Directional Symptom Dependencies in Multiple Sclerosis and Parkinson’s Disease: A Comparative Bayesian Network Analysis

Abstract

Background: Multiple sclerosis (MS) and Parkinson’s disease (PD) are progressive neurological disorders characterized by complex interactions among motor symptoms, psychological disturbances, sleep problems, fatigue, and pain. Because conventional correlation-based approaches cannot determine effect direction, this study used Bayesian Network (BN) analysis to identify and compare directional pathways among disease characteristics, physical function, psychological measures, sleep quality, fatigue, pain, and physical activity in individuals with MS and PD.

Methods: Cross-sectional data from individuals with MS (n=104) and PD (n=54) were analyzed. Variables included demographics, disease duration, disability, physical function, cognition, anxiety, depression, fatigue, sleep quality, pain, and physical activity. Missing data (< 20%) were handled using multiple imputations with predictive mean matching. Bayesian networks were estimated using the Hill-Climbing algorithm with Gaussian BIC scoring continuous data. Bootstrap analysis (500 resamples) assessed edge reliability, and linear regression on standardized (z-scored) variables quantified relationship strengths.

Results: The MS network comprised 18 nodes and 27 directed edges; bootstrap stability analysis (500 resamples) indicated that 21 of 27 edges (77.8%) were stable at the ≥ 50% threshold. Anxiety emerged as a central hub, directly predicting physical activity (β = 0.37, p < 0.001), sleep quality (β = 0.51, p < 0.001), physical fatigue (β = 0.24, p = 0.016), and pain severity (β = 0.30, p < 0.001). Pain interference acted as a mediator linking sleep and depression to both cognitive (β = 0.23, p = 0.005) and physical fatigue (β = 0.40, p < 0.001). The PD network comprised 14 nodes and 15 directed edges; 14 of 15 edges (93.3%) were bootstrap-stable. Anxiety was again the central hub, strongly predicted by depression (β = 0.78, p < 0.001) and directly predicting pain interference (β = 0.59, p < 0.001) and balance (β = − 0.48, p < 0.001). Age directly predicted physical activity (β = − 0.50, p < 0.001) and balance (β = − 0.40, p < 0.001).

Conclusion: BN analysis revealed distinct disease-specific dependency structures. Anxiety emerged as a shared central hub in both MS and PD, although its downstream pathways differed. In MS, pain interference appeared to act as a key mediator, whereas the PD network was more parsimonious and dominated by age- and anxiety-related pathways. Because the data are cross-sectional, these findings should be interpreted as directional dependency structures that generate testable hypotheses for longitudinal and interventional work, rather than as confirmed causal pathways. They nonetheless point to anxiety as a plausible candidate target for future intervention studies for both conditions.