Classification based levodopamine response prediction in parkinson’s disorder

Parkinson’s disease (PD) is a progressive neurodegenerative disorder that affects a sizable fraction of the population and degrades the quality of life. Levodopamine (L-Dopa) is the first-line treatment drug for PD and remains ubiquitously used. However, the drug response prediction of L-Dopa is still an exigent task and there is an unresolved absence of any substantial biomarkers for a robust prediction of L-Dopa response for a robust prediction of L-Dopa response in Parkinson’s disease. The present study intends to develop a robust prediction model to predict the L-Dopa drug response in PD using machine learning approaches. This work intended to utilize the MJFF Levodopa Response Study data of Parkinson’s subjects with conclusive pre-clinical and clinical assessments for resolving the significantly impending task of drug response prediction. The problem was identified as a classification task which employed four different supervised machine learning classification algorithms for data analysis and predictive learning. The underlying task of predictive classification of drug response classified the responders as “good” and “bad,” based on comprehensive analysis on the selected feature space which identified the participants with an improvement in the symptoms as “good” responders and the ones with degraded or no improvement in the symptoms as “bad” responders. The decision tree’s (classification and regression tree) classification accuracy was 88.89% (area under receiver operating characteristics curve 0.9) for predicting levodopamine response. The presence or absence of the symptoms along with Unified Parkinson’s Disease Rating Scale (UPDRS) scores and Hoehn and Yahr scale (H and Y) scores were recognized as the most distinguishing feature subset. The research stipulates the required preliminary evidence to the adaptive advancement of decision trees as an illuminating technique that can facilitate the prediction of the drug treatment response for Parkinson’s disorder, however, an extended effort is necessary to provide efficient predictive performance.