Impact of performance and information feedback on medical interns' confidence-accuracy calibration

Abstract

Diagnostic errors are a major, largely preventable, patient safety concern. Error interventions cannot feasibly be implemented for every patient that is seen. To identify cases at high risk of error, clinicians should have a good calibration between their perceived and actual accuracy. This experiment studied the impact of feedback on medical interns' calibration and diagnostic process. In a two-phase experiment, 125 medical interns from Dutch University Medical Centers were randomized to receive no feedback (control), feedback on their accuracy (performance feedback), or feedback with additional information on why a certain diagnosis was correct (information feedback) on 20 chest X-rays they diagnosed in a feedback phase. A test phase immediately followed this phase and had all interns diagnose an additional 10 X-rays without feedback. Outcome measures were confidence-accuracy calibration, diagnostic accuracy, confidence, and time to diagnose. Both feedback types improved overall confidence-accuracy calibration (R²_{No Feedback} = 0.05, R²_{Performance Feedback} = 0.12, R²_{Information Feedback} = 0.19), in line with the individual improvements in diagnostic accuracy and confidence. We also report secondary analyses to examine how case difficulty affected calibration. Time to diagnose did not differ between conditions. Feedback improved interns' calibration. However, it is unclear whether this improvement reflects better confidence estimates or an improvement in accuracy. Future research should examine more experienced participants and non-visual specialties. Our results suggest that feedback is an effective intervention that could be beneficial as a tool to improve calibration, especially in cases that are not too difficult for learners.

Keywords: Calibration; Clinical reasoning; Diagnostic error; Feedback; Medical education.

Fig. 1

Study design

Fig. 2

Mean accuracy and confidence results of the test phase per feedback condition. Error bars represent the 95% CI

Fig. 3

Interns’ mean diagnostic accuracy and confidence scores per feedback condition for easy cases. Error bars represent the 95% CI

Fig. 4

Interns’ mean diagnostic accuracy and confidence scores per feedback condition for difficult cases. Error bars represent the 95% CI

Fig. 5

a Filler task in the control condition, b performance feedback, c information feedback. After the experiment, all participants received information feedback (c) on cases they had not previously received feedback for

Fig. 6

Scatterplot of the mean confidence and mean diagnostic accuracy of each participant in the no feedback condition.

Fig. 7

Scatterplot of the mean confidence and mean diagnostic accuracy of each participant in the performance feedback condition

Fig. 8

Scatterplot of the mean confidence and mean diagnostic accuracy of each participant in the information feedback condition