We retrospectively conducted a study to investigate whether the HRR synergizes in predicting the risk of new-onset CVE in peritoneal dialysis patients. Patients who commenced PD between November 1, 2005, and December 31, 2016, were qualified for our study. Among them, 104 cases were excluded because of age younger than 18 years or older than 80 years (n=17), missing of data (n=9), duration of PD less than 3 months (n=53), severe infection (n=0), history of hematological or autoimmune disease, or taking glucocorticoid or immunosuppressant (n=25). The above patients were excluded because these factors might impact HRR levels (Fig. 1). Finally, 1474 patients were included in the study. The Institutional Review Board approved this retrospective study. Written informed consent was not required because we needed to collect the hospital's pre-existing medical data.

Fig. 1.

Flow chart-including patient enrollment and outcomes.

Baseline demographic data and laboratory biochemical indicators were collected 3 months after PD therapy was initiated. The main result was new-onset CVD. Baseline demographic and clinical data were collected at the initiation of PD therapy. Biochemical parameters were collected within 90 days after PD therapy was initiated. We considered patients to have diabetes if they were taking insulin or oral hypoglycemic agents and/or had a clinical diagnosis of type 1 or type 2 diabetes mellitus. Hypertension was noted if patients took antihypertensive drugs or had two blood pressure measurements of 140/90mmHg or higher. New-onset CVE was identified if any of the following appeared in the patient's medical records after the initiation of PD: arrhythmia, coronary atherosclerotic heart disease, myocardial infarction, heart failure, ischemic or hemorrhagic cerebrovascular accidents, or if a patient who died from any of the above causes. Laboratory measurements were obtained using standard methods in the clinical laboratory. Total Kt/V was calculated by PD Adequest software 2.0 (Baxter et al.). Medicine use was recorded based on prescriptions. The patients returned to these centers for quarterly evaluation, and trained nurses interviewed the patients by telephone monthly to assess general conditions.

The outcome was the first occurrence of CVE since PD therapy. The follow-up endpoint was the first-time CVE, death, transferring to hemodialysis therapy, kidney transplantation, transferring to other PD centers, and losing to follow-up or censoring on December 31, 2016.

Considering the potential relationship between HRR levels and poor prognosis in PD patients, we assumed that there might be a non-linear relationship between them in PD patients. Eligible patients were divided into two groups according to HRR levels (HRR ≤5.835) and (HRR >5.835). All continuous variables were skewed distribution described as median (25th to 75th percentile), and categorical variables were given as frequency and percentage. The chi-square test was used for categorical variables, and the Mann–Whitney U test was used for skewed continuous variables. The univariate logistic regression model was used to examine the association between the patient's characteristics and poor prognosis, and then multivariate logistic regression was used to analyze the patient's characteristics with predictive odds adjusted for covariates with (P<0.05 in univariate logistic analysis). The incidence of new-onset CVE was analyzed using the Kaplan–Meier curve, and the differences between groups were assessed using the log-rank test. Multivariate COX regression models were constructed using qualified covariates, which were significantly associated with new CVE in multivariate analysis or the characteristics listed in Table 1 or for the importance of clinical concern.

In the COX regression models, time at risk was from study entry until new-onset CVE, all-cause death, transferring to hemodialysis therapy, transferring to kidney transplantation, transferring to other centers, and losing to follow-up or censoring on December 31, 2016. Competitive risk analysis was used to exclude the effects of the above follow-up end events on new-onset CVE. The interaction between subgroup variables of interest, including sex, age, and history of hypertension or diabetes mellitus. A forest plot was used to show the relationship between HRR and new-onset CVE in each subgroup. The statistical analysis was completed using SPSS25.0 and R software (version R-4.2.2, www.project.org). All tests were performed bilaterally, and P<0.05 was considered statistically significant.

A total of 1578 PD patients were included. The median follow-up period was 47 months. One hundred four (7.1%) new-onset CVEs were recorded. Other endpoints included transfer to hemodialysis (n=192), renal transplantation (n=82), transfer to other PD centers (n=11), all-cause death (n=364), and loss of follow-up (n=36). Finally,1474 patients were followed up to 31 December 2017. A comparison of baseline data of different HRR groups is shown in Table 1. The median age of the patients was 51 (40.0, 61.0) years old, of which 797 were male and 677 were female. A total of 305 (20.7%) of the patients had a history of diabetes, 1132 (76.8%) patients had a history of hypertension, 150 (10.2%) patients had a history of CVD, and 8 (0.5%) patients had a history of gastrointestinal bleeding (GIB). Compared to the higher group (HRR >5.835), the lower group (HRR ≤5.835) had fewer females and younger ages, additionally even had a lower prevalence of diabetes, hypertension, and CVD. Regarding treatments, the lower group had lower use of CCB, ACEI or ARBs, α-ketoacid, calcitriol, insulin, and stain but higher EPO and iron. The lower group also had higher creatinine, blood urea nitrogen, chlorine, and phosphorus.

