|| Checking for direct PDF access through Ovid
Delays in appropriate triage of bleeding trauma patients result in poor outcomes. Clinical gestalt is fallible and objective measures of risk stratification are needed. The objective of this review is to identify and assess prediction models and predictors for the early identification of traumatic hemorrhage patients requiring massive transfusion, surgery, or embolization.We searched electronic databases through to September 31, 2016, for studies describing clinical, laboratory, and imaging predictors available within the first hour of resuscitation for identifying patients requiring major intervention for hemorrhage within the first 24 hours.We included 84 studies describing any predictor-outcome association, including 47 multivariable models; of these, 26 (55%) were specifically designed for prediction. We identified 35 distinct predictors of which systolic blood pressure, age, heart rate, and mechanism of injury were most frequently studied. Quality of multivariable models was generally poor with only 21 (45%) meeting a commonly recommended sample size threshold of 10 events per predictor. From 21 models meeting this threshold, we identified seven predictors that were examined in at least two models: mechanism of injury, systolic blood pressure, heart rate, hemoglobin, lactate, and focussed abdominal sonography for trauma. Pooled odds ratios were obtained from random-effects meta-analyses.The majority of traumatic hemorrhagic prediction studies are of poor quality, as assessed by the Prognosis Research Strategy recommendations and Critical Appraisal and Data Extraction for Systematic Reviews of Modeling Studies checklist. There exists a need for a well-designed clinical prediction model for early identification of patients requiring intervention. The variables of clinical importance identified in this review are consistent with recent expert guideline recommendations and may serve as candidates for future derivation studies.