohorts are being labeled using the triggers. Once completed, study data can inform important parameters for future clinical decision support systems to help identify risks that contribute to diagnostic errors. CONCLUSIONS Using large datasets to investigate risk factors (patient, provider/care team, and system-level) in the diagnostic process can provide mechanisms for future monitoring of diagnostic safety.