Selective processing of clinical information related to correct and incorrect diagnoses: An eye-tracking experiment

Diagnostic errors are often attributed to erroneous selection and interpretation of patients' clinical information, due to either cognitive biases or knowledge deficits. However, whether the selection or processing of clinical information differs between correct and incorrect diagnoses in written clinical cases remains unclear. We hypothesised that residents would spend more time processing clinical information that was relevant to their final diagnosis, regardless of whether their diagnosis was correct.

In this within-subjects eye-tracking experiment, 19 internal or emergency medicine residents diagnosed 12 written cases. Half the cases contained a correct diagnostic suggestion and the others an incorrect suggestion. We measured how often (i.e. number of fixations) and how long (i.e. dwell time) residents attended to clinical information relevant for either suggestion. Additionally, we measured confidence and time to diagnose in each case.

Residents looked longer and more often at clinical information relevant for the correct diagnostic suggestion if they received an incorrect suggestion and were able to revise this suggestion to the correct diagnosis (dwell time: M: 6.3 seconds, SD: 5.1 seconds; compared to an average of 4 seconds in other conditions; number of fixations: M: 25 fixations, SD: 20; compared to an average of 16-17 fixations). Accordingly, time to diagnose was longer in cases with an incorrect diagnostic suggestion (M: 86 seconds, SD: 47 seconds; compared to an average of 70 seconds in other conditions). Confidence (range: 64%-67%) did not differ depending on residents' accuracy or the diagnostic suggestion.

Selectivity in information processing was not directly associated with an increase in diagnostic errors but rather seemed related to recognising and revising a biased suggestion in favour of the correct diagnosis. This could indicate an important role for case-specific knowledge in avoiding biases and diagnostic errors. Future research should examine information processing for other types of clinical information.