This study included 106 adult patients who underwent allo-HSCT at the Third Xiangya Hospital of Central South University between January 2019 and December 2023 and had complete longitudinal lipid profiles. Patients with underlying diseases of dyslipidemia or incomplete lipid data were excluded based on screening of medical records. This study was approved by the Ethics Committee of the Third Xiangya Hospital (Ethic committee, Third Xiangya Hospital, Central South University, No.23067). All participants provided written informed consent in accordance with the Declaration of Helsinki.

The conditioning regimens were defined according to the consensus from The Chinese Society of Hematology (CSH) (7). Most patients with hematological malignancies received a standard myeloablative regimen of modified busulfan and cyclophosphamide. Elderly patients (age ≥ 55 years old) and patients with a high risk of comorbidity underwent a reduced-intensity regimen (RIC) in which cyclophosphamide was substituted (or partially substituted) with fludarabine. Patients with refractory leukemia underwent an intensified conditioning regimen. Additionally, anti-thymocyte globulin (ATG) was added to the conditioning regimen for mismatched unrelated donor transplantations or transplantations for nonmalignant diseases. All patients received graft-versus-host disease (GVHD) prophylaxis consisting of cyclosporine A (CsA), methotrexate (MTX), and mycophenolate mofetil (MMF).

Diagnoses and severity grading of acute GVHD (aGVHD) and chronic GVHD (cGVHD) followed the criteria outlined in the Chinese Expert Consensus on Acute GVHD (8) and Chinese Expert Consensus on Chronic GVHD (9), respectively. aGVHD comprises two distinct subtypes: (i) Classic aGVHD—manifesting inflammatory injury to the skin, gastrointestinal tract, and/or liver within 100 days post-transplant (+100 d); and (ii) Late-onset aGVHD—exhibiting identical diagnostic features but emerging >100 days post-transplant. Chronic GVHD (cGVHD) encompasses the broad spectrum of disease occurring beyond +100 d, not exclusively defined by temporal boundaries but rather by clinicopathological manifestations.

Neutrophil recovery was defined as the first of three consecutive days with ANC ≥0.5×10⁹/L, while platelet recovery required seven consecutive transfusion-free days with platelet counts ≥20×10⁹/L (10).

Lipid levels during HSCT, including HDL-C, LDL-C, TC and TG levels, were collected from electronic medical charts. The lipid levels were classified according to the normal limits recommended by the National Cholesterol Education Program (NCEP) guidelines (11). Dyslipidemia was defined as levels above or below the normal range for one or more lipids. Lipid values were defined as follows: low HDL‐C, <1 mmol/L in men and <1.3 mmol/L in women; high TG, >2.25 mmol/L; high TC, > 5 mmol/L, and high LDL-C, > 3.2 mmol/L.

SPSS 26.0 and GraphPad Prism 8.0 were used for statistical analysis and drawing. Continuous variables are expressed as the median (range). The Mann–Whitney U test was used to evaluate differences in general characteristics and laboratory results between groups. Categorical variables are expressed as counts and proportions and were evaluated using the chi-squared test or Fisher’s exact test. The overall survival (OS) and GVHD-free/relapse-free survival (GRFS) were estimated by the Kaplan–Meier method and compared by the log-rank test.

Data analysis was performed in R v4.4.3 using generalized linear mixed models (GLMMs) to evaluate dynamic associations between high-density lipoprotein (HDL) levels and graft-versus-host disease (GVHD) probability over post-transplant time. After preprocessing the longitudinal dataset, converting patient IDs to factors, and applying square-root transformation to time variables, four GLMM variants were compared: 1) Base model (HDL + Time + random intercept); 2) Interaction model (HDL×Time); 3) Nonlinear model (HDL²); 4) Time-stratified model (early/late phase). Model selection via likelihood ratio tests and AIC/BIC minimization identified the interaction model as optimal, enabling quantification of time-dependent HDL effects. Final visualizations included trajectory plots at 8 timepoints with 95% CIs and 3D interaction surfaces, with residual diagnostics confirming model adequacy.

Cumulative incidence functions (CIFs) for GVHD were estimated using Fine-Gray proportional subdistribution hazards models via the R package cmprsk with GVHD and non-GVHD mortality as competing events. All patients were followed from transplantation until GVHD onset, competing event occurrence, or last contact. Event status was coded as: 0=censored, 1= GVHD, 2= GVHD and non-GVHD mortality. Subdistribution hazard ratios (sHR) with 95% confidence intervals quantified HDL effects on GVHD risk in the presence of competing events.

Univariate and multivariate logistic regression models were used to evaluate the relationships between lipid levels and GVHD, and unadjusted and adjusted odds ratios (ORs) and 95% confidence intervals (Cis) were calculated. In the multivariate adjusted models, sex, age and diagnosis (including acute lymphoid leukemia, acute myeloid leukemia (AML), aplastic anemia (AA), myelodysplastic syndrome (MDS), and other disease) were included. A P value < 0.05 (two-tailed) was considered statistically significant.

A total of 106 patients were enrolled in our study. The median age of the patients in the retrospective cohort was 39 years (range, 18–58 years). Of the patients, 64 (60.38%) had AML, 23 (21.7%) had acute lymphoid leukemia (ALL), 11 (10.37%) had MDS, 2 (1.89%) had AA, and 6 (5.66%) had other diseases. The median number of MNCs infused for the stem cell transplantation was 9.47×10⁸/kg (range, 0.87-19.9×10⁸/kg), and the median number of CD34+ cells infused was 4.37×10⁶/kg (range, 0.62-16.8×10⁶/kg). The median times to neutrophil and platelet engraftment were 11 (range, 8-19) and 13 (range, 7-46) days, respectively. Most patients had a matched related donor (n=86, 81.13%). The demographics are summarized in Table 1 .

