Balancing Risk and Benefit in Individualized Treatment Regimens for Parkinson's Disease

About This Grant

Project Summary: Parkinson's disease (PD) is a chronic, progressive neurological disorder marked by heterogeneous symp- toms and outcomes. Clinical decision-making in PD often involves balancing therapeutic benefits (e.g., improved motor function) against significant risks (e.g., long-term motor complications). While individualized treatment reg- imens (ITRs) offer a promising approach, current methodologies fall short in addressing the complex, dynamic trade-offs inherent in PD management. Existing methods for optimizing treatment regimens across benefit and risk outcomes suffer from several key limitations. First, most are restricted to single-time point settings, failing to accommodate the progressive nature of PD and the need for time-adaptive treatment strategies. Second, they often rely on ad hoc composite outcomes to represent benefit-risk trade-offs, which lack a principled structure and impair interpretability. Third, they typically consider only one benefit and one risk outcome, excluding the multifaceted clinical priorities that arise in real-world PD care. Fourth, they do not provide valid statistical in- ference to assess the relative optimality of treatment decisions, leaving clinicians without clear guidance. Fifth, they depend on complex, unstable optimization algorithms that may yield clinically implausible recommendations. Finally, no existing framework accommodates time-to-event or recurrent event outcomes, which are essential for capturing long-term PD progression and cumulative treatment burden. This project closes these gaps by devel- oping novel statistical methods for constructing data-driven ITRs that simultaneously optimize multiple benefit and risk outcomes, including both continuous and time-to-event endpoints. Building on and extending marginal struc- tural models, we propose to estimate “regimen-response curves”—functions that quantify the expected outcomes under different treatment strategies—and use them to guide optimal clinical decision-making. Unlike existing approaches, our methods will enable valid inference, accommodate multiple outcomes, and support flexible, in- terpretable treatment rules that evolve with disease progression. Aim 1 develops nonparametric methods for estimating regimen-response curves in settings with multiple continuous outcomes. Aim 2 extends these meth- ods to time-to-event and recurrent outcomes, incorporating dependent censoring and terminal events. Aim 3 introduces regularized, post-selection inference techniques to identify sparse and clinically interpretable treat- ment rules. Aim 4 focuses on software dissemination and application to a harmonized longitudinal dataset from five PD clinical trials. The ability to individualize PD motor treatment strategies in a way that balances benefit and risk does not currently exist and would offer substantial clinical value. The proposed work fills this critical need by delivering validated methods and open-source tools that support evidence-based, patient-centered care. Broader implications include generalizability to other chronic diseases where treatment optimization must account for mul- tiple, competing outcomes over time. The tools developed through this project will empower clinicians to improve long-term outcomes and quality of life for patients living with complex, evolving health conditions.