Comparison of Different Data Mining Methods to Determine Disease Progression in Dissimilar Groups of Parkinson’s Patients

Abstract

Parkinson’s disease (PD) is the second after Alzheimer’s most popular neurodegenerative disease (ND). Cures for both NDs are currently unavailable. OBJECTIVE: The purpose of our study was to predict the results of different PD patients’ treatments in order to find an optimal one. METHODS: We have compared rough sets (RS) and others, in short, machine learning (ML) models to describe and predict disease progression expressed as UPDRS values (Unified Parkinson’s Disease Rating Scale) in three groups of Parkinson’s patients: 23 BMT (Best Medical Treatment) patients on medication; 24 DBS patients on medication and on DBS therapy (Deep Brain Stimulation) after surgery performed during our study; and 15 POP (Postoperative) patients who had had surgery earlier (before the beginning of our research). Every PD patient had three visits approximately every six months. The first visit for DBS patients was before surgery. On the basis of the following condition attributes: disease duration, saccadic eye movement parameters, and neuropsychological tests: PDQ39 (Parkinson’s Disease Questionnaire - disease-specific health-related quality-of-life questionnaire), and Epworth Sleepiness Scale tests we have estimated UPDRS changes (as the decision attribute). RESULTS: By means of RS rules obtained for the first visit of BMT/DBS/POP patients, we have predicted UPDRS values in the following year (two visits) with global accuracy of 70% for both BMT visits; 56% for DBS, and 67%, 79% for POP second and third visits. The accuracy obtained by ML models was generally in the same range, but it was calculated separately for different sessions (MedOFF/MedON). We have used RS rules obtained in BMT patients to predict UPDRS of DBS patients; for the first session DBSW1: global accuracy was 64%, for the second DBSW2: 85% and the third DBSW3: 74% but only for DBS patients during stimulation-ON. ML models gave better accuracy for DBSW1/W2 session S1(MedOFF): 88%, but inferior results for session S3 (MedON): 58% and 54%. Both RS and ML could not predict UPDRS in DBS patients during stimulation-OFF visits because of differences in UPDRS. By using RS rules from BMT or DBS patients we could not predict UPDRS of POP group, but with certain limitations (only for MedON), we derived such predictions for the POP group from results of DBS patients by using ML models (60%). SIGNIFICANCE: Thanks to our RS and ML methods, we were able to predict Parkinson’s disease (PD) progression in dissimilar groups of patients with different treatments. It might lead, in the future, to the discovery of universal rules of PD progression and optimise the treatment.