Method: In a mixed factorial design, we presented participants with 3 patient cases on computer. The cases were based on real patients for whom the diagnosis was known. Each case consisted of a number of cues (items of information) presented sequentially and only for 4 seconds each, to increase working memory load. Participants were allocated randomly to one of 3 thinking modes: Immediate response (limited to 20 seconds), Distracted (participants completed an unrelated memory task before diagnosing each case) or Self-paced (participants took as long as they needed to diagnose). After each case, participants gave their diagnosis and indicated their confidence in the diagnosis.
Result: Participants were 116 family physicians. Only 27% of responses were correct, i.e. matched the patients’ real diagnoses. Thinking mode was related neither to diagnostic accuracy (p=0.43) nor to confidence (p=0.15). Physicians in the Self-paced condition did not take time to think and tended to diagnose within seconds (median 7 seconds). A significant, inverse relationship was found between diagnostic accuracy and confidence (t=3.03, df=329, p<0.01).
Conclusion: The study did not replicate the DWA in medical diagnosis, despite maintaining the conditions of the experimental paradigm. It thus resolves the uncertainty surrounding the effect on clinical diagnosis. The quick responses of physicians in the Self-paced condition and the lack of differences in accuracy suggest that all three groups employed similar cognitive processes, constructing their diagnoses "online" rather than at the end. The period of distraction did not improve information processing. The higher confidence associated with inaccurate diagnoses suggests that high confidence in an initial diagnosis may discourage physicians from revising it to account for subsequent, inconsistent information.