Statistical Models of Parkinson's Disease Progression: Predictive Validity in a 3-Year Follow-up.

Abstract

The rate of Parkinson's disease (PD) progression varies widely between patients. Current knowledge does not allow to accurately predict the evolution of symptoms in a given individual over time. To develop regression-based models of PD progression and to explore its predictive value in a three-year follow-up. At baseline, 300 consecutive PD patients were assessed using the Unified Parkinson's Disease Rating Scale (UPDRS) - subscales II and III, Hoehn & Yahr (H&Y) and Schwab and England Independence Scale (S&E); and the Freezing of Gait Questionnaire (FOG-Q). UPDRS-III and H&Y were applied in OFF and ON medication conditions. An axial index was derived from the UPDRS-III. Based on multiple linear regression coefficients, algorithms were developed to adjust test scores to the characteristics of each individual. Sixty-eight patients were reevaluated three years later. In the construction of the models, disease duration, age ≥70, age at disease onset ≥55, tremor as the first symptom alone, and medication description explained between 35% (UPDRS-III in ON) and 57% (axial index in ON) of the variance of test scores. The predictive r2 of the models in a 10-fold cross-validation ranged between 33% (UPDRS-III in ON) and 55% (axial index in ON and S&E in OFF). All measures, except UPDRS-III OFF, H&Y ON, and S&E ON, had moderate/good absolute agreement (intraclass correlation coefficient between 0.60 and 0.72) between baseline and follow-up. A cross-sectional assessment of a PD population allowed the development of models of disease progression, whose predictive value was validated on a three-year longitudinal study.