Impact of hindsight bias on interpretation of nonenhanced computed tomographic head scans for acute stroke.

摘要

OBJECTIVE: To determine whether knowledge of the presence or the absence of acute stroke from review of follow-up studies, including diffusion-weighted magnetic resonance imaging (DW-MRI) head scans, biases the retrospective review of a radiologist's detection of acute stroke on nonenhanced computed tomographic (CT) scans of the head. METHODS: Forty-four consecutive patients with acute cerebral infarction as defined by positive DW-MRI and a CT scan within 48 hours before the MRI were identified. In addition, 30 control patients with negative DW-MRI results and a head CT scan within 48 hours before the MRI were chosen. Medical records were examined, and the time from onset of symptoms was recorded when this was available. The CT examinations were shown to 6 board-certified radiologists (of which 4 were certificate of added qualification neuroradiologists). The radiologists were asked to provide a probability estimate on whether an acute stroke is present or absent, the location of the stroke if present, and the presence of early ischemic signs including parenchymal hypoattenuation, sulcal effacement, and loss of gray-white matter differentiation. The radiologists had access to the patients' age, sex, and date of study but were not provided with the presenting history. After a minimum 10-day washout period and after reshuffling the cases, the radiologists were asked to reinterpret the examination results after first being shown the DW-MRI of the study in question and allowed access to reports and other follow-up studies. Receiver operating characteristic analysis was performed. RESULTS: For infarctions in the middle cerebral artery, the posterior fossa, and anterior cerebral artery territories, there was a significant increase in performance with knowledge of DW-MRI results than without prior knowledge of the DW-MRI findings. When the basal nuclei and the posterior cerebral artery were included, the difference in performance of the readers without and those with DW-MRI approached significance (t = 2.35, P = 0.0571). CONCLUSIONS: Knowledge of the presence or the absence of an acute infarct influences a radiologist's ability to detect the lesion on CT. The presence of hindsight bias may influence the testimony of expert witnesses in legal cases and have an impact on quality assurance procedures on a hospital level.