A Machine Learning Prototype for the Identification of Patients who are Not Likely to Respond to Medical Treatment of BPH with alpha-Blockers

This machine learning study aims to identify prior to treatment the patients who are not likely to respond to drug therapy with α-blockers of Benign Prostatic Hyperplasia (BPH). Medical records of 310 BPH patients are used. The data set is divided into two classes, namely positive and negative-response groups, with respect to the change in the patient’s International Prostate Symptom Score (IPSS) after drug therapy. Age, IPSS, Prostate Volume, Post Void Residual Urine (PVR), Prostate Specific Antigen (PSA), free-PSA, and Creatinine are examined as baseline clinical factors, along with our novel synthetic features in order to propose a decision tree to classify a new patient. Our learning scheme, which uses synthetic variable generation and selection routines along with the C4.5 Decision Tree Algorithm correctly classifies 71.5% of patients, with sensitivity 71.3%, and specificity 71.6%. Synthetic variables do enhance its performance. Our prototypical classifier’s success rate reveals that failure is predictable prior to medical treatment of BPH.