Uncovering Causal Links Between Electrophysiological Biomarkers and Parkinson's Disease Symptom Severity Using Machine Learning Approaches

Introduction: Understanding the relationship between electrophysiological biomarkers and clinical symptoms in Parkinson’s Disease (PD) is critical for advancing treatment strategies. Though correlations exist, causal connections remain unclear. This study aims to apply nonlinear machine learning to identify causal links between electrophysiological biomarkers and PD symptoms. We hypothesize that low beta cortical local field potential bursts hold the strongest association with symptom severity.

Methods: We analyzed electrophysiological data from 40 PD patients who underwent Deep Brain Stimulation (DBS) using a Random Forest regression model. The biomarkers included power, burst duration, rate, and amplitude from theta to gamma bands, with Unified Parkinson’s Disease Rating Scale (UPDRS) scores as the dependent variable. The model's performance was assessed using Root Mean Squared Error (RMSE) and Pearson correlation over 100 iterations with and without cross-validation. Feature importance was calculated to identify optimal predictors.

Results: A series of Random Forest models trained on 80% of the data and tested on 20% identified five key predictors of UPDRS scores: pre-central low beta burst rate, pre-central low beta burst duration, GPi alpha burst amplitude, central low beta burst rate, and GPi gamma burst rate. Notably, power was not among the top predictors. On average, models using 2.3 predictors achieved significantly lower RMSE (22.56) than random variables (26.17, p< 0.001). Models on seen data yielded a mean R² of 0.74. Cross-validation on unseen data resulted in a decrease in mean R² to 0.30, still outperforming the random shuffles (mean R² = 0.12, p< 0.001).

Conclusion : The Random Forest algorithm effectively predicted UPDRS scores using electrophysiological measures. A few neural biomarkers, especially beta burst, indicate a causal relationship with PD symptomatology. These findings highlight the potential of electrophysiological biomarkers in predicting clinical outcomes in PD, aiding in improved patient management and personalized treatment strategies.