Application of Classification Tree Modeling in the Development of Adult Vehicular Trauma Triage Decision Rules
John E. Kolassa, Rutgers University 
Linda Scheetz, SUNY New Paltz 
*Jane Zhang, Bristol-Myers Squibb Company 

Keywords: Data mining, trauma triage

Data mining techniques are gaining in popularity among investigators studying health outcomes. Classification tree analysis is a data mining technique known for its simplicity of interpretation, ability to explore sophisticated interactions among variables and flexibility to handle complex data repository such as data with missing values. The classification tree decision rules can be used to predict rates of certain events in clinical practice and may have impact on existing health policies. In this talk, the application of classification tree analysis to the National Automotive Sampling System Crashworthiness Data System database will be presented. Across all age groups in the United States, vehicular crashes rank among the leading causes of death and disability. Crash scene identification of severely injured persons who should be transported to a trauma center challenges emergency care providers nationwide. Algorithms designed to enable such identification have fallen short of targeted sensitivity goals, particularly for older adults. Construction and validation of a classification tree model, including the concepts of variable splitting, prior probabilities and cost ratio will be discussed. The relevance of missing data and how to handle it will also be addressed.