Association between red cell distribution width and mortality in patients undergoing continuous ambulatory peritoneal dialysis

Although red cell distribution width (RDW) has emerged as a biomarker of clinical prognostic value across a variety of clinical settings in the last two decades, limited evidence is available for its role in end-stage renal disease. We enrolled 313 incident patients undergoing continuous ambulatory peritoneal dialysis (CAPD) in this retrospective observational study from 2006 to 2015. In the fully adjusted model of Cox regression analysis, the adjusted hazard ratios for the high RDW group versus the low RDW group were 2.58 (95% confidence interval (CI) = 1.31–5.09, p = 0.006) and 3.48 (95% CI = 1.44–8.34, p = 0.006) for all-cause and cardiovascular disease (CVD)-related mortality, respectively. Based on area under the receiver operating characteristic curve (AUC) analysis, RDW (AUC = 0.699) had a stronger predictive value for all-cause and CVD-related mortality than other biological markers including hemoglobin (AUC = 0.51), ferritin (AUC = 0.584), iron saturation (AUC = 0.535), albumin (AUC = 0.683) and white blood cell count (AUC = 0.588). Given that RDW is a readily available hematological parameter without the need for additional cost, we suggest that it can be used as a valuable index to stratify the risk of mortality beyond a diagnosis of anemia.


Results
Patient characteristics. The study cohort included 313 patients undergoing CAPD from 2006 to 2015. The baseline characteristics of these patients stratified by median RDW value (15.3%) are shown in Table 1. The mean age was 54.5 ± 15.9 years, and 164 (52.4%) were male. The three leading causes of ESRD were chronic glomerulonephritis (34.1%), diabetes mellitus (28.1%) and hypertension (17.8%). Most of the patients (248, 79.2%) had pre-dialysis CKD before initiating PD. At baseline, the patients in the higher RDW group (> 15.3%) were older, had lower urine output and lower residual renal function. With regards to laboratory examinations, the patients in the high RDW group had lower levels of albumin, calcium, hemoglobin, and cholesterol, and higher levels of alkaline phosphate, ferritin and blood urea nitrogen (BUN) compared to the lower RDW group (≦ 15.3%). With regards to pharmacotherapy, more patients in the low RDW group used iron preparations.
Association of RDW with all-cause and CVD-related mortality. During the study period, 27 patients (17.4%) died in the low RDW group and 64 patients (40.5%) died in the high RDW group (p < 0.001). Of these 91 patients, 48 died of CV events, which was the leading cause of mortality. There was also a significant difference in CVD-related mortality rate between the two groups, with 14 patients (9%) in the low RDW group and 34 (21.5%) in the high RDW group (p = 0.003). Kaplan-Meier survival curves showed that the high RDW group had higher all-cause and CVD-related mortality rates compared to the low RDW group (Figs 1 and 2; p < 0.001, p < 0.001, respectively). In the unadjusted and adjusted Cox proportional regression models, the high RDW group was associated with an increased risk of all-cause and CVD-related mortality compared with the low RDW group ( Table 2). In the fully adjusted model (model 5), the adjusted HRs for the high RDW group versus the low RDW group were 2.58 (95% CI = 1.31-5.09, p = 0.006) and 3.48 (95% CI = 1.44-8.34, p = 0.006) for all-cause and CVD-related mortality, respectively. Subgroup analyses showed that the patients with higher RDW levels had higher rates of all-cause and CVD-related mortality in the adjusted models compared to those with lower RDW levels (Figs 3 and 4).

Sensitivity analysis.
Three levels of sensitivity testing were performed as shown in Table 2. In the fully adjusted model, a higher RDW level was associated with a higher risk of overall and CVD-related mortality in all of the three sensitivity analyses. Table 3 shows the strength of association and correlation of RDW with other parameters by Pearson correlation test and linear regression analysis. A negative association was disclosed between RDW and albumin, hemoglobin, intact parathyroid hormone, body mass index, and cholesterol. However, these associations were weak (the absolute value of Pearson correlation < 0.3). Albumin had the strongest correlation with RDW, with a 0.64% decrease with every 10-g/L increase in albumin level (Pearson coefficient, − 0.196).

Association of RDW with different variables.
Predictive value of RDW. We first calculated the area under the curve (AUC) in receiver operating characteristic (ROC) analysis to compare the predictive value of a single variable in predicting overall and CVD-related mortality within 1-year, 3-year and 5-year periods. As shown in Table 4, RDW had the highest predictive value compared to the other variables over the study period, except for albumin which had the highest value in predicting 3-year overall mortality. We then calculated the AUC after adding each variable to model 4 (Table 4). Adding RDW to model 4 resulted in the highest AUC compared to the other variables, implying that RDW was a better index in predicting 1-year, 3-year and 5-year overall and CVD-related mortality.

