PDF
Citation
Print
INTRODUCTION
According to the American Heart Association, millions of patients undergo cardiac surgery yearly [1]. The mortality rate has declined in recent decades owing to advances in surgical techniques and perioperative management. However, even the lowest-risk procedures (such as isolated coronary artery bypass grafts or isolated aortic valve replacements) still have a mortality rate of 2% [2]. Additionally, higher mortality rates are associated with more complex or combined procedures and patients with high risks [3, 4]. Consequently, increasing emphasis has been placed on assessing mortality risks for cardiac surgery.
However, even widely used risk models, such as the European System for Cardiac Operative Risk Evaluation (EuroSCORE) [5], have their limitations. The EuroSCORE is widely used to predict mortality after cardiac surgery, but its predictive power and reliability are decreasing due to changes in cardiac surgery and outcomes. Among patients who underwent degenerative mitral valve repairs, the EuroSCORE II showed low discrimination, accuracy, and calibration scores because it overpredicted the 30-day mortality [6]. A new risk factor that enables simple and cost-effective clinical use, with improved methods for predicting and preventing mortality following cardiac surgery, should be proposed.
The red blood cell distribution width (RDW), an easily accessible laboratory parameter, is expressed as the ratio between the standard deviation (SD) of the red blood cell (RBC) volume and the mean corpuscular volume (MCV). The RDW can reflect heterogeneity in the RBC size distribution, and an increase in the RDW may imply the presence of various abnormalities, namely, telomere-length shortening, oxidative stress, inflammation, RBC fragmentation, a poor nutritional status, hypertension, dyslipidemia, and abnormal erythropoietin function [7, 8].
Generally, the RDW is primarily used for the differential diagnosis of anemia. According to MCV values, anemia can be classified as microcytic, normocytic, and macrocytic anemia. Combining MCV with RDW values can enable further sub-classification [9]. Recent findings showed that RDW could be used to predict the prognosis for various diseases, such as heart failure [10], acute myocardial infarction (MI) [11], and sepsis [12]. The RDW can also be used to predict the mortality of patients after coronary artery bypass graft surgery [13] and percutaneous coronary intervention [14]. However, those studies were limited to coronary artery disease, and whether RDW has similar predictive capacities for other cardiac procedures remains unknown. To our knowledge, this study was the first to assess the prognostic ability of RDW for all populations who underwent cardiac surgery. We aimed to investigate the utility of RDW as an early mortality predictor for cardiac surgery using a public database with a large sample size.
MATERIALS AND METHODS
Data source
The Medical Information Mart for Intensive Care-IV (MIMIC-IV) Database (version 2.0) [15] of the Massachusetts Institute of Technology (MIT), Cambridge, MA, USA, was used to enroll patients undergoing cardiac surgery. From 2008 to 2019, Beth Israel Deaconess Medical Center’s intensive care unit (ICU) treated more than 50,000 adult patients. One author of this study (Weiqiang Chen) obtained full access to the database and extracted the data (certification number 41021697). Our study protocol followed the guidelines for observational studies [16]. As the data were obtained from a publicly available source, the requirement for informed consent was waived.
Patients (age >18 yrs) undergoing cardiac surgery based on International Classification of Disease (ICD) codes (Supplemental Data Table S1) from the MIMIC-IV Database (version 2.0) were included. Patients without RDW records during hospitalization were excluded. We used the date of the first ICU admission for patients admitted to an ICU at least once [17].
Variates
According to existing literature and clinical judgment, we incorporated the following ICU variables, which are considered confounders of cardiac surgery outcomes.
Demographic information: age, sex, and body–mass index (BMI).
Comorbidities: MI, congestive heart failure (CHF), cerebrovascular disease (CVD), chronic pulmonary disease (CPD), diabetes mellitus (DM), renal disease, and liver disease. The Charlson comorbidity index (CCI) and severity at admission were measured using the Glasgow Coma Scale (GCS), Simplified Acute Physiology Score II (SAPS II), and Sequential Organ Failure Assessment (SOFA) score.
