Linear predictive coding electroencephalography algorithms predict Parkinson's disease mortality using out-of-sample tests.

Background: Parkinson's disease (PD) increases mortality is difficult to predict because of its heterogeneity and the availability of very few reliable which prognostic markers. Objectives: We used electroencephalography (EEG) and the Linear Predictive Coding EEG Algorithm for PD (LEAPD) for binary classification of 3-year mortality status and correlation between LEAPD indices and time to death. Methods: 2-minutes resting-state EEG from 94 PD patients (59 channels, 22 deceased within 3 years of recording) was used for binary classification of 3-year mortality status. Single-channel classification using a balanced dataset of 44 was performed using leave-one-out cross-validation (LOOCV). Robustness was evaluated by truncating the recordings. LOOCV Spearman's correlation coefficient (ρ) was obtained between LEAPD indices and time to death. Optimum hyperparameters obtained from a balanced training dataset of 30 were tested on the remaining 64 patients by 10,000 randomized comparisons of 7 vs 7, using 5 channel combinations Hyperparameters for the best ρ, using the same training dataset were for the out-of-sample correlation for the remaining 7 deceased. Results: In LOOCV analysis several channels yielded 100% accuracy with robust performance from five. The correlations ranged between ρ = -0.59 to -0.86; were significant after adjusting for age, cognitive and motor impairment. Out-of-sample testing using the best-performing 5-channel combination yielded a mean accuracy of 83%. Out-of-sample Spearman's ρ was -0.82. Conclusion: LEAPD provides a robust approach for binary classification of mortality in PD from resting-state EEG. LEAPD indices correlate with survival duration, independent of clinical predictors, suggesting potential utility as a continuous neurophysiological biomarker.