Improving the performance of the Revised Trauma Score using Shock Index, Peripheral Oxygen Saturation, and Temperature–a National Trauma Database study 2011 to 2015

Background: The Revised Trauma Score is the standard physiologic injury severity indicator used in trauma research and quality control. Shock index, peripheral oxygen saturation, and temperature have emerged as strong predictors for mortality and morbidity. We hypothesized that replacing systolic blood pressure and respiratory rate with age-adjusted shock index and peripheral oxygen saturation and adding temperature would generate a more accurate model, valid across all ages. Methods: This is a retrospective database analysis using children and adults from the National Trauma Data Bank for years 2011 to 2015. Glasgow Coma Scale, systolic blood pressure, heart rate, respiratory rate, peripheral oxygen saturation, temperature, and shock index (calculated as heart rate/systolic blood pressure) were used as predictor variables, alone or in combination, in logistic models with survival as primary outcome. Bayesian information criterion and area under the receiver operator characteristic curve were used to compare models’ performances. To adjust for age, models tested on the entire population (children and adults) used Z-scores derived on age-based homogenous intervals rather than the raw value. Results: The analysis included 283,724 pediatric and 1,555,478 adult patients. Overall mortality was 0.7% and 2.7%, respectively. The Glasgow Coma Scale + shock index + peripheral oxygen saturation + temperature model outperformed the revised trauma score in both adults (Bayesian information criterion 296,345.94 vs 298,494.72; area under the receiver operator characteristic curve 0.831 vs 0.809, P < .001) and children (Bayesian information criterion 12,251.48 vs 12,283.48; area under the receiver operator characteristic curve 0.974 vs 0.968, P = .05) cohorts. On the merged (children and adults) cohort the Glasgow Coma Scale + Z-scores derived on age-based homogenous intervals + peripheral oxygen saturation + temperature model outperformed the Revised Trauma Score (Bayesian information criterion 313,814.78 vs 317,781.31; area under the receiver operator characteristic curve 0.852 vs 0.809, P < .001). Conclusions: Replacing systolic blood pressure and respiratory rate with shock index and peripheral oxygen saturation in the Revised Trauma Score model and adding temperature generated a more accurate model in both children and adults. Adjusting shock index for age rendered the model accurate across all ages. Calibration on population-derived nomograms of vital signs would further increase the model's accuracy and precision.