Objective: To develop machine learning models that predict advanced Parkinson’s disease (APD) by integrating quantitative clinical data from the Parkinson’s Progression Markers Initiative (PPMI) with expert-based criteria from the Advanced Parkinson Disease questionnaire (CDEPA).

Background: APD represents a critical stage of Parkinson’s disease (PD), marked by severe motor and non-motor complications that detrimentally affect patient quality of life. Despite its clinical significance, a universally accepted quantitative definition of APD is lacking. Adapting qualitative screening tools like the CDEPA to structured datasets may improve early identification and facilitate timely, personalized interventions.

Method: Data from 1,302 PD patients in the PPMI study were curated by mapping 2,857 available variables to 216 features relevant to CDEPA criteria, as determined by a panel of neurology and neuropsychology experts. Multiple machine learning techniques—including logistic regression, support vector machines, random forests, and gradient boosting (XGBoost)—were employed to classify APD status using both binary (APD vs. non-APD) and multiclass (definite, probable, possible, and non-APD) approaches. Model performance was evaluated at 5-year and 8-year follow-ups via bootstrapping, area under the receiver operating characteristic curve (AUC-ROC), and feature importance analyses.

Results: The binary predictive models achieved high accuracy, with top-performing approaches recording AUC-ROC values above 0.85. Key predictors included motor and non-motor assessment scores, disease duration, and functional impairment measures. Binary classification models outperform multiclass classification models when reliably identifiyng patients at risk of progressing to APD, demonstrating robust generalizability across follow-up periods.

Conclusion: Integrating expert-driven qualitative criteria with data-driven machine learning methods offers a promising framework for early APD prediction. This approach may enhance clinical decision-making by enabling timely interventions and the personalization of treatment strategies for patients with PD.

« Back to 2025 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/ai-driven-prediction-models-for-advanced-parkinsons-disease-a-data-driven-approach-using-ppmi-and-cdepa-criteria/