Chronic severe fatigue is a disabling comorbidity in autoimmune diseases such as systemic lupus erythematosus (SLE), primary Sjögren's syndrome (pSS), and multiple sclerosis (MS), sharing highly homologous underlying mechanisms characterized by neuro-immune microenvironment abnormalities and proinflammatory cytokine network cascade activation. Using MS—a classic central autoimmune disease—as a translational research model, this study aimed to quantify and prioritize the main effects of diverse exercise modalities and workload intensities on immune-mediated fatigue via network meta-analysis (NMA), thereby providing high-level evidence for targeted non-pharmacological interventions and personalized rehabilitation prescriptions in rheumatology.

Adhering strictly to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses for Network Meta-Analyses (PRISMA-NMA) statement, a comprehensive and highly sensitive literature search was executed across core medical databases to identify randomized controlled trials (RCTs) evaluating exercise interventions for immune-related fatigue. Following rigorous quality control, bias assessment, and data extraction protocols, a total of 37 high-quality RCTs were ultimately included in the quantitative synthesis. This extensive dataset encompassed 43 independent intervention arms and an aggregated cohort of 2,112 patients, ensuring high statistical power. A frequentist network meta-analysis model was employed to deconstruct the interventions into 11 distinct topological nodes based on training type and physiological load: multicomponent training (SCT, alongside its moderate and high-workload subgroups SCT_M, SCT_H), resistance training (RES, with its subgroups RES_M, RES_H), high-intensity interval training (HIT), continuous aerobic training (CAT), general comprehensive rehabilitation (REH), mind-body interventions (YOP), and passive control (PC). Global effect sizes for fatigue amelioration were evaluated using standardized mean differences (SMD) and corresponding 95% confidence intervals (CIs). The P-score metric was utilized to quantitatively rank the cumulative probability of each node emerging as the optimal intervention. Furthermore, network geometry was mapped, and potential publication bias or small-study effects were rigorously assessed utilizing comparison-adjusted funnel plots and Egger's regression test to validate the integrity of the findings.

Based on the robust baseline and endpoint clinical data derived from the 2,112 patients, an 11-node closed-loop network architecture was successfully constructed, with passive control (PC) serving as the central reference baseline. Pooled effect size analysis demonstrated that, compared to PC, moderate-intensity multicomponent training (SCT_M) exhibited the most significant and statistically robust main effect in alleviating fatigue (SMD = -0.50, 95% CI [-0.87, -0.14]). The overarching multicomponent training category (SCT) also displayed strong statistical significance (SMD = -0.56, 95% CI [-0.86, -0.27]), while the point estimates for CAT, REH, YOP, and baseline RES indicated steady clinical benefits, ranging from -0.30 to -0.35. Spatial ranking via the P-score cumulative probability model established SCT (0.878), SCT_M (0.786), and HIT (0.703) as the top-tier recommended interventions. Crucially, detailed subgroup analysis revealed a striking non-linear dose-response characteristic regarding exercise workload intensity. Moderate-workload SCT_M significantly outperformed its high-workload counterpart SCT_H (P-score: 0.786 vs 0.509). More importantly, the effect size for high-intensity, single-muscle-group resistance training (RES_H) failed to reach statistical significance, with its 95% CI widely crossing the line of no effect (SMD = -0.11). Consequently, its P-score plummeted to 0.222, approaching the baseline passive control group (0.074). This quantitative drop-off strongly indicates that mechanical or metabolic stress workloads exceeding a specific threshold may exacerbate autoimmune decompensation, subsequently worsening central fatigue rather than relieving it. Furthermore, the adjusted funnel plot displayed excellent visual symmetry, and Egger's test yielded a value of p = 0.9381, confirming that the current network model possesses exceptional global consistency and is devoid of significant confounding by publication bias.

This large-scale network meta-analysis provides compelling evidence that specific exercise interventions are highly effective for managing autoimmune-mediated chronic fatigue, yet they present significant clinical heterogeneity and strict non-linear dose-response relationships. Moderate-intensity multicomponent training (SCT_M) yields the optimal benefit-risk profile, likely by promoting anti-inflammatory myokine release without triggering systemic exhaustion. Conversely, excessively high-workload interventions—particularly high-intensity resistance training—may induce adverse immune stress, leading to diminished therapeutic efficacy and potential clinical deterioration. The conceptualization of an "immune-exercise workload adaptive window," distilled from this robust MS translational model, provides critical, cross-disciplinary evidence. It cautions against the uncalibrated application of intense physical therapy and establishes a foundational framework for developing precision rehabilitation guidelines tailored to the fragile physiological limits of patients suffering from chronic fatigue in SLE, pSS, and related rheumatologic conditions.