Key points are not available for this paper at this time.
Background: Anemia affects up to 85% of hemodialysis patients and is associated with increased morbidity and mortality. Immune dysfunction in end-stage renal disease (ESRD) may contribute to anemia through inflammatory pathways. This exploratory study used machine learning to investigate associations between immune markers, particularly total immunoglobulin G (IgG) concentration and varicella zoster virus (VZV) serostatus, and anemia in dialysis patients. Methods: This cross-sectional study enrolled 351 participants (179 dialysis patients, 172 healthy controls) between November 2023 and February 2024. Demographic and clinical characteristics were compared using conventional statistics. Within the dialysis cohort, seven machine learning classification models were trained to identify correlates of anemia (hemoglobin <12.0 g/dL for women, <13.0 g/dL for men). Feature importance analysis quantified the relative contribution of clinical and serological variables. Nested cross-validation with 1000 bootstrap iterations provided bias-corrected performance estimates with 95% confidence intervals. Results: Dialysis patients were older (median 47 vs. 34 years, p<0.01) and had higher prevalence of diabetes (34.6% vs. 4.7%, p<0.01), hypertension (82.1% vs. 11.1%, p<0.01), and VZV IgG seropositivity (92.2% vs. 80.2%, p<0.01). In machine learning analysis restricted to dialysis patients, random forest modeling achieved the best performance (cross-validated accuracy 0.91, 95% CI: 0.87-0.95). Feature importance analysis identified random blood glucose (24.6%), total IgG concentration (22.0%), and age (19.9%) as the strongest correlates of anemia. VZV serostatus alone showed minimal predictive value (importance 2.1%) when total IgG was included in the model. Conclusion: These findings suggest that total IgG concentration, a marker of broader immune activation, is strongly associated with anemia in dialysis patients, whereas VZV-specific serostatus shows minimal independent association. The results should be interpreted as hypothesis-generating, and prospective studies are needed to validate these associations and explore underlying mechanisms.
Sandokji et al. (Tue,) studied this question.