The impact of hindsight bias on the diagnosis of perioperative events by anesthesia providers: A multicenter randomized crossover study

Abstract

Study objective: Hindsight bias is the tendency to overestimate the predictability of an event after it has already occurred. We aimed to evaluate whether hindsight bias influences the retrospective interpretation of clinical scenarios in the field of anesthesiology, which relies on clinicians making rapid decisions in the setting of perioperative adverse events.

Design: Two clinical scenarios were developed (intraoperative hypotension and intraoperative hypoxia) with 3 potential diagnoses for each. Participants completed a crossover study reviewing one case without being informed of the supposed ultimate diagnosis (i.e., no 'anchor' diagnosis), referred to as their foresight case, and the other as a hindsight case wherein they were informed in the leading sentence of the scenario that 1 of the 3 conditions provided was the ultimate diagnosis (i.e., the diagnosis the participant might 'anchor' to if given this information at the start). Participants were randomly assigned to (1) which scenario (hypotension or hypoxia) was presented as the initial foresight case and (2) which of the 3 potential diagnoses for the second case (the hindsight case, which defaulted to whichever case the participant was not assigned for the first case) was presented as the ultimate diagnosis in the leading sentence in a 2 (scenario order) x 3 (hindsight case anchor) between-subjects factorial design (6 possible randomization assignments).

Setting: Two academic medical centers.

Participants: Faculty, fellow, and resident anesthesiologists and certified nurse anesthetists (CRNAs).

Interventions: None.

Measurements: After reading each clinical scenario, participants were asked to rate the probability (%) of each of three potential diagnoses to have caused the hypotension or hypoxia. Compositional data analysis (CoDA) was used to compare whether diagnosis probabilities differ between the hindsight and the foresight case.

Main results: 113 participants completed the study. 59 participants (52%) were resident anesthesiologists. Participants randomized to the hypotension scenario as a hindsight case were 2.82 times more likely to assign higher probability to the pulmonary embolus diagnosis if provided as an anchor (95% CI, 1.35-5.90; P = 0.006) and twice as likely to assign higher probability to the myocardial infarction diagnosis if provided as an anchor (95% CI, 1.12-3.58; P = 0.020). Participants randomized to the hypoxia scenario as a hindsight case were 1.78 times more likely to assign higher probability to the mainstem bronchus intubation diagnosis if provided in the anchor statement (95% CI, 1.00-3.14; P = 0.048) and 3.72 times more likely to assign higher probability to the pulmonary edema diagnosis if provided as an anchor (95% CI, 1.88-7.35; P < 0.001).

Conclusions: Hindsight bias influences the clinical diagnosis probabilities assigned by anesthesia providers. Clinicians should be educated on hindsight bias in perioperative medicine and be cognizant of the effect of hindsight bias when interpreting clinical outcomes.

Keywords: Clinical diagnosis; Cognitive bias; Hindsight bias; Perioperative events.