Using Data Mining Techniques to Extract Clinical Disorders Affecting Mortality in Trauma Patients

Trauma is one of the most common causes of death in the world, which often occurs as a result of road accidents. Prompt identification of patients with acute injury, leads to take the appropriate medical actions and thus, save lives and also avoid enormous cost of treatment. Finding the best data mining algorithms to identify clinical disorders resulting in death in trauma patients. 1,073 trauma patients hospitalized in Poursina Hospital in Rasht with their 52 recorded clinical conditions (features) have been analyzed in this research. In order to automatically identify emergency cases, a number of classification algorithms have been modified for the task, such as decision tree, K-nearest neighbor, and neural network methods. These algorithms have been trained over a wide range of features and their performance has been investigated using 10-fold cross validation. Totally, 82.8% (888) of the surveyed patients were male and17.2% (185) were female. 22.1% died, most of them (30%) in the first week after their hospitalization and 23.6% on the first day. No significant relationship has been found between the duration of hospitalization and mortality. Among the classification algorithms, decision tree and k-nearest were able to recognize death cases with higher precision, (i.e. 91% and 89%, respectively). In order to find effective factors on training a better Decision Tree classifier, the Best First algorithm was used which then selected and could identify 18 effective features (of 52 initial features). Given the high accuracy of some data mining algorithms, like Decision Tree algorithm, we are able to differentiate severe trauma cases which may lead to death from those with mild injuries. Hence, their application to predict mortality in trauma patients and identify those at life risk can be investigated in real environment.