Am I Right When I Am Sure? Data Consistency Influences the Relationship Between Diagnostic Accuracy and Certainty

PURPOSE: When gauging diagnostic accuracy cognitive biases may lead to inaccurate estimates of certainty, predisposing clinicians to diagnostic errors. This study explored the relationship between diagnostic accuracy and measures of certainty for diagnoses based on consistent or inconsistent information. METHOD: The authors analyzed three experiments among 180 to 190 postgraduate trainees performing cardiac physical diagnoses using a simulator from 2010 to 2012. Each asked participants to assess diagnostic certainty. One experiment used a seven-point certainty scale and provided only simulated physical findings. Two assessed certainty continuously (probability 1%-100%) and included cases with inconsistent clinical information in addition to simulated physical findings. Relationships between certainty and accuracy were explored through descriptive statistics and nonparametric tests. RESULTS: Measures of certainty ranged widely (between 2 and 7, and 5%-100%). Relationships between accuracy and certainty varied depending on information consistency. In experiments providing only simulated findings, or consistent clinical data, diagnostic accuracy was associated with higher certainty (median 90% versus 75%, and 5/7 versus 4/7, both P < .001). Studies providing inconsistent data generated similar certainty among participants regardless of accuracy (median 75% versus 75%, P = .36; and 80% versus 85%, P = .60). CONCLUSIONS: Diagnostic accuracy was moderately associated with higher certainty only when clinical data were consistent. This correlation disappeared when incon sistent data were provided, possi bly reflecting changes in reasoning strategies among diagnostically success ful trainees. The relationship between certainty and diagnostic accuracy is context dependent. Certainty is an unreliable surrogate for diagnostic accuracy.