Discussion
In this study, we investigated the relationship between baseline RDW levels and patient survival in 313 incident CAPD patients over a period of 10 years at a single PD center, and found a robust and consistent relationship between high RDW and overall and CVD-related mortality independent of other common risk factors. In addition, RDW was superior to albumin, ferritin, WBC, iron saturation and hemoglobin in predicting the risk of overall and CVD-related mortality based on AUC analysis in both univariate and multivariate models. In contrast to previous studies on RDW, the association between RDW and laboratory variables was weak with a negative association with albumin being the strongest correlation.
RDW has been used to differentiate the causes of anemia in clinical practice. A high degree of heterogeneous red blood cell (RBC) size is called anisocytosis. An elevated RDW is commonly encountered in patients with impaired erythrocyte production or increased erythrocyte destruction. In recent studies, RDW has been reported to be a significant prognostic marker for the risk of mortality in various diseases, and especially cardiovascular diseases 14 . Considerable and convincing evidence has indicated the close relationship between RDW and acute coronary syndrome, ischemic cerebrovascular disease, peripheral artery disease, atrial fibrillation and heart failure 14 . A high RDW level has also been reported to predict adverse outcomes in patients with these conditions. A meta-analysis of 17 cohort studies by Huang et al. showed the prognostic role of RDW on admission and discharge in patients with congestive heart failure with a 10% increase in the overall risk of mortality for every 1% increase in baseline RDW 15 . Furthermore, in a Chinese population of 1,442 patients with stable angina, a higher RDW on admission was shown to increase the risk of 1-year cardiac mortality and 1-year acute coronary syndrome 16 .
In addition to the prediction of a higher risk of mortality in patients with cardiovascular diseases, RDW has also been implicated in the clinical setting of kidney diseases. Oh et al. reported that RDW at the initiation of continuous renal replacement therapy was an independent predictor of 28-day all-cause mortality after multiple adjustments 17 . Later, a single-center, prospective longitudinal study of 100 hemodialysis patients reported that a RDW value above the median was associated with a hazard ratio of 5.15 for 1-year mortality compared to a lower   Although the exact pathophysiological mechanisms are unknown, substantial evidence suggests a robust and independent relationship between RDW and clinical outcomes in many human diseases. Several plausible explanations have been postulated to explain the relationship between RDW and adverse outcomes. First, a high RDW level indicates a great degree of heterogeneity in RBC size (anisocytosis). In addition, accelerating RBC destruction and/or ineffective erythropoiesis, which are common in patients undergoing dialysis, and bone marrow dysfunction can lead to a high RDW level. Bone marrow-derived mesenchymal stem cells have been reported to play a crucial role in the restoration of many injured vital organs 22 . Therefore, disordered hematopoiesis in dialysis patients may contribute to the high risk of mortality associated with a high RDW. Second, inflammation, which is prevalent in CKD patients, has been closely linked with RDW in many patient populations. Proinflammatory cytokines are well known to inhibit erythropoietin-induced RBC maturation and proliferation 23 . Solak et al. reported a strong association between RDW and CRP in predialysis CKD patients 13 . A similar relationship has also been found in patients undergoing PD and HD 19,20 . Furthermore, inflammation has been shown to be associated with mortality in dialysis patients. Third, the presence of malnutrition and/or protein energy wasting, which is common in dialysis patients, is known to increase RDW. RDW has also been reported to be significantly and inversely correlated with nutritional index in a wide array of medical conditions 12,24 . Fourth, distinct from the investigation by Peng et al., we found that residual renal function was greater in the low RDW group than that in the high RDW group. Residual renal function has been shown to have a beneficial effect on patient survival, and especially in patients undergoing PD 25 . In addition, the loss of residual renal function has been implicated in malnutrition and increased inflammation 26 .
Other potential mechanisms have also been proposed to explain the association between RDW and unfavorable outcomes. RDW has been associated with slow coronary flow and left ventricular filling pressure in patients with diastolic heart failure 27,28 . In patients with stage 1-5 CKD, RDW has been shown to independently predict endothelial dysfunction, which may be responsible for the high CVD burden in patients with CKD 13 . In addition, a recent study of patients with advanced CKD by Leszek et al. reported a relationship between RDW and left ventricular diastolic dysfunction 29 . Elevated RDW may lead to increased mortality through impaired microcirculation, ischemia and thrombosis as a result of reduced RBC deformability 30 . Finally, anisocytosis also enhances the accumulation of erythrocytes in the atherosclerotic lesions, leading to the neutralization of vasodilators, growth, ulceration and thrombosis of the fibrous cap 31 .
There are several limitations to this study. First, as a retrospective and observational study, we could not prove a causal relationship between RDW and mortality. Second, a single measurement of RDW may underestimate the true relationship with study outcomes. Third, a single-center investigation limits the extrapolation of the findings to the whole PD population. Fourth, the possibility of an epiphenomenon reflecting the complex interactions between RDW and other un-evaluated risk factors cannot completely be excluded. For example, inflammation, oxidative stress and poor nutrition usually accompany CKD and they are associated with impaired production of erythropoietin, which promotes the release of erythrocytes of heterogeneous size from the bone marrow 31,32 . However, oxidative stress was not evaluated in this study. Therefore, RDW should not be considered a causative factor, but rather a valuable marker in assessing the risk of mortality. Fifth, Lippi et al. reported the lack of harmonization of RDW by using four different hematological analyzers 33 . The optimal cut-off value of RDW for the prediction of mortality risk might vary and depend on the used hemocytometer.
In conclusion, we found that a higher RDW was an independent risk factor for overall and CVD-related mortality in patients undergoing CAPD, and that its predictive value was better than other makers of anemia. It is unclear whether RDW is a true risk factor for mortality or merely an integrated biomarker reflecting anemia, inflammation, malnutrition and impaired kidney function. Nevertheless, given that RDW is a readily available hematological parameter without the need for additional cost, we suggest that it can be used as a valuable index to stratify the risk of mortality beyond a diagnosis of anemia. Further investigations are required to verify our findings and elucidate the underlying mechanisms.

