This study was a multicenter retrospective cohort including patients diagnosed with HTG-AP who were hospitalized between January 2020 and March 2025 at Zhongshan Hospital (Xiamen), Fudan University, the 900th Hospital of Joint Logistics Support Force of Chinese People’s Liberation Army (PLA), and Zhangzhou Municipal Hospital of Fujian Province.

Inclusion criteria: eligible patients had confirmed HTG-AP during the index hospitalization and were either first-episode cases or had an adequately documented history of prior pancreatitis episodes.

Exclusion criteria: (1) age <18 years; (2) confirmed chronic pancreatitis; (3) history of pancreatic surgery; (4) severe cardiac, hepatic, or renal insufficiency; (5) malignancy; (6) pregnancy or lactation; and (7) missing essential clinical or follow-up data. The study flowchart is shown in Figure 1.

This study was approved by the Ethics Committees of Zhongshan Hospital (Xiamen) Fudan University (B2025-026), 900th Hospital of PLA Joint Logistic Support Force (2025–068), and Zhangzhou Municipal Hospital (2025KYZ249).

Diagnosis of acute pancreatitis was established according to the Guidelines for diagnosis and treatment of acute pancreatitis in China (2021, 15), requiring at least two of the following criteria: (1) characteristic symptoms of acute pancreatitis such as acute upper abdominal pain; (2) serum amylase or lipase levels elevated to more than three times the upper limit of normal; and (3) imaging findings consistent with acute pancreatitis. Diagnosis of HTG-AP was made if serum triglyceride levels were ≥11.3 mmol/L at presentation, or between 5.65 and 11.3 mmol/L after excluding other possible causes. The severity of disease was graded based on the 2012 revised Atlanta classification and definitions: mild acute pancreatitis—no organ failure or local/systemic complications; moderately severe acute pancreatitis—local/systemic complications or transient organ failure (≤48 h); and severe acute pancreatitis—persistent organ failure lasting >48 h (16).

Recurrence was defined as a new episode of acute pancreatitis meeting standard diagnostic criteria and occurring ≥3 months after the index attack, following complete clinical resolution with a symptom-free interval, to minimize misclassification of early re-admissions attributable to persistent inflammation or local complications of the index episode (17). All recurrent episodes were identified according to clinical diagnoses documented in the medical records and verified by corresponding imaging examinations. Metabolic comorbidities included diabetes, hypertension, fatty liver, and gout. Obesity was defined by a body mass index (BMI) of ≥28.0 kg/m². Lipid-lowering therapy was defined as continuous use of statins, fibrates, or niacin for at least 1 month before admission.

Data were retrospectively extracted from the electronic medical record systems of all centers, covering demographic information at admission (age, gender, BMI, marital status, education level), lifestyle habits (smoking and drinking history), comorbidities (diabetes, hypertension, gout, and fatty liver), and history of lipid-lowering drug use. Disease-related data were also recorded, including previous episodes, severity classification of the current episode, and laboratory findings within 24 h of admission, such as routine complete blood count, hepatic and renal function tests, electrolytes, total cholesterol, triglycerides, fasting blood glucose (FBG), LDL-C, HDL-C, C-reactive protein (CRP), and procalcitonin (PCT). To ensure data accuracy, all information was independently extracted and cross-checked by two trained investigators using a standardized data collection form. Any discrepancies were adjudicated by a third researcher.

Admission lipid profiles (triglycerides, total cholesterol, LDL-C, and HDL-C) were extracted from the laboratory information systems of each participating center; only laboratory-validated results released in the system were used for analysis. Blood sampling was performed within 24 h of admission as part of routine clinical care. Lipid sampling was obtained when patients met each center’s pre-analytical requirements for lipid testing; fasting status was assessed by diet history at presentation. Serum lipids were measured in each center’s central laboratory using enzymatic assays on automated chemistry analyzers, and HDL-C was determined using a direct homogeneous method according to manufacturers’ instructions and local standard operating procedures.

Inter-center comparability was supported by national clinical laboratory quality management requirements, including routine internal quality control and participation in applicable external quality assurance programs. Lipid values were harmonized for analysis using uniform units. Specimens with severe lipemia or hemolysis were flagged by the laboratory; results deemed analytically unreliable were not released as valid results and were therefore excluded from analysis. When repeat sampling and re-testing were performed as part of routine clinical care after specimen rejection, the subsequent validated result was used when available.

