Prognostic analysis of hip fracture in elderly women with data mining methods

Hip fracture has become one of the most public health concerns because it is an increasing cause of significant morbidity, mortality, and costs associated with osteoporosis. Although many studies claimed physical activity is the leading cause of hip fracture, other risk factors should be investigated. Most studies in this area employ either descriptive statistical analysis or traditional regression techniques to assess the association between hip fracture and small sets of clinical risk factors. Consequently, not only many potentially important variables such as pre-fracture health conditions are neglected in their analysis, but the results produced by such regression models may not represent relevant risk factors and pattern recognition of hip fracture appropriately. Thus, in this study, we examine whether more complex analytical models using several data mining techniques can better predict and explain the causes of increasing hip fracture in elderly patients. These techniques including logistic regression model, decision tree, and artificial neural network (ANN) are effective ways to analyze data sets with multiple predictor variables, which include both clinical and non-clinical-related risk factors. The preliminary analysis results show that physical activity, traditional medicine, race, BMI, underlying cerebrovascular disease, and alcohol consumption are among the key risk factors.