Vital signs: heart rate, mean arterial pressure (MAP), respiratory rate (RR), and peripheral oxygen saturation (SpO2) at the time of ICU admission.
Laboratory results: hemoglobin (Hb), white blood cell (WBC) count, platelet count, blood urea nitrogen (BUN), creatinine, glucose, potassium, lactate, partial pressure of oxygen (PaO2): fraction of inspired oxygen (FiO2) ratio, and base excess (BE).
Prognostic information: acute kidney injury (AKI), sepsis, the lengths of stay in a hospital and ICU.
Outcomes
The primary outcome was the 30-day mortality, measured from the day of admission to an ICU. Secondary outcomes were the lengths of stay in a hospital and ICU.
Statistical analysis
We found that <30% of covariates had missing data. We used multiple imputations to impute the missing values for the covariates. The Kolmogorov–Smirnov test was applied to test for normality in the quantitative data. Normally distributed, continuous data were presented as the mean (SD) and compared using Student’s t-test. Non-normally distributed data were presented as the median (interquartile range [IQR]) and compared using the Wilcoxon rank-sum test. Chi-square tests were used to compare categorical variables, and differences were presented as percentages (%). Kaplan–Meier survival curves were plotted to evaluate the 30-day survival probabilities. To determine whether the elevated-RDW group was significantly different from the normal-RDW group, we applied the log-rank test (RDW <14.5 and ≥14.5) [18].
Additionally, to determine whether the RDW was independently associated with the 30-day mortality rate, we constructed five different multivariate Cox proportional hazard models. Model 1 was only adjusted for age, sex, and BMI. Model 2 was adjusted for additional comorbidities, such as MI, CHF, CVD, CPD, DM, and renal disease. Model 3 was additionally adjusted for the CCI, SAPS II, and SOFA scores. Model 4 was additionally adjusted for the heart rate, MAP, RR, and SpO2. Model 5 was additionally adjusted for Hb, WBC, and platelet counts; BUN, creatinine, glucose, and potassium lactate levels; the PaO2: FiO2 ratio; and BE.
We conducted a sensitivity analysis to ensure that our analysis was robust. To identify modifications and interactions, subgroup analyses were applied based on sex (male or female), age (<70 or ≥70 yrs), sepsis (yes or no), AKI (yes or no), CHF (yes or no), CVD (yes or no), CPD (yes or no), DM (yes or no), or renal disease (yes or no). To reduce publication bias, we also conducted sensitivity analysis after excluding patients with missing data. The restricted cubic spline (RCS) method was used to explore the relationship between the RDW and 30-day mortality of patients after undergoing cardiac surgery. A receiver operating characteristic (ROC) curve was plotted for RDW to assess the ability to predict the 30-day mortality. We determined the area under the curve (AUC) to test whether the predictive power of existing systems could be improved by incorporating the RDW. We also tested the robustness of the results by using RDW as a categorical variable. The lengths of hospital and ICU stays were also analyzed using multivariate linear regression. We used Free Statistics software (version 1.7; Beijing Free Clinical Medical Technology Co.) and R software (version 4.1.2; R Foundation for Statistical Computing, Vienna, Austria; http://www.r-project.org) in the RStudio environment for statistical analysis. Differences were considered statistically significant when P was <0.05.
RESULTS
Basic characteristics of the study population
In total, 11,634 patients were identified based on ICD codes recorded in the MIMIC-IV Database (version 2.0). Of these, 132 patients without RDW values or who died on the first day in the ICU were excluded, leaving 11,502 patients who underwent cardiac surgery in the final cohort (Fig. 1).
According to Table 1, all patients enrolled had the following basic demographic characteristics. Approximately 67.7% of the patients were men. The average age of all patients was 68.8 (12.6). The 30-day mortality rate was 4.2% (485/11,502), and the lengths of their hospital and ICU stays were 6.8 (4.8, 10.0) and 1.9 (1.2, 3.2) days, respectively. Based on these criteria [15], the patients were divided into a normal-RDW group and an elevated-RDW group. The elevated-RDW group had a higher age; higher heart and respiratory rates; an elevated platelet count; higher BUN, creatinine, glucose, and lactate levels; and higher CCI, SAPS II, and SOFA scores. Patients in the normal-RDW group had higher Hb, MAP, PaO2: FiO2, and GCS values; were more likely to have MI, CHF, CVD, CPD, diabetes, AKI, and sepsis; and a lower mortality rate than those in the elevated-RDW group (Table 1, P<0.001).
