A data-driven framework for long-term risk stratification of advanced Parkinson's disease using PPMI

Area of Science:

• Neurology
• Biostatistics
• Clinical Informatics

Background:

• Current staging for advanced Parkinson disease lacks standardization, hindering longitudinal studies and treatment comparisons.
• Qualitative assessment tools impede objective, reproducible tracking of disease progression.

Purpose of the Study:

• To develop and validate a reproducible operationalization for staging advanced Parkinson disease.
• To generate longitudinal labels for tracking disease certainty over time.
• To forecast the onset of advanced Parkinson disease using baseline data.

Main Methods:

• Translated the 13-item Diagnostic Criteria for Advanced Parkinson Disease questionnaire into structured variables.
• Developed a pipeline for generating longitudinal labels of advanced disease certainty.
• Applied the pipeline to the Parkinson's Progression Markers Initiative cohort (n=1,302) for trajectory analysis.
• Utilized baseline clinical and genetic data to forecast advanced disease at 7-11 years.

Main Results:

• The developed pipeline successfully generated longitudinal labels and characterized disease trajectories.
• A year-9 forecasting model achieved an AUC of 0.89 (95% CI 0.81-0.97) for predicting advanced disease.
• Model performance attenuated in an independent real-world cohort (AUC 0.55-0.61), indicating dataset shift.

Conclusions:

• The operationalized staging tool provides a reproducible method for longitudinal Parkinson disease assessment.
• Forecasting models show potential for predicting future disease advancement, aiding proactive therapeutic strategies.
• Further validation in diverse cohorts is necessary to address dataset shift and improve generalizability.