Abstract
Failure to appreciate the importance of the frequency of a disorder in the appropriate population (the base rate) may lead to the misinterpretation of the diagnostic significance of unexpected test results (unexpected test result defined in this context as a test result that is positive in a higher proportion of cases of an alternative diagnosis than in the diagnosis considered most likely before the test). This study aimed to determine whether pathologists are vulnerable to this error. Pathologists were asked to estimate the probability of tumour B in a scenario in which, prior to the immunostaining result, an experienced pathologist considers there to be a 99% chance that the patient has tumour A and a 1% chance that they have tumour B. Antibody X is positive in 80% of cases of tumour B and negative in 90% of cases of tumour A and is positive in the case described in the scenario. The estimates made by consultant pathologists ranged from 0 to 100% (mean 29.7%). The Bayesian answer would be 7.5%. These findings suggest that base-rate error may lead some pathologists to overestimate the implications for the likelihood of a diagnosis in the light of an unexpected immunohistochemical result.
Highlights
Base-rate error in medical diagnosis refers to the cognitive bias in which doctors may underestimate the importance of the frequency of the relevant disorder in the appropriate population when considering the implications for the diagnosis of the result of tests which have less than 100% sensitivity and specificity
79.2% of patients with breast cancer are assumed to have a malignant result on mammography and 90.4% of patients without cancer to have a benign mammogram result
Number responding to survey Number answering question Median predicted probability of tumour B (%) Mean predicted probability of tumour B (%) Range (%) Standard deviation
Summary
Base-rate error in medical diagnosis refers to the cognitive bias in which doctors may underestimate the importance of the frequency of the relevant disorder in the appropriate population (the pretest probability) when considering the implications for the diagnosis of the result of tests which have less than 100% sensitivity and specificity. Eddy has previously reported base-rate error in the interpretation of the diagnostic contribution of mammogram results by physicians [3] In his scenario, 79.2% of patients with breast cancer are assumed to have a malignant result on mammography (sensitivity) and 90.4% of patients without cancer to have a benign mammogram result (specificity). He asked what he described as “an informal group of approximately 100 physicians” what the probability of a patient having breast cancer would be if they had a 1% risk of the disease prior to the mammogram but had a malignant diagnosis on mammography In this scenario, the mathematically correct answer (by Bayesian analysis) would by 7.7%. He reported that approximately 95% of physicians estimated the risk as approximately 75% (almost 10 times the actual risk)
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have