This multicenter retrospective cohort study was conducted among adult patients who initiated PD between January 2005 and December 2021 in six tertiary PD centers in China. The study followed the STROBE guidelines for observational studies. Eligible participants were those aged ≥ 18 years who commenced PD as their first renal replacement therapy and remained on PD for ≥ 3 months. Exclusion criteria included: (1) prior hemodialysis or kidney transplantation; (2) active malignancy; (3) acute infection at baseline; (4) outliers of lipid data; and (5) missing baseline lipid parameters required for calculating the LCI.

The study protocol adhered to the principles of the Declaration of Helsinki and was approved by the Ethics Committee of the First Affiliated Hospital of Nanchang University [Approval No. 852 IIT (2025)] and by the ethics committees of participating centers. Patient data were anonymized prior to analysis.

Baseline data were obtained from electronic medical records at the time of PD initiation. Demographic and clinical variables included age, sex, body-mass index (BMI), systolic and diastolic blood pressure (SBP, DBP), primary kidney disease, diabetes mellitus, hypertension, history of CVD, and medication use (statins, aspirin, β-blockers, renin–angiotensin-system inhibitors, and calcium-channel blockers). PD-specific parameters, such as Kt/V, weekly CrCl, and peritoneal transport status, were not included, as these are typically measured during follow-up after the initiation of PD. Although we considered incorporating these PD-related data from the first 6 months of dialysis, significant missing data across the multicenter cohort led to their exclusion to preserve the integrity and robustness of the analysis.

Laboratory variables included fasting lipid profile, TC, TG, LDL-C, and HDL-C, together with hemoglobin, serum albumin (ALB), uric acid (UA), alkaline phosphatase (ALP), calcium, phosphate, C-reactive protein (CRP), creatinine, and RRF (estimated by 24-h urine collection). All tests were performed in the certified laboratories of each center using unified methods and regular external quality control. The LCI was computed as (TC × TG × LDL-C)/HDL-C, with all lipid values expressed in mmol/L. Participants were categorized into quartiles (Q1 - Q4) according to baseline LCI values, with Q1 indicating the lowest and Q4 the highest lipid burden. The cut-points for quartiles were determined using the 25th, 50th, and 75th percentiles of LCI values.

Patients were reviewed quarterly in each PD center and contacted monthly by trained dialysis nurses via outpatient visits or telephone. The primary endpoint was all-cause mortality; the secondary endpoint was cardiovascular mortality. Cardiovascular death was defined as death resulting from acute myocardial infarction, heart failure, arrhythmia, sudden cardiac death, or cerebrovascular events (stroke or intracranial hemorrhage) (15).

Follow-up continued until death, kidney transplantation, transfer to hemodialysis, loss to follow-up, or 31 December 2021, whichever occurred first. When death occurred outside hospital, the cause was verified by physician review of family-reported circumstances and prior medical records.

Continuous variables were presented as mean ± standard deviation (SD) or median (interquartile range, IQR) according to data distribution, and categorical variables as counts (percentages). Baseline differences across LCI quartiles were compared using one-way ANOVA or Kruskal-Wallis tests for continuous data and the chi-square test for categorical variables.

Missing baseline covariates were imputed using multiple imputation by chained equations (MICE) with predictive mean matching (five imputations, fifty iterations, seed = 500). Although some variables, including pre-existing CVD, had higher missing proportions (≈48%), all clinically relevant variables were included in the imputation model to minimize bias and preserve statistical power. The results from the multiple imputed datasets were pooled using Rubin’s rules to obtain unbiased overall estimates and standard errors. To further assess the impact of missing data, we performed a Missing Not at Random (MNAR) sensitivity analysis. We used MICE with the selection model for imputation under the MNAR assumption, generating five imputed datasets. The results from these datasets were pooled using Rubin’s rules. The association between LCI and mortality outcomes was evaluated using Cox proportional hazards regression models, with results stratified by center. We compared the findings from the MNAR analysis with those from the original analysis (based on the MAR assumption) to assess whether the missing data affected the observed associations.

Survival probabilities were estimated by Kaplan–Meier curves, and log-rank tests were applied to assess inter-quartile differences. The association between LCI and mortality was examined using Cox proportional-hazards regression models, reporting hazard ratios (HRs) and 95% confidence intervals (CIs). The proportional-hazards assumption was verified by Schoenfeld residuals. To account for potential baseline hazard differences among participating PD centers, all Cox models were stratified by center (i.e., strata(center)). Three hierarchical models were constructed: Model 0: unadjusted; Model 1: adjusted for age and sex; Model 2: further adjusted for BMI, SBP, DBP, diabetes, pre-existing CVD, hemoglobin, serum ALB, UA, calcium, phosphate, CRP, ALP, RRF, and use of statins or aspirin. In the multivariate Cox proportional hazards regression, variables with a p value < 0.1 in univariate analysis were initially selected. Additional variables were included based on clinical relevance and their known association with lipid metabolism and mortality.

To explore non-linear relationships between continuous LCI and mortality, restricted cubic spline (RCS) analyses were performed. Pre-specified subgroup analyses were conducted according to sex (male/female), age (< 60 vs. ≥ 60 years), diabetes (yes/no), CVD (yes/no), serum ALB (< 35 vs. ≥ 35 g/L), and RRF (<median vs. ≥ median), with multiplicative interaction terms tested for significance.

Fine-Gray competing-risk models were used for cardiovascular mortality, treating non-cardiovascular death as a competing event. Sensitivity analyses were undertaken by (1) excluding deaths within 6 months of PD initiation, (2) reclassifying LCI into tertiles instead of quartiles, and (3) repeating the analysis using complete-case data without imputation.

All analyses were performed using R (version 4.3.3). Two-sided p values < 0.05 were considered statistically significant.

