Cognitive Performance and Learning Parameters Predict Response to Working Memory Training in Parkinson's Disease.

Area of Science:

• Neuroscience
• Cognitive Science
• Artificial Intelligence

Background:

• Working memory (WM) training (WMT) is used for cognitive decline in Parkinson's disease (PD).
• Patient response to WMT varies, indicating a need for personalized prediction models.

Purpose of the Study:

• To develop a multivariate model predicting post-intervention verbal WM in PD patients using machine learning.
• To assess the predictive value of novel WMT learning parameters against traditional demographic, clinical, and cognitive data.

Main Methods:

• 37 PD patients underwent a 5-week WMT.
• Four random forest regression models were built using cognitive variables, learning parameters, combined variables, and all available data (demographic, clinical, cognitive, learning).
• Models predicted immediate and 3-month follow-up WM.

Main Results:

• The comprehensive model ('all' model) demonstrated the lowest root mean square error (RMSE) for predicting WM at both immediate (0.184) and 3-month follow-up (0.216).
• Baseline cognitive parameters were key predictors within the comprehensive model.
• A model combining cognitive and learning parameters significantly outperformed one using only cognitive variables.

Conclusions:

• Demographic, clinical, and cognitive variables robustly predict WMT response in PD.
• Incorporating training-specific learning parameters enhances prediction model accuracy.
• Improved prediction may lead to augmented training benefits from cognitive interventions in PD.