Acute graft-versus-host disease (aGVHD) remains a leading contributor to morbidity and transplant-related mortality after allogeneic hematopoietic stem cell transplantation (allo-HSCT). Existing prognostic tools largely rely on clinical grading and organ involvement; circulating biomarker panels (for example, MAGIC ST2 and REG3α) offer biologic insight but often require specialized assays. The lactate dehydrogenase-to-albumin ratio (LAR) integrates a marker of metabolic stress (LDH) with a proxy for nutritional and inflammatory reserve (albumin) and has shown prognostic value in multiple malignancies. This single-center retrospective cohort study assessed whether LAR measured at the time of aGVHD diagnosis is associated with all-cause mortality (ACM) and non-relapse mortality (NRM) after allo-HSCT.
This retrospective analysis included 109 consecutive patients who developed aGVHD after allo-HSCT at Henan Cancer Hospital between January 2019 and December 2024. Exclusion criteria were active uncontrolled infection at aGVHD onset, incomplete LDH or albumin within 48 hours of diagnosis, or loss to follow-up. The study received ethics committee approval and participants provided informed consent.
Patients were stratified by LAR tertiles: T1 4.37–8.84, T2 8.84–16.02, and T3 16.02–178.02. For regression, LAR was natural-log transformed (ln-LAR) to reduce skewness. Serum LDH and albumin were measured on a Beckman Coulter AU5800 platform and LAR calculated as LDH (U/L) divided by albumin (g/L).
Clinical covariates included demographics, HSCT indication, stem cell source, transplant type (matched sibling, unrelated, haploidentical), conditioning intensity (myeloablative vs reduced-intensity), MNC and CD34+ cell counts, interval from transplant to aGVHD, aGVHD grade (I–II vs III–IV), neutrophil engraftment time, CMV and EBV viremia, routine labs, and infectious complications. Supportive care and GVHD prophylaxis followed institutional protocols.
The primary endpoint was ACM; the secondary endpoint was NRM (death without preceding relapse). Follow-up ran from aGVHD diagnosis to death or last contact, censored December 31, 2024. Statistical approaches included Cox proportional hazards models with three hierarchical adjustments: Model 1 (age, sex); Model 2 (+transplantation covariates); Model 3 (+post-transplant clinical covariates). Restricted cubic spline (RCS) regression characterized dose-response shapes and two-piecewise Cox models assessed threshold effects. Subgroup, interaction, and sensitivity analyses (including exclusion of early deaths <3 months) tested robustness.
From 521 allo-HSCT recipients, 116 developed aGVHD; after exclusions 109 patients composed the analytical cohort. Mean age was 29.4 ± 15.3 years; 62.4% were male. Over a median follow-up of 30 months, there were 69 ACM events (63.3%) and 66 NRM events (60.6%).
Baseline characteristics varied across LAR tertiles. The proportion with grade III–IV aGVHD increased across tertiles (T1 22.2% → T2 55.6% → T3 81.1%; P < 0.001). Higher LAR tertiles had greater creatinine and blood urea nitrogen, higher aspartate aminotransferase, and more frequent CMV viremia. T1 patients had shorter median interval from transplantation to aGVHD diagnosis versus higher tertiles.
In Cox regression, continuous ln-LAR independently predicted outcomes after full adjustment (Model 3). For ACM, ln-LAR yielded HR 1.73 (95% CI 1.07–2.77, P = 0.024). For NRM, ln-LAR yielded HR 1.69 (95% CI 1.03–2.77, P = 0.038). Comparing tertiles, the highest LAR group (T3) experienced approximately threefold greater risk versus T1 (ACM HR 2.96, 95% CI 1.27–6.90, P = 0.012; NRM HR 3.25, 95% CI 1.36–7.74, P = 0.008).
RCS analyses demonstrated nonlinear inverted-L dose–response patterns with statistically significant nonlinearity (P = 0.006 for ACM, P = 0.009 for NRM). Two-piecewise models estimated threshold effects with breakpoints near LAR ≈53 for ACM and ≈19 for NRM. One outlier LAR value was excluded from segmented analyses to avoid undue influence; that patient remained in primary Cox models.
Subgroup and interaction testing did not identify significant effect modifiers (all P for interaction >0.05). Sensitivity analyses excluding early deaths (<3 months) produced qualitatively consistent associations and in some cases amplified effect sizes.
In this cohort of allo-HSCT recipients who developed aGVHD, higher LAR at diagnosis was independently associated with increased risk of both ACM and NRM after adjusting for demographic, transplantation, and post-transplant clinical covariates. The magnitude of association—approximately threefold risk in the highest tertile versus lowest—suggests the composite index may capture biological risk dimensions not fully represented by clinical grade alone. The observed nonlinear dose–response and estimated thresholds indicate the relationship between LAR and mortality may not be strictly linear.
The study leverages readily available laboratory measures (LDH and albumin) which could facilitate broader applicability compared with specialized biomarker assays. The LAR concept aligns with prior evidence linking elevated LDH to poor outcomes and hypoalbuminemia to increased NRM.
This analysis is retrospective and single-center, limiting causal inference and generalizability. Sample size was modest (n=109) and event rates were high; residual confounding remains possible despite multivariable adjustment. The study excluded patients with missing LDH or albumin within 48 hours and those with active infection at aGVHD onset, which may introduce selection bias. External validation in independent cohorts and prospective evaluation are required before clinical implementation.
Elevated LAR at the time of aGVHD diagnosis was independently associated with higher all-cause and non-relapse mortality in this single-center retrospective cohort of allo-HSCT recipients. These findings are hypothesis-generating and warrant external validation to determine the clinical utility of LAR for risk stratification and management decisions in aGVHD.