Serum Pyroptosis-Related Cytokines as Biomarkers for Diagnostic Assessment and Risk Stratification of Ocular Graft-versus-Host Disease: A Case-Control Study

Abstract

Background: Ocular graft-versus-host disease (oGVHD) often presents with subtle and nonspecific symptoms following allogeneic hematopoietic stem cell transplantation (allo-HSCT). Current diagnostic methods primarily depend on subjective clinical evaluations with limited sensitivity. This study aimed to identify serum pyroptosis-related cytokines associated with oGVHD and to develop a cytokine-based diagnostic model. Methods: In this prospective case–control study, 116 allo-HSCT recipients (61 with oGVHD and 55 without) and 47 healthy controls were enrolled. A sandwich antibody array was used to screen differentially expressed proteins in a pilot cohort (n = 4 per group), followed by pathway enrichment analysis. Ocular surface parameters and serum levels of NLRP3, TLR4, CCL2, IL-18, IL-6, and TNF-α were measured. Linear correlation, logistic regression, receiver operating characteristic (ROC) analyses, and internal bootstrap validation were performed to evaluate diagnostic performance. A cytokine-based risk score model was established. Results: Thirty-four upregulated proteins were enriched in immune and inflammatory pathways. Patients with oGVHD exhibited severe dry eye features and significantly higher serum levels of NLRP3, TLR4, CCL2, IL-18, and IL-6 (all p < 0.001), which inversely correlated with ocular surface parameters. A logistic model combining these five cytokines achieved excellent diagnostic accuracy (AUC = 0.960). Internal 10-fold cross-validation with 1,000 bootstrap iterations yielded a consistent mean AUC of 0.953, confirming model robustness. A simplified risk score stratified patients into low-, intermediate-, and high-risk categories with strong discriminatory power (p < 0.001). Conclusion: A serum panel of NLRP3, TLR4, CCL2, IL-18, and IL-6 demonstrates high diagnostic accuracy and stability for oGVHD. As the diagnostic cutoffs were derived from the same dataset, potential overfitting cannot be excluded, and independent validation in larger multicenter cohorts is warranted.