Optimizing Parkinson's Disease progression scales using computational methods

Parkinson's disease is a highly heterogeneous condition with symptoms spanning motor and non-motor domains. Clinical scales like the Movement Disorder Society's Unified Parkinson's Disease Rating Scale (MDS-UPDRS), are standard in clinical trials where disease progression is monitored. They rely on summing item values, assuming uniform item importance and score increments. Here we propose a novel data-driven approach to optimize weights for such scales - so that total scores better reflect the underlying disease severity. By leveraging large-scale longitudinal data from the Parkinson's Progression Markers Initiative (PPMI), our methods identified which items (and value increments) most strongly indicate PD progression, down-weighting or excluding less informative items. The learned weights substantially improve the monotonic relationship between total scores and clinical progression. We validated our weights using both held-out PPMI data and an independent dataset (BeaT-PD), demonstrating their robustness. Applying such weights in clinical trials may increase power and reduce the required sample size.