61 Gestalt in Acute Trauma: Evaluating Clinician Accuracy in Predicting Abdominal Injuries Among Trauma Patients

Abstract

We sought to evaluate the accuracy of physician gestalt in predicting clinically significant abdominal injuries among trauma patients presenting to the emergency department (ED) by comparing physician predictions to patient outcomes based on computed tomography (CT) reports and hospital course, including admissions and procedures. As part of an observational multi-center prospective study of CT use in acute abdominal trauma, we surveyed physicians regarding their estimated likelihood of clinically significant injuries. The study was conducted from January 2017 - March 2020 at EDs within three major trauma centers. Patients presenting with blunt trauma who received abdominal CTs were included in the study. Exclusion criteria consisted of penetrating trauma, presentation to ED > 72 hours after trauma, age <15, occurrence of emergency procedures prior to abdominal radiography, and viewing of radiographic scans or reports by physicians prior to patient evaluation. The primary outcome was clinically significant injury, defined as an injury on CT requiring either hospital admission or procedure. The primary predictor was clinician gestalt: physicians were asked to estimate the likelihood of clinically significant injury, with responses including <2%, 2-10%, 10-20%, 20-40%, >40%. We calculated sensitivity, specificity, interval likelihood ratios (LR), injury proportions, and the area under the receiver operating characteristic curve (AUROC). Of 1677 patients, 326 (19%) had an injury on CT and 222 (13%) had a clinically significant injury. The proportion of clinically significant injuries increased with estimated risk of injury: of those estimated to have <2% risk of injury, 5.6% had a clinically significant injury and among those estimated to have >40% risk, 29.8% had a clinically significant injury. When comparing physician predictions across estimates, the interval LR associated with clinically significant injuries increased with increasing suspicion of injury, with the interval LR for <2% likelihood of injury at 0.38, the LR for 2-10% at 0.45, the LR for 10-20% at 1.2, the LR for 20-40% at 1.4 and the LR for >40% at 2.8. The sensitivity for detecting clinically significant injuries using <2% as the threshold for low risk of injury was 95%, while the sensitivity for detecting clinically significant injuries using <10% as the threshold was 75%. Employing a <2% threshold, attending physicians showed 100% sensitivity and resident physicians showed 95% sensitivity. Using a <10% threshold, the sensitivity was 100% among attending physicians and 73% among resident physicians. Clinician gestalt tracked monotonically with patient outcomes, but physicians overestimated likelihood of clinically significant injuries at the high end of predictions and underestimated likelihood at the low end of predictions. Results showed that clinically significant injuries were found even when clinicians predicted the likelihood of injury as <2%, which suggests that physician gestalt may not demonstrate sufficient sensitivity in detecting clinically significant injuries and should not be solely relied upon to determine whether or not to CT trauma patients. A clinical decision aid demonstrating higher accuracy than physician gestalt in predicting clinically significant injuries may help reduce the number of missed cases.View Large Image Figure ViewerDownload Hi-res image Download (PPT)