A New CPXR Based Logistic Regression Method and Clinical Prognostic Modeling Results Using the Method on Traumatic Brain Injury

Prognostic modeling is central to medicine, as it is often used to predict patients´ outcome and response to treatments and to identify important medical risk factors. Logistic regression is one of the most used approaches for clinical prediction modeling. Traumatic brain injury (TBI) is an important public health issue and a leading cause of death and disability worldwide. In this study, we adapt CPXR (Contrast Pattern Aided Regression, a recently introduced regression method), to develop a new logistic regression method called CPXR (Log), for general binary outcome prediction (including prognostic modeling), and we use the method to carry out prognostic modeling for TBI using admission time data. The models produced by CPXR (Log) achieved AUC as high as 0.93 and specificity as high as 0.97, much better than those reported by previous studies. Our method produced interpretable prediction models for diverse patient groups for TBI, which show that different kinds of patients should be evaluated differently for TBI outcome prediction and the odds ratios of some predictor variables differ significantly from those given by previous studies, such results can be valuable to physicians.