We performed a retrospective multi-center cohort study of individuals selected from the 6,126 participants in the Beijing Hemodialysis Quality Control and Improvement Project, during which data was collected at 138 dialysis centers between 1 January 2012 and 31 December 2019. The inclusion criteria were age ≥ 18 years and hemodialysis three times a week for at least 3 months. The exclusion criteria were as follows (1): duration of hemodialysis < 3 months; (2) previous treatment by peritoneal dialysis; (3) organ transplantation; (4) malignant disease; (5) autoimmune disease, or chronic or acute infectious disease; and (6) missing baseline data. The existence of acute or chronic infection was determined based on the admission diagnosis of patients that explicitly stated infection (for example, pneumonia) or that provided evidence of infection (detected bacteria or virus). After the application of these criteria, 3,856 patients were eligible for enrolment in the present study ( Figure S1 ). The study was performed in accordance with the principles of the Declaration of Helsinki and was approved by the Human Ethics Committee of Beijing Friendship Hospital.

Variables and potential confounders were selected for study by considering clinical guidelines and the results of previously published studies. Only data collected during the first assessment made after the initiation of hemodialysis were included. The baseline demographic data (age and sex) and biochemical/hematological data (hemoglobin, albumin, platelet count, neutrophil count, lymphocyte count, C-reactive protein, creatinine, urea, calcium, parathormone, phosphorus, Fe, ferritin and UIBC) were obtained within the first month of hemodialysis. The biochemical/hematological data were standardized before being recorded in the Beijing Hemodialysis Quality Control and Improvement Project database to minimize variability in the measurements made by the various dialysis laboratories. All of these data were collected from the project database. The patients were followed from the initiation of hemodialysis until they were transferred to another hemodialysis center, they were changed to peritoneal dialysis, they underwent kidney transplantation, they were lost to follow-up, they died, or the end of the study period on 31 December 2019. The LCR was calculated as lymphocyte count (10⁹/L)/C-reactive protein (mg/L) × 10⁴. For the subsequent stratified analyses, on the basis of the laboratory reference ranges and the Kidney Disease Outcome Quality Initiative (KDOQI) guidelines, the biochemical data were defined as normal or abnormal. The normal values were as follows: hemoglobin 100–130 g/L, platelet count 125–350×10⁹/L, neutrophil count 1.8–6.3×10⁹/L, albumin >35 g/L, calcium 2.1–2.52 mmol/L, parathormone 150–300 pg/μl, phosphorus 1.13–1.78 mmol/L, ferritin 200–500 ng/ml.

The overall survival time was defined as the interval between the initiation of hemodialysis and the date of death, transfer to another dialysis center, the change to peritoneal dialysis, kidney transplantation, withdrawal from the study, or the end of the study period (31 December 2019). We aimed to evaluate the relationship between LCR and overall survival and to identify the most appropriate cut-off value of LCR for prognostic use in patients undergoing regular hemodialysis.

Continuous datasets with skewed distributions are presented as median (interquartile range) and were compared using the Mann-Whitney test. Categorical data are expressed as numbers (percentages) and were compared using Pearson’s χ² test. Spearman correlation analysis was used to identify linear relationships between LCR and selected variables. Univariate and multivariate Cox proportional hazards regression models were used to identify the risk factors for mortality and to provide hazard ratios (HRs) for each variable. Sex and age were included in the multivariate Cox regression model and other potentially confounding variables were selected based on the results of univariate analysis (p<0.1 was defined as indicating a significant association with all-cause mortality). For the time-to-event analysis, survival curves were generated using the Kaplan-Meier method and compared using the log-rank test.

