Study group

Data from consecutive patients with stable coronary artery disease undergoing stent implantation between 2007 and 2011 at our institution (the Silesian Center for Heart Diseases) were analyzed. To identify patients with stable coronary artery disease, we screened all patients with the diagnosis codes of I25.0 and I25.2, as well as patients with other diagnoses who met the following criteria: i) elective hospital admission and ii) stent implantation. Patients undergoing concomitant transcatheter aortic valve implantation procedure, patients undergoing hybrid revascularization, and patients after orthotropic heart transplant were not considered in the first place. We have identified 2774 patients with stable coronary artery disease. Patients who died during hospitalization (n = 4), patients on dialysis (n = 11), those with advanced valve disease (n = 203), a history of cancer (n = 26) or other diseases potentially limiting survival (n = 18) were excluded from the analysis. Final cohort consisted of 2550 patients.

Data source

Starting in 2006, it has been compulsory for every attending physician at our Institution to fill out a complex report form for all admitted patients. This report form includes clinical data, past medical history and performed procedures. The form includes detailed data on a patient’s medical history and clinical characteristics at admission, and it resembles the cardiac report form used in clinical studies. Before patient documentation is given to the hospital information archive, the course of hospitalization is entered, and the report form is checked for completeness. Despite these strict measures, 11 (0.4%) patients were found to have missing data regarding information on the family history (FH) of premature coronary heart disease (CHD), Canadian Cardiovascular Society class, heart rate or systolic blood pressure (SBP) at admission. Data on ejection fraction were available for 2322 (91.1%) patients. Hemodynamic data were available for all patients and were taken from angiography descriptions. Creatinine, sodium levels and complete blood counts were available for all patients. The complete blood counts were performed using the Sysmex XS1000i and XE2100 (Sysmex Corporation, Kobe, Japan). Red cell distribution width (RDW) is calculated using the following formula: RDW = (standard deviation of red blood cell corpuscular volume)/(mean corpuscular volume (MCV)) × 100 [%]. The creation of the database of patients with stable coronary artery disease used in this study was supported by the National Science Center – Dec-2011/01/D/NZ5/04387. Study was approved by ethics committee at district chamber of physicians.

Statistical analysis

The continuous variables are presented as the means and standard deviations. The categorical variables are presented as percentages. Patients were divided into subgroups according to RDW quartiles. Group I (n = 607) comprised patients with an RDW < 13.1% (1^st quartile), group II (n = 574) comprised patients with an RDW value of ≥13.1% and < 13.6% (2^nd quartile), group III (n = 663) comprised patients with RDW values of ≥ 13.6% and <14.1% (3^rd quartile) and group IV (n = 706) comprised patients with RDW values ≥ 14.1% (4^th quartile). To test for differences across all groups, the chi square and Kruskall-Wallis tests were used.

Survival and regression analysis

The associations between the RDW quartiles and mortality were analyzed using the Kaplan-Meier method with log-rank testing. To assess the impact of the RDW on prognosis, a multivariate Cox regression analysis was performed. RDW was imputed as a continuous variable. To minimize the impact of missing data on the Cox regression analysis, the multiple imputation method was used to impute missing data for the variables that were to be included in the Cox regression procedures. The model was adjusted for age, sex, heart failure, atrial fibrillation, hypertension, previous myocardial infarction (MI), previous PCI, previous coronary artery bypass graft surgery (CABG), previous sudden cardiac death (SCD), peripheral vascular disease (PVD), previous stroke, diabetes, lipid abnormalities, obesity, chronic obstructive pulmonary disease (COPD), chronic kidney disease (CKD), smoking, New York Heart Association (NYHA) and CCS class, heart rate, blood pressure, ejection fraction, number and type(s) of stent implanted, number of PCI vessels, hemoglobin, MCV. Additionally stratified analyses according to gender, age (over or under 75 years) diabetes, CKD, anemia, and heart failure status were performed. Difference in Cox-model including aforementioned variables was compared with Cox model including additionally RDW using likelihood ratio test. Discriminative ability of those two models was assessed using Harrell’s C-statistics [12]. We have also compared those models by means of Net Reclassification Improvement (NRI) and Integrated Discrimination Improvement (IDI). NRI estimates if the addition of new variable correctly increases or decreases the predicted risk for events and non-events [13–16]. We have calculated NRI using predicted probabilities estimated by Cox regression analysis at the end of follow-up time. IDI describes differences in integrated sensitivity and integrated one minus specificity between models [13].

Follow-up data

Information on survival was based on the National Health Fund (NFZ) insurance status, which can be electronically verified. Because the NFZ insurance policy is obligatory for all Polish citizens, patients who were insured were marked as alive. We made an attempt to contact the relatives of uninsured patients and/or the relevant local registry office to obtain the exact date of death. Follow-up data were available for 2535 (99.4%) of patients. The mean follow-up period was 915.4 ± 525.3 days. During the observation period, there were 233 reported deaths. All reported p-values are two-sided. The analyses were performed using Number Crunching Statistical Systems 8.0 (NCSS, Kaysville, UT, USA) and in R software [17].