Table 1
Baseline characteristics of the study participants and their outcome parameters
Data were presented as mean±SD or median (first quartile, third quartile) based on their distribution.
Abbreviations: RDW, red blood cell distribution width; BMI, body–mass index; MI, myocardial infarction; CHF, congestive heart failure; CVD, cerebrovascular disease; CPD, chronic pulmonary disease; DM, diabetes mellitus (with complications); CCI, Charlson comorbidity index; SOFA, sequential organ failure assessment; SAPS II, Simplified Acute Physiology Score II; HR, heart rate; bpm, beats/min; MAP, mean artery pressure; RR, respiratory rate; SpO2, peripheral oxygen saturation; Hb, hemoglobin; PLT, platelet; WBC, white blood cell; BUN, blood urea nitrogen; CR, creatinine; GLU, glucose; K, potassium; LAC, lactate; P/F, partial pressure of oxygen: fraction of inspired oxygen ratio; BE, base excess; GCS, Glasgow Coma Scale; AKI, acute kidney injury; Los, length of stay; ICU, Intensive care unit.
Association between RDW and 30-day mortality
The Kaplan–Meier survival curves (Fig. 2) show that the elevated-RDW group had a lower 30-day survival probability than that in the normal-RDW group (P<0.001), suggesting that the elevated-RDW group had a higher 30-day mortality rate. Table 2 shows hazard ratios (HRs) and 95% confidence intervals (CIs). In all five models, the HRs were robust between the unadjusted and adjusted models (P<0.05). The unadjusted model showed a 19% increase in the 30-day mortality for every unit difference of RDW (HR=1.19, 95% CI, 1.17–1.21). Increasing the RDW by one unit increased the 30-day mortality difference by 17% in the minimally adjusted model (Model 1; HR=1.17, 95% CI, 1.16–1.19). With Model 5, as the RDW increased by one unit, the 30-day mortality increased by 12% (HR=1.12, 95% CI, 1.10–1.14). Overall, the statistical results showed that the RDW was independently associated with the 30-day mortality rate and that this association was robust with all models tested.
Fig. 2
Kaplan–Meier curves showing the association between the red blood cell distribution width (RDW) and the 30-day mortality of patients after undergoing cardiac surgery.
Table 2
Multivariate Cox regression to assess the association between the RDW and 30-day mortality
| Variables | Normal RDW | Elevated RDW | RDW | ||||
|---|---|---|---|---|---|---|---|
| HR | HR (95% CI) | P | HR (95% CI) | P | |||
| Unadjusted | 1 | 2.32 (2.14–2.52) | <0.001 | 1.19 (1.17–1.21) | <0.001 | ||
| Model 1* | 1 | 2.20 (2.02–2.39) | <0.001 | 1.17 (1.16–1.19) | <0.001 | ||
| Model 2† | 1 | 2.00 (1.83–2.18) | <0.001 | 1.16 (1.14–1.18) | <0.001 | ||
| Model 3‡ | 1 | 1.83 (1.68–2.00) | <0.001 | 1.13 (1.11–1.15) | <0.001 | ||
| Model 4§ | 1 | 1.84 (1.69–2.01) | <0.001 | 1.13 (1.11–1.15) | <0.001 | ||
| Model 5ІІ | 1 | 1.79 (1.64–1.96) | <0.001 | 1.12 (1.10–1.14) | <0.001 | ||
*Models 1–5 were derived from multivariate Cox regression models. Model 1 covariates were adjusted for sex, age, and body–mass index (BMI).
†Model 2 covariates were adjusted for sex, age, BMI, myocardial infarction, congestive heart failure, cerebrovascular disease, chronic pulmonary disease, diabetes with complications, and renal disease.
