Decision Support in Trauma Management: Predicting Potential Cases of Ventilator Associated Pneumonia

Ventilator Associated Pneumonia (VAP) is a complication of intubated trauma patients and a leading cause in Intensive Care Unit (ICU) mortality. Since early diagnosis, by specimen culture takes days to complete, an overuse of broad spectrum antibiotics is the usual treatment. As a result there is the risk of developing antibiotic resistant strains. Using an Artificial Neural Network (ANN) derived model to predict those at risk would result in reduced risk of resistant strains, a lowering of mortality rates and considerable savings in treatment costs. Artificial Neural Networks work well on classification problems, using feed-forward/back propagation methodology. Using the National Trauma Data Bank (V6.2) data files, Tiberius Software created the ANN models. Best models were identified by their Gini co-efficient, ability to predict the complication outcome selected, and their RMSE scores. The model ensemble for the complications recorded in the registry were determined, variables ranked and model accuracy recorded. Results show an effective model, able to predict to 85% of those likely to contract VAP and similar figures for those unlikely to contract VAP. This equates to 1 in 10 patients being missed, and 1 in 10 falsely being flagged for treatment. Important variables in model development are not related to physiological factors, but injury status and the treatment received (intubation and expected ICU stay more than 2 days). Application of a predictive model could reduce the number of false positives being treated in an ICU and identify those most at risk, thereby lowering treatment costs and potentially helping improve mortality rates.