Classification of health level from chronic pain self reporting

This paper proposes an approach to identify patients' health levels based on the information gathered following a process of self reporting based on the patient's current condition. The goal of approach is the accurate provision of information to assist with self management of chronic pain. Four supervised classifiers, namely decision tree, naive Bayes, support vector machine and multilayer perceptron, have been applied to classify the health level of patients suffering from chronic pain based on information collected from self reports from three treatment stages - pre-treatment stage, post-treatment stage and 3-month follow-up stage. Three binary classification problems, i.e. pre-treatment vs. post-treatment, pre-treatment vs. 3-month follow-up and post-treatment vs. 3-month follow-up, were investigated. The classification accuracy and area under Receiver Operating Characteristics (ROC) curve ranged from 66.7% 94.7% and 0.689 0.989 respectively. The multilayer perceptron classifier achieved the best performance with a classification accuracy of 94.7% and area under ROC curve of 0.981 for the pre-treatment vs. post-treatment classification. The results from this study have demonstrated that it is feasible to apply automated classification techniques to identify patients' health level from their self reports. This data may be used as an important indicator in automated approaches to chronic disease self management, an area which is currently receiving much attention. Further work will investigate the presence of optimal features derived from questionnaires to improve the classification performance.