‡Model 3 covariates were adjusted for sex, age, BMI, myocardial infarction, congestive heart failure, cerebrovascular disease, chronic pulmonary disease, diabetes with complications, renal disease, Charlson comorbidity index, sequential organ failure assessment (SOFA) score, and Simplified Acute Physiology Score (SAPS) II.
§Model 4 covariates were adjusted for sex, age, BMI, myocardial infarction, congestive heart failure, cerebrovascular disease, chronic pulmonary disease, diabetes with complications, renal disease, Charlson comorbidity index, SOFA score, SAPS II, heart rate, mean arterial pressure (MAP), respiratory rate, and peripheral oxygen saturation (SpO2).
ІІModel 5 covariates were adjusted for sex, age, BMI, myocardial infarction, congestive heart failure, cerebrovascular disease, chronic pulmonary disease, diabetes with complications, renal disease, Charlson comorbidity index, SOFA score, SAPS II, heart rate, MAP, respiratory rate, SpO2, hemoglobin, platelets, white blood cells, blood urea nitrogen, creatinine, glucose, potassium, lactate, partial pressure of oxygen: fraction of inspired oxygen ratio, Glasgow Coma Scale, and base excess.
Further sensitivity analysis was conducted by dividing the continuous variable, RDW, into two subgroups and comparing the results against those of the normal-RDW group as a reference baseline. The 30-day mortality rate was higher in the elevated-RDW group than in the normal-RDW group (HR=1.79, 95% CI, 1.64–1.96). Missing covariate values were imputed with multiple imputation. Supplemental Data Table S2 shows that the RDW was independently associated with 30-day mortality.
A Cox regression model (adjusted for all confounders mentioned above and the RDW) was developed to predict the 30-day mortality. The survival probability was calculated using the regression model based on the grouping criteria (RDW <14.5 or ≥14.5). A significant difference in survival probabilities was observed between both groups (P<0.001, Supplemental Data Table S3).
Subgroup analysis of RDW and other factors
Subgroup analyses were performed to examine whether the association between the RDW and the 30-day mortality was stable among the distinct subgroups (Fig. 3). No significant interaction was identified between the RDW and sex, sepsis, AKI, CHF, CVD, CPD, DM, renal disease, sepsis, or the SOFA score (interaction P all >0.05). However, a significant interaction was observed between RDW and age (P<0.05).
Fig. 3
Forest plot showing subgroup analysis of the association between red blood cell distribution width (RDW) and 30-day mortality.
Abbreviations: HR, hazard ratio; CI, confidence interval; Sepsis 3.0, the third international consensus definitions for sepsis and septic shock; SOFA, sequential organ failure assessment; AKI, acute kidney injury; MI, myocardial infarction; CHF, congestive heart failure; CVD, cerebrovascular disease; CPD, chronic pulmonary disease; DM, diabetes mellitus.
Association between the RDW and clinical outcomes
The RCS method revealed a linear correlation between the RDW and 30-day mortality after cardiac surgery (Supplemental Data Fig. S1). The ROC was calculated to further show the predictive power of the RDW (Supplemental Data Fig. S2). The results in Supplemental Data Fig. S3 suggest that the AUC increased significantly when the RDW was combined with the SOFA score or SAPS II. The AUC increased when the RDW was treated as a categorical variable (Supplemental Data Table S4).
Associations between the RDW and the lengths of hospital and ICU stays
When considering RDW as a continuous variable, each unit increase in the RDW correlated with an increased stay of 0.32 days in a hospital or an increase of 0.08 days in an ICU (Supplemental Data Table S5).
DISCUSSION
We examined the relationship between the RDW and the 30-day mortality rate by conducting a population-based analysis with adjusted variables. An elevated RDW was associated with a higher 30-day mortality. Additionally, the RDW was positively associated with 30-day mortality in patients after cardiac surgery in an ICU setting. The predictive power of existing scoring systems could be improved by incorporating the RDW. An elevated RDW was linked to a longer hospital stay and more days in an ICU.
