Accuracy of physician gestalt in prediction of significant abdominal and pelvic injury in adult blunt trauma patients.

OBJECTIVE

Focusing on potential missed injury rates and sensitivity of low-risk of injury predictions, we sought to evaluate the accuracy of physician gestalt in predicting clinically significant injury (CSI) in the abdomen and pelvis among blunt trauma patients presenting to the emergency department (ED).

METHODS

We collected gestalt data on physicians caring for adult blunt trauma patients who received abdominal/pelvic computed tomography (CT) at three Level I and one Level II trauma centers. The primary outcome of CSI was defined as injury on abdominal/pelvic CT requiring hospitalization or intervention. Physicians evaluating trauma patients estimated the likelihood of CSI prior to abdominal/pelvic CT review (response choices: <2%, 2%-10%, 11%-20%, 21%-40%, >40%). We evaluated potential missed injury rates (prevalence of CSI) and sensitivity for prediction categories, as well as calibration and area under the receiver operating characteristic (AUROC) curve for overall physician gestalt.

RESULTS

Of 2030 patients, 402 (20%) had an injury on abdominal/pelvic CT and 270 (13%) had CSI. The <2% risk of CSI gestalt cutoff had a potential missed injury rate of 5.6% and a sensitivity of 95.2% (95% confidence interval [CI] 91.7%-97.3%). The 0%-10% cutoff of CSI gestalt had a potential missed injury rate of 6.3% (95% CI, 5.0%-7.9%) and a sensitivity of 75.2% (95% CI 69.5%-80.1%). With an overall AUROC of 0.699 (95% CI 0.679-0.719), physician gestalt was moderately accurate and calibrated for the midranges of predicted risk but poorly calibrated at the extremes.

CONCLUSIONS

Physician gestalt for the prediction of adult abdominal and pelvic CSI is moderately accurate and calibrated. However, the potential missed CSI rate and low sensitivity of the low perceived risk of injury cutoffs indicate that gestalt by itself is insufficient to direct selective abdominal/pelvic CT use in adult blunt trauma patient evaluation.