Presented at the Mining Big Data to Improve Clinical Effectiveness Workshop at the 2015 IEEE International Conference on Big Data, San Jose, CA (October 2015).

Trauma patients suffer from a wide range of injuries, including vascular injuries. Such injuries can be difficult to immediately identify, only becoming detectable after repeated examinations and procedures. Large data sets of Shock Trauma patient treatment and care exist, spanning thousands to millions of patients, but machine learning techniques are needed to analyze this data and build appropriate models for predicting patient injury and outcome. We developed an initial approach for ensemble prediction of vascular injury in trauma care to aid doctors and medical staff in predicting injury and aiding in patient recovery. Of the classifiers tested, we found that stacked ensemble classifiers provided the best predictions. Prediction accuracy varied among vascular injuries (sensitivity ranging from 1.0 to 0.21), but demonstrated the feasibility of the approach for use on massive clinical datasets.