Materials and Methods
This retrospective longitudinal study was performed at a single PD center at Changhua Christian Hospital, Taiwan. We recruited all incident patients who started CAPD as renal replacement therapy between 1 January 2006 and 31 October 2014. Patients were excluded if they were under 18 years of age (n = 5) and had received PD for less than 3 months (n = 8). The final study cohort consisted of 313 adult patients undergoing CAPD, all of whom were followed from the index date, defined as the date of initiating CAPD, until the date of death or the end of the study period (31 October 2015), whichever occurred first.
Measurement of red blood cell parameters was carried out using the automatic hematology analyzer (DxH 800, Beckman Coulter). RDW was calculated using standardized methods and routinely reported as part of the complete blood cell count as a percentage. The whole study cohort was divided into two groups by the median RDW value (15.3%) to assess the predictive value of RDW on study outcomes as: the high RDW group (> 15.3%), and low RDW group (≦ 15.3%). CVD was the leading cause of mortality, followed by infection. The study outcomes were all-cause and CVD-related mortality. This study was approved by the institutional review board of Changhua Christian Hospital and conducted in compliance with the declaration of Helsinki. Written informed consent was not required for this retrospective study due to its non-intrusiveness and patient anonymity.
Statistical analysis. Descriptive data were expressed as number (N) and proportion, or mean ± standard deviation (SD) for categorical and continuous data, respectively. Differences between patients with high and those low RDW were compared using the Student's test or Mann-Whitney test for continuous variables, and the chi-square test or Fisher's exact test for categorical variables. Survival curves were calculated using the Kaplan-Meier method, and differences in survival were assessed using the log-rank test. Cox proportional hazard analysis was used to evaluate the association between RDW and study outcomes, including all-cause and CVD-related mortality, initially without adjustments. Multivariate Cox regression analysis was then performed with adjustments for the covariates which showed a significant correlation (p < 0.05) with the outcome of interest, including sex, age, BMI, smoking status, comorbidities, laboratory variables, medications and PD-related parameters.
Five models were used: model 1, adjusted for sex, age, smoking status and BMI; model 2, adjusted for all variables in model 1, plus medications; model 3, adjusted for all variables in model 2, plus comorbidities; model 4, adjusted for all variables in model 3, plus PD-related parameters; model 5, adjusted for all variables in model 4, plus laboratory data. Three sensitivity analyses were performed to increase the robustness of our results. First, the hazard ratio (HR) of RDW was calculated per 1% increment in RDW level to maximize the predictive value of RDW for the clinical outcomes of interest. Second, the entire cohort was divided into three tertiles by the RDW level with the first tertile group as the reference in relation to the risk of mortality. Third, the optimal RDW value determined from the ROC analysis was used to divide the patients for statistical analysis.
The associations between RDW and laboratory variables were tested using Pearson rank correlation test, and linear regression was used to determine the expected changes in RDW with each unit change in these continuous variables.
ROC analysis with AUC analysis was conducted to compare the predictive value of RDW for all-cause and CVD-related mortality within 1 year, 3 years and 5 years of initiating PD with other markers of anemia (hemoglobin, ferritin and iron saturation), albumin and WBC. The predictive value of these variables was further examined by the AUC with each variable added to those variables in Cox regression model 4. We also tested the possible