Baseline characteristics of the entire population

Survival analysis and predictors of outcome

During the observation period, there were 233 reported deaths. Mortality in quartiles 1 to 3 was significantly lower as compared with quartile 4 (quartile 1 – 26 (4.3%); quartile 2 – 39 (6.8%); quartile 3–48 (7.3%); quartile 4–120 (17.1%), p < 0.0001, respectively) Figure 1. RDW was highly significantly associated with mortality in the entire cohort and in a subset of patients stratified by age, gender, anemia status, CKD, diabetes and heart failure (Table 4). Adding RDW to model based on clinical variables, ejection fraction hemodynamic data creatinine, and MCV resulted significantly improved the model (p < 0.001) although Harrell’s-C statistic increased only slightly from 0.76 (95% CI, 0.69÷0.82) to 0.77 (95% CI, 0.70÷0.83). Adding RDW to model based on clinical, hemodynamic and laboratory parameters (including MCV) however did not change IDI significantly (IDI - 0.019 95% CI: - 0.005 ÷ 0.051, p =0.120). There was also non-significant change in the risk of death predicted by the models with and without RDW (NRI - 0.251 95% CI: - 0.163 ÷ 0.512 p = 0.213).

RDW as a marker of disease burden

An increased percentage of patients with serious comorbidities compared to patients with lower RDW values suggests that RDW is a universal marker of disease burden. Data from analyses referring to the wide spectrum of diseases seem to confirm that finding [3, 18–28]. Associations between higher RDW levels and CKD are quite easy to explain because anemia, which is frequently found in patients with impaired renal function [29, 30], is associated with increased RDW values. Moreover, RDW values are often used for anemia classification [31]. An explanation similar to that for CKD may hold for heart failure because Jankowska E et al. showed that iron deficiency develops during the progression of heart failure [32]. In our study, the frequency of heart failure and the percentage of patients with advanced heart failure symptoms were higher in patients with RDW values within the 4^th quartile. The association of RDW with other comorbidities such as COPD, diabetes and peripheral vascular disease may be more difficult to elucidate. Possible explanations come from publications describing the increased inflammatory status and increased oxidative stress in those conditions [33–38]. Therefore, in more advanced stages of atherosclerosis, COPD and diabetes, the detrimental effects of oxidative stress on erythrocyte membrane fluidity affects the lifespan of red blood cells, which in turn leads to higher RDW values [39]. However, Fornal M et al. and Lippi G et al. [40, 41] reported the existence of a potential link between inflammatory biomarkers and RDW values that may also be of importance, because the inflammation may impair iron metabolism and inhibit both the production of and the response to erythropoietin [18, 42]. Veranna V. et al. reported that RDW values of 12.6% and increased CRP levels above 3 mg/dl are associated with a higher risk of mortality in a cohort free of coronary heart disease [43]. Another interesting observation is the lower frequency of multivessel coronary artery disease (MVD) in patients with RDW values in the 1^st quartile compared to the other quartiles, which is in concordance with results of Isik T. et al. and MA F-L [44, 45]. Those findings may indirectly confirm the association of increasing values of RDW with extensiveness and possibly the duration of the atherosclerotic process and the role of RDW as a marker of the disease process.

Red blood cell heterogeneity as a potential causative factor

The association between RDW values and comorbidities does not entirely explain the increased mortality in patients with the highest RDW values because RDW remained highly significantly associated with mortality after adjusting for clinical echocardiographic, hemodynamic and laboratory parameters in the entire cohort and in subgroups, although in our study it did not improve risk prediction as estimated by NRI and IDI measures. There have been however studies showing that RDW improves prediction of bleeding after acute coronary syndromes [46] and mortality in kidney transplant recipients [47].

Currently, it is not clear why higher RDW levels are so strongly associated with worse long term prognoses in various diseases. Luneva O.G. et al. provided insight into the pathophysiology of the relationship between mortality and RDW values by finding a significant correlation between RDW values and the cholesterol content of erythrocyte membranes, which also determines erythrocyte membrane fluidity [39]. It has also been reported that greater variation in RDW is associated with impaired blood flow through the microvascular system, which may cause tissue hypoxia, even in patients without anemia [48].

Strengths and limitations

A limitation of this study is the retrospective design. Nonetheless, the potential disadvantages of this retrospective analysis are diminished by the fact that the patient data are inputted into an electronic database from report forms filled out by the attending physician upon the patient’s admission to our center. Other limitations include a lack of iron status and other biomarkers such as high sensitivity troponins.

Strengths of this study include large cohort, detailed data on clinical, echocardiographic hemodynamic and laboratory parameters, and long follow-up period with very little patients lost to follow-up.