A multi-timepoint immune profiling model for differentiating gvhd- and infection-associated diarrhea following allo-HSCT

Abstract Introduction Allogeneic hematopoietic stem cell transplantation (allo-HSCT) is a curative therapy for various hematologic malignancies. However, post-transplant diarrhea is a frequent and clinically significant complication. Among its leading causes, graft-versus-host disease (GVHD) and infection often present with overlapping gastrointestinal symptoms, yet require different treatment approaches. Prompt and accurate differentiation between these etiologies is therefore critical for guiding therapeutic decision-making and improving patient outcomes. This study evaluated weekly monitoring of lymphocyte subsets and cytokines starting from white blood cell engraftment to develop a predictive model for differentiating infection- and GVHD-associated diarrhea after allo-HSCT. Methods We retrospectively analyzed 58 patients who underwent endoscopic evaluation with biopsy for post-HSCT diarrhea at Beijing Lu Daopei Hospital between April 2021 and December 2024. Based on histopathology, 34 were diagnosed with GVHD and 24 with infection. Lymphocyte subsets and cytokine levels were monitored weekly following white blood cell engraftment, including the week prior to diarrhea onset. Patients were follow-up continued until May 31, 2025. Results 1、Clinical characteristics did not differ significantly between the GVHD group (n=34) and infection group (n=24), including pre-transplant disease status, transplant type, conditioning regimen, ATG formulation, stem cell source, third-party donor use, graft cell dose, and cGVHD incidence(p all ＞0.05). Overall survival was comparable between groups (P=0.408). GVHD-associated diarrhea occurred earlier (median onset: 31 vs. 58 days) and lasted longer (86 vs. 28 days) than infection-related diarrhea. Frequency and volume of diarrhea were similar between groups (χ²=1.501, P=0.139; χ²=0.858, P=0.394). Both groups commonly presented with rash and hemorrhagic cystitis; however, hematochezia (26.5% vs. 8.3%) and hyperbilirubinemia (20.6% vs. 4.2%) were more frequent in the GVHD group. 2、At one week prior to diarrhea onset, the GVHD group showed significantly higher levels of Reg3α (2205.74 pg/mL [1174.11–3293.06] vs. 1333.70 [1058.83–1634.65], P=0.037), CD3⁺ T% (76.13% [47.45–84.27] vs. 50.33% [34.09–63.18], P=0.005), CD8⁺ T% (48.29±19.33 vs. 34.53±19.88, P=0.011), naive CD4⁺ T% (2.28% [0.75–5.73] vs. 1.06% [0.30–2.19], P=0.021), and CD4⁺ Th2% (40.18±21.11 vs. 27.77±20.61, P=0.030). Conversely, the GVHD group had lower NK cell proportion (17.80% [9.70–40.13] vs. 36.66% [15.46–46.78], P=0.042) and NK cell count (48.00/μL [22.25–73.25] vs. 80.00/μL [39.00–164.75], P=0.021). 3、Based on t-test or M-U test results, effect sizes for intergroup differences at one week prior to diarrhea onset were summarized using a forest plot. CD3⁺T%, CD8⁺ T%, absolute NK cell count (cells/μL), and the proportion of naïve CD4⁺T cells among total CD4⁺T cells demonstrated moderate effect sizes in distinguishing GVHD- from infection-associated diarrhea. Reg3αlevels and NK% showed small effect sizes. The direction of these effect sizes was consistent with the differences identified in comparative analyses.Using cytokine and immunophenotypic data from 1 week prior to diarrhea, a bivariate LASSO model incorporating NK cell count and CD4⁺CD28⁺% was constructed, yielding an AUC of 0.799. The optimal cutoff value was 0.408, with sensitivity 0.708 and specificity 0.882. 4、To improve diagnostic accuracy, a multivariable predictive score (PS) was developed using significant features from weeks 4, 5, and pre-diarrhea via LASSO selection. The final model incorporated 9 parameters:PS=2.6714+(CD8+T%)*(-0.0116)+NK*0.0049+(CD4+CD28+T%)*(-0.0217)+(CD4+Th2%)*(-0.0256)+(TNFRIweek4)*(-0.0005)+(NKweek4)*(0.0006)+(GM-CSFweek5)*(-1.6570)+(NK week5)*(0.0038)+(CD3+CD38+T% week5)*(0.0243).Multivariate logistic regression showed that each 1-unit increase in PS was associated with a 5.133-fold higher odds of infection-related diarrhea (OR=5.133; 95% CI: 2.47–14.71; P<0.001). The model demonstrated excellent discriminatory power (AUC=0.908), with an optimal cutoff of 0.669 (sensitivity: 0.875; specificity: 0.794). The multi-timepoint model outperformed single-timepoint predictions. Conclusion Immune profiling before diarrhea onset enables differentiation between GVHD and infection. A multi-timepoint model showed strong diagnostic accuracy (AUC=0.908) and may support clinical decision-making.