The non-linear relationship between LCR and HR was visualized using restricted cubic splines. The optimal cut-off value for LCR was determined using the ‘surv_cutpoint’ formula in the ‘survminer’ R package, which is an outcome-oriented method that provides a cut-off value with the closest relationship with the outcome. The participants were divided into high- and low-LCR groups based on the optimal cut-off value. Trend tests were performed by assigning a median value to each quartile of the LCR, which was then modeled as a continuous variable, and the Wald test was used to assess statistical significance. Forest plots were used to visualize the results of the analysis of the effects of the interactions of LCR with other variables on the overall survival of the participants. Calibration curves were used to visualize the differences between the predicted and observed probabilities. The accuracy of predictions made using the LCR risk score was assessed using the area under the ROC curve. All the tests performed were two-sided and P<0.05 was considered to represent statistical significance. Statistical analyses were performed using R software version 4.0.2 (R Foundation for Statistical Computing, Vienna, Austria), and the packages ‘survminer’, ‘survival’, ‘rms’,’ timeROC’, ‘forestplot’, ‘ggrisk’, ‘ggplot2’, ‘ggsci’, ‘tableone’, and ‘dplyr’.

From the original total of 6,126 patients, we excluded 2,270 for the reasons listed above, leaving a total of 3,856 participants in the cohort. They had a median age of 62 (53, 73) years and 1,581 (41%) were female. The overall characteristics of the participants and the results of their stratification according to the calculated LCR cut-off value are presented in Table 1 . In summary, 1,129 deaths were recorded over a median follow-up period of 75.1 months. The overall mortality rate for the participants 60 months after their initial assessment was 38.1% (95% CI 36%–40.1%), resulting in a rate of 93.41 events per 1,000 patient-years. The log LCR was compared among the various sex and age groups comprising the sample. The results showed that young (<65 years old) participants had a significantly higher LCR than older (≥65 years old) participants (P<0.001) ( Figure S2 ). Moreover, Spearman analysis of the relationships of LCR with various parameters that are clinically relevant for patients undergoing hemodialysis showed significant correlations with age, hemoglobin level, neutrophil count, albumin concentration, and calcium concentration. On the basis of these results, we also conducted a stratified analysis of participants according to their age (≥65 years old vs. <65 years old) and sex (female vs. male), which showed that hemoglobin level, albumin concentration, and calcium concentration positively correlated with LCR in individuals of differing age and sex, while the neutrophil count negatively correlated ( Figure 1 ).

Potential risk factors for all-cause mortality were identified using univariate and multivariate Cox proportional hazards regression analyses ( Table 2 ). Univariate analysis revealed that age, platelet count, lymphocyte count, CRP, creatinine, urea, albumin, phosphorus, ferritin, and LCR were associated with the overall survival of the participants. The multivariate analysis revealed that age, platelet count, albumin, phosphorus, and LCR were independent predictors of mortality. Then, according to the results of the multivariate analysis, we evaluated the prognostic value of combinations of LCR with the other independent prognostic factors, and the 1, 3, 5, and 7-year calibration curves showed that these combinations would be extremely useful for the prediction of survival ( Figure S3 ). The cut-off point of LCR determined by using Kaplan-Meier curves was 1513.1 ( Figure 2A ), of which hemodialysis patients with an LCR <1513.1 were found to be associated with higher mortality.

To further evaluate the relationship between LCR and mortality in patients undergoing hemodialysis, four categories of LCR were defined. First, as a continuous variable, the restricted cubic spline plot showed that LCR had an L-shaped dose-response relationship with the risk of all-cause mortality risk in the participants ( Figure 3 ). Second, the Cox regression models of the relationship between LCR and OS showed that LCR positively correlated with prognosis (HR 0.85 per SD increase, 95% CI 0.75–0.95, P=0.006) after adjustment for sex, age, hemoglobin level, platelet count, neutrophil count, creatinine, urea, albumin, calcium, parathormone, phosphorus, Fe, ferritin, and unsaturated iron-binding capacity (UIBC) ( Table 3 ). Third, using the cut-off value calculated for LCR, the participants were divided into two groups: a low-LCR group and a high-LCR group. Compared with the low-LCR group, participants in the high-LCR group had a consistently better prognosis (HR 0.75, 95% CI 0.66–0.85, P<0.001) after adjustment for the variables listed above. Finally, the participants were divided into quartiles according to their LCR; and compared with the first quartile (Q1, <1,054.3), the second (1,054.3–3,144.9), third (3,144.9–9,068.6), and fourth quartiles (≥9,068.6) all had a better prognosis (P for trend <0.05). After adjustment for the potential confounding factors, the HRs for all-cause mortality were 0.78 (95% CI 0.66–0.92, P=0.003), 0.73 (95% CI 0.62–0.87, P<0.001), and 0.76 (95% CI 0.64–0.91, P=0.003) for the second, third, and fourth quartiles, respectively.

