Comparing the Predictive Performances of Rule Based, Regression and Artificial Neural Network Models within a Managed Care Population

Keywords:Rule Base System, Ordinal Regression, Artificial Neural Network

Abstract: The assessment of patient outcomes is important in the efforts to contain costs, streamline patient management and codify medical practices. As such, it is necessary to develop predictive models that will make accurate predictions of these outcomes. A study is developed to examine the feasibility of predicting patient outcomes by creating discrete models from the application of statistics and artificial intelligence (primarily machine learning). It focuses on the determination of four diagnostic categories of asthma (Definitive Diagnosis; High Probable Diagnosis; Probable Diagnosis and Suspicious Diagnosis) within a managed care patient population. There are three phases to the study. In Phase One, the patient population is divided into training and testing sets. Using the training set, a rule based system is developed to classify the patient population into the four diagnostic categories. It is validated by comparing the outcomes of a subset against the professional diagnoses of the attending physician. In Phase Two, ordinal logistic regression and artificial neural network models are created based upon the predictors identified by the knowledge base. In Phase Three, the models are used to predict outcomes within the testing set. The predictive performances of the rule base, regression and artificial neural network models are then compared.