Use of a structured approach and virtual simulation practice to improve diagnostic reasoning.

Michael Dekhtyar, University of Illinois
Yoon Soo Park, Harvard University
Judy Kalinyak, i-Human Patients
Saumil M. Chudgar, Duke University
Kelly Bossenbroek Fedoriw, University of North Carolina
Khadeja J. Johnson, Morehouse School of Medicine
Craig F. Knoche, i-Human Patients
Lisa Martinez, Florida Atlantic University
Nina Mingioni, Thomas Jefferson University
Amber T. Pincavage, University of Chicago
Rachel Salas, Johns Hopkins Medicine
Fred Sanfilippo, Emory University
Stephen M. Sozio, Johns Hopkins University
Nancy Weigle, Duke University
Sarah Wood, Florida Atlantic University
Jillian Zavodnick, Thomas Jefferson University
Scott Stern, University of Chicago

This article is the authors’ final published version in Diagnosis, July 2021.

The published version is available at Copyright © Walter de Gruyter


OBJECTIVES: In 2015, the National Academy of Medicine IOM estimated that 12 million patients were misdiagnosed annually. This suggests that despite prolonged training in medical school and residency there remains a need to improve diagnostic reasoning education. This study evaluates a new approach.

METHODS: A total of 285 medical students were enrolled in this 8 center, IRB approved trial. Students were randomized to receive training in either abdominal pain (AP) or loss of consciousness (LOC). Baseline diagnostic accuracy of the two different symptoms was assessed by completing a multiple-choice question (MCQ) examination and virtual patient encounters. Following a structured educational intervention, including a lecture on the diagnostic approach to that symptom and three virtual patient practice cases, each student was re-assessed.

RESULTS: The change in diagnostic accuracy on virtual patient encounters was compared between (1) baseline and post intervention and (2) post intervention students trained in the prescribed symptom vs. the alternate symptom (controls). The completeness of the student's differential diagnosis was also compared. Comparison of proportions were conducted using χ 2-tests. Mixed-effects regressions were used to examine differences accounting for case and repeated measures. Compared with baseline, both the AP and LOC groups had marked post-intervention improvements in obtaining a correct final diagnosis; a 27% absolute improvement in the AP group (p<0.001) and a 32% absolute improvement in the LOC group (p<0.001). Compared with controls (the groups trained in the alternate symptoms), the rate of correct diagnoses increased by 13% but was not statistically significant (p=0.132). The completeness and efficiency of the differential diagnoses increased by 16% (β=0.37, p<0.001) and 17% respectively (β=0.45, p<0.001).

CONCLUSIONS: The study showed that a virtual patient platform combined with a diagnostic reasoning framework could be used for education and diagnostic assessment and improved correct diagnosis compared with baseline performance in a simulated platform.