According to the cut-off point calculated for LCR, the 3,856 participants were divided into two groups: a low-LCR group (LCR <1513.1, n=1,223) and a high-LCR group (LCR ≥1513.1, n=2633). The Kaplan-Meier curves and log-rank test results revealed that the high-LCR group had a better prognosis than the low-LCR group ( Figure 2B ). Table 1 presents a comparison of the demographics and clinical characteristics of the low- and high-LCR groups. Briefly, the participants in the low-LCR group were older; and had lower hemoglobin, higher neutrophil counts, lower lymphocyte counts, higher CRP, lower creatinine, lower urea, lower albumin, lower calcium, lower phosphorus, lower Fe, higher ferritin, and lower UIBC than those in the high-LCR group.

Stratified analyses were conducted to evaluate the relationships between LCR and the HR for overall mortality in various subgroups ( Figure 4 and Table S1 ). Overall, high LCR was consistently associated with a lower risk of death in the participants, irrespective of their subgrouping. However, although the same trend was also present in participants with a normal calcium concentration, it was not statistically significant (P=0.234). Moreover, an analysis was performed to explore the interactions between high LCR and the other variables, but this showed no associations between high LCR and low overall risk of mortality in the participants (P for interaction >0.05 in all instances). In addition, the LCR and covariates were then cross-classified to better understand the effects of each variable ( Table S2 ). This analysis showed that an abnormal status with regard to any of a number of variables and a low LCR had an additive effect to increase the risk of mortality. Kaplan-Meier curves also showed that combinations of low LCR and abnormalities in other variables had a deleterious effect on mortality. Specifically, participants with an LCR of <1513.1 had the worst survival rate when they were ≥ 65 years old and had an abnormal albumin concentration ( Figure S4 ).

To validate the finding that LCR is a useful predictor of mortality in patients undergoing hemodialysis, a sensitivity analysis and internal validation were performed to assess the robustness of the results ( Table 4 ). An analysis performed after excluding the participants who died within 6 months of the first assessment showed that LCR remained an independent predictor of mortality (adjusted HR 0.85, 95% CI 0.75–0.96, P=0.007 for a high LCR per SD). Subsequently, the full cohort was randomly assigned at a 7:3 ratio to validation cohort A (n=2,681) or validation cohort B (n=1,175) using computer-generated random numbers ( Table S3 ). Similar results were obtained for cohort A (adjusted HR 0.87, 95% CI 0.78–0.98, P=0.026 for high LCR per SD) and cohort B (adjusted HR 0.69, 95% CI 0.49–0.97, P=0.034 for high LCR per SD). Moreover, Kaplan-Meier curves showed that participants with a high LCR had a better prognosis when in either validation cohort A or B ( Figures S5A, B ).

LCR values were naturally log-transformed to make it easier to assess the prognosis of patients undergoing hemodialysis, and then a risk score model was constructed using the β regression risk coefficient derived from the multivariate Cox regression model and the log LCR. The formula for the risk score was determined to be −0.1669 × log LCR. Using the calculated risk scores, a risk plot heatmap and a time-dependent ROC curve were created. The results, shown in Figure 5A , indicate that a high LCR is associated with a low risk score and a better prognosis. Figure 5B shows AUCs of 62.0, 56.5, 58.3, and 54.9 for 1, 3, 5, and 7 years, respectively.