Cognitive mechanisms underlying anchoring bias in diagnosis: a randomized controlled experiment

Anchoring bias occurs when physicians fail to revise an incorrect diagnosis triggered by salient distracting features (SDFs) despite contradictory evidence. The cognitive mechanisms underlying anchoring bias are unclear. We examined differences in processing of biasing (SDFs) and critical diagnostic information (discriminating features, DFs), when physicians give an SDF-related-diagnosis or an accurate diagnosis.

Randomized, crossover experiment with medical residents in Brazil. Each participant diagnosed six clinical vignettes while eye movements were tracked and subsequently recalled each vignette's clinical findings. Each vignette had three versions: without SDFs, with SDFs at the beginning, with SDFs at the end. Each participant diagnosed two vignettes in each version in counterbalanced manner.

Fixations on DFs did not significantly differ between SDF-related-diagnoses and accurate diagnoses nor with the position of SDFs. Fixations on SDFs did not significantly differ with diagnostic accuracy. However, SDFs at the end of the vignette received fewer fixations than when at the beginning. Regardless of the SDFs position, physicians recalled more DFs and fewer SDFs when they gave an accurate diagnosis than an SDF-related-diagnosis (respectively, DFs: 0.47; 95 % CI, 0.40-0.53 vs. 0.20; 95 % CI, 0.06-0.33; SDFs: 0.16; 95 % CI, 0.08-0.24 vs. 0.37; 95 % CI, 0.24-0.50).

DFs and SDFs attracted physicians' attention similarly extensively, but the recall suggests different cognitive processes were taking place. Overcoming anchoring bias apparently required engaging in inhibiting the SDFs while corroborating the diagnostic value of DFs. Clinical teaching should strengthen knowledge of DFs. Digital diagnostic support could possibly be optimized by directing physicians' attention to DFs during diagnosis.