To accurately characterize the temporal evolution of serum lipid profiles relative to transplantation, we included 106 patients with complete lipid measurements at predefined time points: pre-transplant (baseline), day 7, day 14, and months 1, 3, 6, 12, and 24 post-HSCT. Analysis revealed significant dynamic changes in lipid levels over time. While a subset of patients exhibited dyslipidemia pre-transplant, a pronounced alteration in lipid profiles became evident post-HSCT. Specifically, many survivors developed significantly lower serum HDL-C and/or higher TG, TC, and LDL-C levels after transplantation compared to baseline ( Figure 1 , Supplementary Table 1 ). The most dramatic shift occurred within the first month post-HSCT, characterized by a steep decline of HDL-C. Notably, HDL-C reached its nadir by day 14, followed by a slow recovery phase. Although a gradual recovery trend in lipid levels emerged after month 1, persistent alterations compared to baseline remained common during long-term follow-up (up to 24 months), particularly for HDL-C. Consequently, the predominant dyslipidemia observed within 3 month post-HSCT featured significantly decreased HDL-C levels and reduced HDL-C/TC and HDL-C/TG ratios. The overall temporal trends and individual patient variations are detailed in Figure 1 and Supplementary Table 1 .

We examined the correlation between lipid levels and the clinical outcomes of HSCT. The median follow‐up of survivors was 34.07 months (range: 0.70–77.47 months). A total of 38 patients (35.85%) had acute GVHD. The incidence of grade I-II aGVHD and grade III-IV aGVHD was 23.58% and 12.26% ( Table 2 ), respectively. The incidence of cGVHD was 10.38%. HDL-C and the HDL-C/TC ratio decreased significantly in the GVHD group, especially at 14 days and/or 1 month after transplantation ( Figure 2 ). The detailed characteristics of lipid levels are shown in Supplementary Table 1 .

Analysis of GVHD outcomes across severity grades (acute grades I-IV and chronic GVHD) revealed that patients with grade III-IV aGVHD exhibited significantly lower serum HDL-C levels at day 14 post-transplantation compared to non-GVHD controls (P=0.045). Conversely, those with grade I-II aGVHD demonstrated a markedly reduced HDL-C/TC ratio at 1 month post-HSCT relative to the non-GVHD group (P=0.034). No significant differences in lipid profiles were observed between patients with cGVHD and non-GVHD counterparts at any measured timepoint ( Table 2 ). Cumulative incidence analysis of GVHD outcomes revealed significantly higher cumulative incidence in patients with low HDL-C levels (≤0.78 mmol/L; P=0.018) and a low HDL-C/TC ratio (≤0.17; P=0.013), compared to those with high levels (HDL-C levels measured at day 14 post-HSCT). However, no association was observed between HDL-C levels or the HDL-C/TC ratio and overall mortality (including both GVHD-related and non-GVHD-related mortality) ( Figure 3 ).

Furthermore, early post-HSCT HDL-C levels (measured at day 7 and day 14 post HSCT) exhibited a significant inverse correlation with time to neutrophil engraftment, indicating that lower HDL-C correlated with delayed myeloid recovery ( Figure 4 ).

Among 106 HSCT recipients, 23 deaths (21.70%) occurred with GVHD complications being the predominant cause (39.13%), followed by relapse (34.78%) and infection (17.39%). The 1-year and 2-year OS rates were 92.5% and 82.9%, respectively, while the 1-year and 2-year GRFS rates were 52.8% and 49.0%. No significant differences in lipid parameters were observed between survivors and non-survivors (P>0.05), and neither HDL-C levels nor HDL-C/TC ratio showed association with OS ( Figure 5 ). Notably, lower HDL-C levels was significantly associated with worse GRFS outcomes ( Figure 5 ). This dichotomy suggests HDL-C’s protective effect manifests primarily through GVHD mitigation rather than overall mortality reduction in this cohort.

To better elucidate the relationship between HDL-C levels and GVHD, we employed generalized linear mixed models (GLMM) to investigate the dynamic effects of HDL-C levels and post-transplantation time on the probability of GVHD occurrence. The GLMM analysis demonstrated a significant interaction between HDL-C levels and GVHD, indicating that the protective effect of HDL-C varies over the post-HSCT course ( Figure 6 ). The 3D surface plot ( Supplementary Figure 1 ) quantified HDL-time effects on GVHD probability. HDL-C <1.0 mmol/L within 6 months post-HSCT caused steep risk elevation. At HDL-C=0.6 mmol/L (Month 1), probability peaked at 12.0%. Levels ≥1.8 mmol/L maintained risk < 4.0%.

Next, the utility of the HDL-C level and HDL-C/TC ratio (day 14 post-transplantation) for predicting GVHD were estimated. The AUC for the HDL-C level was 0.640 (P =0.014), and the AUC for the HDL-C/TC ratio was 0.632 (P =0.02) ( Figure 7 ). Univariate analysis found that HDL-C ≤ 0.84 mmol/L was risk factors for GVHD. Further multivariate analysis showed that HDL-C ≤ 0.84 mmol/L was independent risk factors for GVHD [OR: 3.15, 95% CI, 1.20-8.26, P=0.020] ( Table 3 ). Therefore, lower HDL-C levels could be simple biochemical markers of GVHD in HSCT. These findings suggest that HDL monitoring may be particularly valuable for GVHD risk stratification during the early post-transplantation period when its protective effects are most pronounced. The time-dependent nature of this relationship highlights the importance of considering both HDL levels and temporal factors in clinical assessment.