Restricted cubic splines (RCS) were used to analyze the relationship between HHR and new-onset CVE, indicating a non-linear relationship between HHR and observed events (P=0.018) (Fig. 2). Therefore, we took HHR=5.835, namely the HRR value at HR=1, as the cutoff to divide patients into two groups.

Fig. 2.

Relationship between HRR and new-onset CVE by restricted cubic spline (RCS). When HR=1, there was intersection point on the RCS curve. The HRR value of the intersection point, which equaled to 5.835, was taken as the cutoff. HR, hazard ratio; HRR, hemoglobin-to-red cell distribution width ratio; RCS, restricted cubic spline.

Kaplan–Meier curves showed that more new-onset CVE (log-rank=9.721, P=0.002) occurred in the lower HRR group than in the higher HRR group (Fig. 3). Associations of HRR with new-onset CVE with defined models (with the higher HRR group as the reference group) were presented in Table 2. Multivariate COX regression showed that elevated HRR was an independent risk factor for new-onset CVE in PD patients after adjusting for age, sex, complications, and biochemical examination. For each one-unit increase of HRR, the risk of new-onset CVE increased by 1.737 times. Excluding patients with a history of CVD or extending the follow-up period to >6 months, HRR was still predictive (P=0.014; P=0.040, respectively). Patients with PD anemia may be on iron or EPO. HRR still impacted new-onset cardiovascular outcomes in patients with PD after excluding patients on iron and/or EPO (P=0.015; P=0.014; P=0.016, respectively). The hazard ratios and P-values for complete factors were put in Table S1. Hb affected new cardiovascular events in this PD population (P=0.047); however, RDW did not show a better predictive role in this PD population (P=0.165) (Table S2). In competitive risk models, the estimated cumulative incidence curves demonstrated significant differences in different HRR groups for new-onset CVE (P=0.0009) and all-cause mortality (P=0.00004) but not for the occurrence of other events (being transferred to hemodialysis therapy, being transferred to kidney transplantation, being transferred to other centers, being lost to follow-up) (Fig. 4).

Fig. 3.

The Kaplan–Meier curves with new-onset CVE by category of the level of HRR. The curves were constructed using the Kaplan–Meier method and compared using the Mantel–Cox log-rank test. Patients in the high NPAR group (HRR ≤5.835) had a higher risk of the new-onset CVE (log-rank P=0.002).

Fig. 4.

Competitive risk models. Estimated cumulative incidence curves between the new-onset CVE and other competing events for each HRR group. The cumulative incidence curves for different HRR groups are highly significant for the new-onset CVE (P=0.0009) and the CVD mortality (P<0.001), while there is no statistical significance for the occurrence of other competing events (being transferred to hemodialysis therapy, being transferred to kidney transplantation, being transferred to other centers, being lost to follow-up).

Fig. S1 shows the subgroup analysis based on diabetes, hypertension, gender, and age. The forest plot indicates that no interaction was found between the subgroups.

Four study groups were reclassified by the median values of hemoglobin (85g/L) and red cell distribution width (14.4%) as follows: group A (n=346), hemoglobin ≤85 and red cell distribution width ≤14.4%; group B (n=319), hemoglobin ≤85 and red cell distribution width >14.4%; group C (n=309), hemoglobin >85 and red cell distribution width ≤14.4%; group D (n=285), hemoglobin >85 and red cell distribution width >14.4%. The repeated Kaplan–Meier curves showed statistical significance between the redefined four groups and new-onset CVE (Fig. 5). Moreover, in the first 140 months of the follow-up period, the patients in group B (n=399), with hemoglobin ≤85 and red cell distribution width >14.4%, had the highest risk of new-onset CVE than those with good hemoglobin and red cell distribution width level.

Fig. 5.

Sensitivity analyses. The whole cohort was categorized into four groups by the median values of Hb (85g/L) and RDW (14.4%). The repeated Kaplan–Meier curves showed the statistical significance of the association between the redefined groups and new-onset CVE (log-rank P=0.002).