Prediction of critical haemorrhage following trauma: A narrative review

Traumatic haemorrhagic shock can be difficult to diagnose. Models for predicting critical bleeding and massive transfusion have been developed to aid clinicians. The aim of this review is to outline the various available models and report on their performance and validation. Methods: A review of the English and non-English literature in Medline, PubMed and Google Scholar was conducted from 1990 to September 2015. We combined several terms for i) haemorrhage AND ii) prediction, in the setting of iii) trauma. We included models that had at least two data points. We extracted information about the models, their developments, performance and validation. Results: There were 36 different models identified that diagnose critical bleeding, which included a total of 36 unique variables. All models were developed retrospectively. The models performed with variable predictive abilities–the most superior with an area under the receiver operating characteristics curve of 0.985, but included detailed findings on imaging and was based on a small cohort. The most commonly included variable was systolic blood pressure, featuring in all but five models. Pattern or mechanism of injury were used by 16 models. Pathology results were used by 15 models, of which nine included base deficit and eight models included haemoglobin. Imaging was utilised in eight models. Thirteen models were known to be validated, with only one being prospectively validated. Conclusions: Several models for predicting critical bleeding exist, however none were deemed accurate enough to dictate treatment. Potential areas of improvement identified include measures of variability in vital signs and point of care imaging and pathology testing.