Diagnosis Behaviors of Physicians and Non-Physicians When Supported by an Electronic Differential Diagnosis Aid

Abstract

The present study examined a diagnostic medical decision aid developed to help inexperienced operators to diagnose and treat a simulated patient. Diagnosis and treatment accuracy using the tool were assessed and compared across both physicians and non-physicians. Initial analysis revealed more accurate diagnostic and treatment choices for non-physicians, but upon further investigation, physicians were found to have recognized signs for another diagnosis and correctly diagnosed and treated based on the limited information in the patient simulation. This fit with other noted behaviors, such as non-physicians opening the diagnostic support tool within the aid more often than physicians, and frequently returning to the tool during the task. In general, non-physicians were supported in choosing the correct diagnosis and treatment by the aid, while physicians disregarded the aid’s recommendations to make decisions based on their own expertise. These results have implications for the development of future decision support aids for non-physicians performing medical procedures.

Research Trends in Artificial Intelligence Applications in Human Factors Health Care: Mapping Review

“…A study included in our review, using a poststudy questionnaire, found that doctors (pathologists) expressed higher trust in SMILY, an AI-based application, due to its better performance, interface, and higher benevolence compared with the conventional app [ 27 ]. By contrast, another study reported lower trust of experienced physicians in an AI-based recommendation tool due to its inefficient performance [ 54 ]. Based on patient data, expert physicians were able to identify the alternative and better explanation for patient health compared to the AI-based tool [ 54 ].…”

“…By contrast, another study reported lower trust of experienced physicians in an AI-based recommendation tool due to its inefficient performance [ 54 ]. Based on patient data, expert physicians were able to identify the alternative and better explanation for patient health compared to the AI-based tool [ 54 ]. A recent study identified the impact of the AI interface on user’s trust [ 68 ].…”

Trust in Clinical AI: Expanding the Unit of Analysis

“…The AI could perform better on certain tasks than others (e.g., diagnosis of different types of lesions) and the clinician would need to adjust their trust as these cases arose [25]. Further, specific environmental contexts can cause the AI to not perform well and reduce trust [8,25,33,98]. For instance, Beede et al found that poor image quality severely impacted usage of their deep learning system that made assessments based on the image of the eye [90].…”