A total of 1,986 incident PD patients were included in the final analysis (Figure 1). The median follow-up duration was 35 months (interquartile range 18–57). During follow-up, 662 (33.3%) patients died, of whom 328 (49.5%) deaths were attributed to cardiovascular causes.

Patients were categorized into LCI quartiles (Q1 ≤ 7.03, Q2 7.03–13.57, Q3 13.57–28.71, Q4 > 28.71). Across increasing quartiles, BMI and the prevalence of diabetes were higher, while HDL-C was lower; TC, TG and LDL-C were progressively higher (all P for trend < 0.001). Hemoglobin and serum ALB were modestly higher in the highest LCI quartile, whereas CRP did not differ materially between groups. SBP and DBP showed no significant differences across quartiles. The proportion of men was lower in Q4 than Q1, statin use was more frequent with higher LCI, and aspirin use was similar across groups. Calcium was slightly higher and phosphate slightly lower with increasing LCI. Although a history of CVD did not differ significantly across quartiles, this finding may partly reflect the substantial missingness for this variable (48.6%), which could limit the reliability of the comparison (Table 1).

During the follow-up, the cumulative survival rates for both all-cause and cardiovascular mortality decreased progressively with increasing LCI quartiles (Figure 2). Kaplan–Meier curves showed significant between-group differences [log-rank p < 0.0001 for all-cause mortality (A); p = 0.00038 for cardiovascular mortality (B)].

In the multivariable Cox proportional-hazards models (Table 2), higher LCI was independently associated with increased risks of both all-cause and cardiovascular mortality. Using the lowest quartile (Q1) as the reference, the adjusted HRs for all-cause mortality were 1.41 (95% CI 1.10–1.80) for Q2, 1.59 (95% CI 1.25–2.02) for Q3, and 1.70 (95% CI 1.34–2.15) for Q4 (P for trend < 0.001). Similarly, for cardiovascular mortality, the corresponding HRs were 1.45 (1.03–2.02), 1.29 (0.92–1.82), and 1.68 (1.22–2.33) (P for trend = 0.006). When analyzed as a continuous variable, each 1- SD increase in LCI was associated with a 16% higher risk of all-cause mortality (HR 1.16, 95% CI 1.08–1.24; p < 0.001) and a 18% higher risk of cardiovascular mortality (HR 1.18, 95% CI 1.07–1.31; p = 0.001).

These findings remained consistent when competing non-cardiovascular deaths were considered using the Fine-Gray subdistribution hazards model, where the highest quartile had a 54% higher subdistribution hazard for cardiovascular mortality compared with Q1 (sHR 1.54, 95% CI 1.11–2.12, p = 0.009) (Supplementary Table S1).

RCS analyses demonstrated a non-linear, positive association between LCI and all-cause mortality (P for non-linearity = 0.020), with mortality risk rising sharply when LCI exceeded approximately 20 (Figure 3). In contrast, the association with cardiovascular mortality appeared linear (p = 0.20).

Subgroup analyses (Figure 4) showed that the positive association between high LCI and mortality was broadly consistent across sex, age (< 60 vs. ≥ 60 years), diabetes status (yes vs. no), pre-exist CVD (yes vs. no), serum ALB (< 35 vs. ≥ 35 g/L), and RRF categories (< medium vs. ≥ medium) (all P for interaction > 0.05). Notably, for cardiovascular mortality, the overall association persisted, and a significant interaction by age was observed, with a stronger effect among younger patients (< 60 years) (P for interaction = 0.0487).

Three sensitivity analyses were performed to assess the robustness of the findings. First, after excluding patients who died within the first six months of PD initiation (n = 1,892), the associations between LCI and mortality persisted. The adjusted HRs (Model 2) for Q4 versus Q1 were 1.74 (95% CI 1.33–2.29, p < 0.001) for all-cause mortality and 1.73 (95% CI 1.20–2.49, p = 0.003) for cardiovascular mortality; each 1-SD increase in LCI corresponded to a 17–20% higher risk of death (Supplementary Table S2). Second, when LCI was categorized into tertiles rather than quartiles, the direction and magnitude of the associations remained unchanged. Compared with T1, the adjusted HRs for T3 were 1.55 (95% CI 1.23–1.94) for all-cause mortality and 1.47 (95% CI 1.08–2.01) for cardiovascular mortality (both P for trend < 0.05) (Supplementary Table S3). Third, In the complete-case analysis including 550 patients (27.7% of the total cohort), the associations between LCI and mortality were attenuated and became statistically non-significant (Q4 vs. Q1: HR 1.05 (0.70–1.57) for all-cause and 0.87 (0.52–1.46) for cardiovascular mortality; both p > 0.05) (Supplementary Table S4). Given the small sample size and possible selection bias among complete cases, these findings were interpreted as reflecting limited statistical power rather than a genuine absence of association. The MNAR sensitivity analysis (Supplementary Table S5) demonstrated that the relationship between LCI and both all-cause mortality and cardiovascular mortality remained consistent with the results from the original analysis (Table 2). Specifically, the hazard ratios for LCI in the MNAR analysis were similar to those observed under the MAR assumption, suggesting that missing data did not significantly alter the observed associations. These findings support the robustness of our results despite the high rate of missing data.

Overall, the sensitivity and competing-risk analyses (Supplementary Tables S1–S5) generally supported the robustness of the association between higher LCI and increased mortality among PD patients, although the effect estimates in the complete-case analysis were attenuated and no longer statistically significant.

Higher baseline LCI levels were independently and progressively associated with increased all-cause and cardiovascular mortality in PD patients. The association exhibited a non-linear pattern for all-cause mortality and remained consistent across most subgroups and analytical strategies.