Follow-up data were obtained through outpatient visits, review of inpatient medical records, and structured telephone interviews. Trained nurses conducted telephone follow-ups at 1, 3, and 6 months after discharge; recurrence at 12 months and longer-term outcomes were ascertained via ongoing outpatient follow-up and supplemented by medical record review. For suspected recurrences managed outside participating centers, available medical documentation (e.g., discharge summaries, laboratory results, and imaging reports) was retrieved and each episode was adjudicated against standard diagnostic criteria before confirmation. The primary endpoint was overall post-discharge recurrence of HTG-AP, with secondary endpoints of recurrence within 6 and 12 months. Participants without recurrence were censored at the last successful follow-up or the study end date (June 30, 2025), whichever occurred first; patients with missing follow-up information were excluded (n = 9; Figure 1).

Continuous variables were summarized as median (interquartile range, IQR) and compared using the Mann–Whitney U test; categorical variables were presented as n (%) and compared using the chi-square or Fisher’s exact test, as appropriate. Recurrence risk was analyzed using Cox proportional hazards models and reported as hazard ratios (HRs) with 95% confidence intervals (CIs). HDL-C was modelled both continuously and by tertiles. Four hierarchical models were prespecified: Model 1, unadjusted; Model 2, adjusted for gender, age, and BMI; Model 3, additionally adjusted for smoking, drinking, metabolic comorbidities, and first episode status; and Model 4, further adjusted for lipid-lowering therapy, admission triglycerides, admission CRP, and acute pancreatitis severity. Kaplan–Meier curves were compared using the log-rank test.

To explore time-dependent effects, Landmark analyses were performed for recurrence risk at 6 and 12 months, with separate Cox models constructed for each time window. An interaction term between HDL-C and log(time) was incorporated to assess the proportional hazards assumption, with adjustment for Model 3 covariates and lipid-lowering therapy use. The restricted mean recurrence-free survival time (RMST) within 12 months was used to quantify absolute risk differences. The dose–response relationship was evaluated using a restricted cubic spline (RCS) model adjusted according to Model 4. Subgroup analyses were stratified by gender, obesity, first onset, and metabolic comorbidities to test for interactions, and joint effects were assessed on an additive scale. Multiple recurrence events were analyzed using negative binomial regression to estimate incidence rate ratios (IRRs).

A Bayesian Cox sensitivity analysis was performed by jointly including admission triglycerides, total cholesterol, LDL-C, and HDL-C (each per 1-SD increase) with the fully adjusted covariates and applying weakly informative regularizing priors. Results are reported as posterior median HRs with 95% credible intervals (CrIs) and posterior probabilities, and convergence was assessed using Rhat and effective sample size.

To reduce confounding, inverse probability of treatment weighting (IPTW) based on propensity scores was applied, and the main analyses were repeated in the weighted cohort. HDL-C was dichotomised using the T3 cut-point (high HDL-C: ≥T3; low HDL-C: <T3). Propensity scores (PS) for low HDL-C were estimated with multivariable logistic regression including prespecified baseline covariates: gender, age, BMI, education level, marital status, smoking history, drinking history, diabetes, hypertension, gout, hepatic steatosis, lipid-lowering agent use, and first episode status. IPTW weights were defined as 1/PS for the low HDL-C group and 1/(1 − PS) for the high HDL-C group, with extreme weights truncated at the 1st and 99th percentiles to improve stability. Covariate balance was assessed using standardized mean differences (SMDs). Weighted Cox models were fitted using robust sandwich variance estimators. All statistical analyses and visualizations were performed using R version 4.4.3 and Python version 3.12.3, with a two-sided p value <0.05 considered statistically significant.

In total, 440 patients diagnosed with HTG-AP were enrolled, with a median age of 39.00 (33.00–46.25) years. Among them, 80.91% (356/440) were male, with the median follow-up period was 14.0 (8.0–26.0) months. Participants were categorized into three groups based on HDL-C levels at admission, and comparisons were made across groups. The results showed that patients in the T1 group had the highest recurrence rate (46.90%), tended to be younger, and had higher proportions of prior episodes and severe cases. Compared with the T2 and T3 groups, patients in the T1 group had significantly higher levels of FBG, triglycerides, and total cholesterol, indicating a heavier metabolic burden and less frequent use of lipid-lowering therapy (Table 1).

Multivariate Cox regression demonstrated a strong negative correlation between HDL-C levels at admission and the overall recurrence risk of HTG-AP. In the fully adjusted model, each 1 mmol/L increase in HDL-C was associated with a 77% reduction in recurrence risk (HR = 0.23, 95% CI: 0.11–0.48, p < 0.001). Compared with the T1 group, recurrence risk decreased by 58% in the T2 group (HR = 0.42, 95% CI: 0.27–0.64) and by 67% in the T3 group (HR = 0.33, 95% CI: 0.19–0.55), with similar protective associations observed for recurrence within 6 and 12 months post-discharge (Table 2; Supplementary Table 1). Kaplan–Meier curves further demonstrated that patients in the T1 group consistently exhibited the highest recurrence risk, which increased rapidly during the early post-discharge period (Log-rank p < 0.001) (Figure 2). Furthermore, negative binomial regression showed that the recurrence risk in the T1 group was 79% higher than that in the T3 group (IRR = 1.79, 95% CI: 1.06–3.02) (Supplementary Table 2), further supporting the inverse association between HDL-C and recurrence risk.

