Predicting the multi-domain progression of Parkinson's disease: a Bayesian multivariate generalized linear...

Area of Science:

• Neurology
• Biostatistics
• Data Science

Background:

• Parkinson's disease (PD) progression is complex, involving multiple domains and longitudinal changes.
• Current statistical models struggle to accurately predict individual patient trajectories.
• Accurate prediction is crucial for effective PD management and treatment.

Purpose of the Study:

• To develop and validate a novel statistical model for predicting clinical progression in Parkinson's disease.
• To leverage multi-domain longitudinal data for enhanced predictive accuracy.
• To compare the proposed model's performance against traditional univariate analyses.

Main Methods:

• Utilized a Bayesian multivariate generalized linear mixed-effect model (GLMM) to analyze longitudinal outcomes (motor, non-motor, postural instability scores from MDS-UPDRS) at baseline, 18, and 36 months.
• Incorporated demographic and clinical data as covariates.
• Performed dynamic predictions and evaluated accuracy using Root Mean Square Error (RMSE), Absolute Bias (AB), and Area Under the Receiver Operating Characteristic (ROC) curve.

Main Results:

• The multivariate GLMM identified key clinical predictors for motor, non-motor, and postural stability scores.
• The model demonstrated improved prediction accuracy, especially for non-motor symptoms (RMSE: 2.89, AB: 2.20) compared to univariate analysis (RMSE: 3.04, AB: 2.35).
• ~80% of observed values for individual longitudinal trajectories fell within the 95% credible intervals in testing data.

Conclusions:

• Multivariate generalized linear mixed-effect models offer a promising approach for predicting individual clinical progression in Parkinson's disease.
• This methodology enhances the ability to forecast patient outcomes using comprehensive longitudinal and multi-domain data.
• The findings support the adoption of advanced statistical modeling for personalized PD care.