GAINING INSIGHT INTO DIAGNOSTIC ERROR: OBSERVING HOW PHYSICIANS DETECT AND CORRECT MISDIAGNOSIS IN REAL TIME

Background: Diagnostic error remains a source of significant harm to patients^1-5. Yet, while much is known about how physicians arrive at a diagnosis^6-10 and why physicians make errors in diagnosis^11-14, we know relatively little about how diagnostic error evolves over time and how physicians detect and correct diagnostic error¹⁵. To better understand diagnostic error evolution, detection, and correction, scholars need to study the process of diagnosis as it unfolds in real time. We identify the medical simulation laboratory as an underutilized context for studying diagnostic error and highlight two methodological approaches – process tracing and video ethnography – which are well-suited to providing a rich description of evolving diagnostic error, including error detection and correction.

Methods: Increasingly, medical simulation labs are being used to train and evaluate complex skills and behaviors (i.e. communication in teams)^16-18 but these labs can also be used to study complex emergent processes, like diagnosis. In the lab, researchers can assess the accuracy of the diagnostic problem solving and complex clinical scenarios can be standardized, recorded, and analyzed so that diagnostic performance can be compared across participants. Inductive methodologies, such as process tracing or video ethnography, are useful for investigating under-theorized topics. Process tracing gives insight into an individual's cognitive activities during a complex work task^{19, 20} and is useful for understanding how emergent processes evolve over time. Video ethnography involves logging, transcribing, and coding video and focuses on understanding how people use language, action, and tools to make sense together from moment to moment^{21, 22}.

Results: Using process tracing, Rudolph^{23, 24} studied anaesthesiology residents and found that only 23% of the residents appropriately diagnosed the patient during a standardized crisis in a high-fidelity OR simulation. This study identified three modes of diagnostic failure: residents either had difficulty generating differential diagnoses to test (5%); had difficulty moving away from an erroneous earlier diagnosis (28%); or moved between various diagnoses without adequately evaluating them (44%). Using video ethnography, Christianson²⁵ studied ED/trauma teams and found that 58% of teams correctly diagnosed the patient in standardized ED scenario and that the time to correct diagnosis ranged from 30 seconds to 16 minutes. Variations in performance were linked to differing levels of interactional competence in teams.

Conclusion: Studying diagnosis in the simulation lab using inductive methodologies such as process tracing or video ethnography holds great promise for developing theory around how diagnostic error evolves and is corrected. (References available upon request)