In a sensitivity analysis to assess the robustness of our findings and to address potential collinearity among admission lipid measures, we fitted a Bayesian Cox proportional hazards model simultaneously including admission triglycerides, total cholesterol, LDL-C, and HDL-C (each standardized per 1-SD increase). The inverse association between HDL-C and recurrence risk remained robust (median HR 0.613, 95% credible interval 0.485–0.771; posterior probability that HR < 1 was 1.00) (Supplementary Table 3; Supplementary Figure 1).

On this basis, a RCS analysis was performed to evaluate the dose–response relationship between HDL-C levels and recurrence risk (Figure 3). After adjusting for covariates in Model 4, low HDL-C levels were significantly associated with an increased recurrence risk, which progressively decreased with rising HDL-C and plateaued at higher concentrations.

To delineate the distinct effects of HDL-C on early versus late recurrence, a landmark analysis was conducted (Figure 4; Supplementary Table 4). The Kaplan–Meier curves indicated that the predictive effect of HDL-C was primarily concentrated within the first 12 months after discharge. Patients with low HDL-C levels had a higher recurrence rate within 0–6 months, with rates of 27.4% in the T1 group versus 7.8 and 5.1% in the T2 and T3 groups, respectively (log-rank p < 0.001). The difference remained significant at 12 months post-discharge (log-rank p = 0.042), but after 12 months, the survival curves for all groups converged progressively (log-rank p = 0.568).

The Cox regression results from the landmark analysis showed a consistent trend (Table 3). Within 6 months after discharge, each 1 mmol/L increase in HDL-C was associated with a 95.5% reduction in recurrence risk (HR = 0.045, 95% CI: 0.012–0.171, p < 0.001). During the 6–12 month period, this protective effect remained significant (HR = 0.176, 95% CI: 0.054–0.568, p = 0.004). However, beyond 12 months, the association between HDL-C and recurrence risk was no longer significant (p = 0.197). According to the RMST analysis (Supplementary Table 4), within 12 months, the mean recurrence-free survival time in T1 and T2 groups was 2.2 months (p < 0.001) and 0.5 months (p = 0.026) shorter than that of the T3 group, respectively. The time-dependent analysis also demonstrated that the protective effect of HDL-C was mainly concentrated in the early post-discharge period and gradually diminished over time.

Subgroup analyses (Figure 5) showed that lower admission HDL-C levels were consistently associated with a higher recurrence risk across the examined strata, although the magnitude of the association varied across subgroups. However, multiplicative interaction tests provided no evidence of effect modification by gender, obesity, diabetes, hypertension, gout, fatty liver, lipid-lowering drug use, or first-onset status (all p for interaction > 0.05). In the additive interaction analysis (Supplementary Table 5), the joint exposure of low HDL-C and metabolic comorbidities was associated with a higher recurrence risk compared with the reference group, but the relative excess risk due to interaction (RERI) was not statistically significant, indicating no clear evidence of additive interaction.

To verify the robustness of the results, we repeated the primary analyses using IPTW. HDL-C levels at admission were divided into high and low groups using the T3 tertile as the cutoff. Before weighting, differences existed between the two groups in age, BMI, marital status, lipid-lowering agents use, and first-onset proportion; after IPTW adjustment, baseline covariates were well balanced and the group distributions were largely comparable (Supplementary Figure 2; Table 4).

After weighting, Cox regression analyses revealed that patients with low HDL-C consistently exhibited a higher recurrence risk than those with high HDL-C at all follow-up time windows (Figure 6; Table 5). Within the first 0–6 months after discharge, the low HDL-C group had a 76% higher recurrence risk (aHR = 1.76, 95% CI: 1.12–2.77, p = 0.014). During the 6–12 month period, the risk further increased to 127% (aHR = 2.27, 95% CI: 1.38–3.73, p = 0.001). This difference was most pronounced during the first year of follow-up (aHR = 3.79, 95% CI: 1.92–7.48, p < 0.001). During extended follow-up, the risk difference became less pronounced, yet the overall recurrence risk remained significantly elevated among patients with low HDL-C (aHR = 1.80, 95% CI: 1.14–2.83, p = 0.011).