Used for the differential diagnosis of anemia, an elevated RDW has been associated with ineffective RBC production (e.g., iron deficiency anemia, vitamin B12 and folic acid deficiency, bone marrow suppression, and hemoglobinopathies), increased RBC destruction (e.g., hemolysis), or blood transfusion. Recent reports showed that the RDW may be useful for predicting the prognoses of various diseases. For example, an elevated RDW correlated significantly with a poor prognosis for people with atrial fibrillation [19], severe heart failure, acute heart attacks, septic shock [20], spontaneous cerebral hemorrhaging [21], and hepatocellular carcinoma [22].
The underlying mechanism linking an elevated RDW with mortality remains unclear. One possible mechanism is related to inflammation [23]. Inflammation can impair bone marrow function, resulting in poor efficiency in terms of RBC production, and affect the RBC membrane permeability, which causes reticulocytes to enter the peripheral blood circulation, increases the proportion of immature RBCs, and increases the RDW [24]. Elevated RDW levels are associated with elevated inflammatory-cytokine levels, which usually result in organ dysfunction and complications [25, 26]. The RDW is linked to mortality because it increases oxidative stress, which reduces RBC survival and increases circulating premature RBCs, thereby leading to anisocytosis [27, 28]. Additionally, the RDW increases during malnutrition [29]. These factors may represent pathophysiological mechanisms that explain the association between RDW and mortality. RDW can be used to predict mortality.
The 30-day mortality rate following cardiac surgery was 4.2%, which was 2.1% higher than that reported previously [30]. The elevated rate may be attributed to the higher percentage of patients in this study undergoing cardiac surgery and hybrid operations with higher risks.
Many findings have shown that age; BMI; residual, lactate, creatinine, and urea nitrogen levels; and complications such as CHF, the SOFA score, and other risk factors can significantly impact mortality in patients after cardiac surgery [31-34]. These factors were well controlled preoperatively in this study. We performed multivariate Cox regression and sensitivity analyses to enhance the robustness and reliability of our results.
Our results revealed that abnormally elevated RDW levels are independent predictors of the 30-day mortality rate. Combining the RDW with risk-scoring tools, such as the SOFA score and SAPS II, significantly improved the diagnostic performance. Therefore, mortality in patients admitted to an ICU after cardiac surgery may be predicted based on their RDW values.
This is the first report describing an independent correlation between the RDW and 30-day mortality rate in patients undergoing cardiac surgery in an ICU. Our results could promote the development of diagnostic or predictive models for patients after cardiac surgery.
This study has several strengths. First, our analysis was based on real-world data collected from a diverse group of people. Second, strict statistical adjustments were utilized in this retrospective observational study. Potential residual confounders were minimized using multivariate Cox regression. Third, we considered the targeted independent variables separately as categorical and continuous variables. This approach reduced random errors in statistical analysis and enhanced the robustness of our conclusions. Finally, we evaluated the predictive value of the RDW and assessed whether incorporating it could improve the predictive power of existing scoring systems, such as the SOFA score and the SAPS II, by evaluating the AUC.
This study also has several limitations. First, this was a single-center study. Another study is needed to determine whether these findings apply to other hospitals or patients in other countries. Second, nutrition information may impact the predictive value of the RDW. We did not adjust for some unknown confounders, such as iron, folic acid, and vitamin B12 levels, and other nutritional information. In addition, the MIMIC-IV Database may have incomplete data for many patients, and other confounding factors (including inflammatory biomarkers, bilirubin levels, underlying disease, the severity of underlying disease, and surgical information such as the transfusion history, bleeding, operation duration, reoperation, and causes of infection) were not collected and analyzed; thus, the current results should be confirmed with a prospective cohort. Third, the predictive performance of the RDW was evaluated in this study without comparison with the EuroSCORE because of missing data in the MIMIC-IV Database. During this study, the baseline RDW was examined at the time of ICU admission, but subsequent changes to the RDW were not included in the dynamic analysis. As with all retrospective analyses, the lack of some necessary data resulted in the exclusion of some patients who underwent cardiac surgery from our study, which may have influenced the results. However, we applied multiple imputation to the entire cohort, followed by multivariate Cox regression, and the results were robust. This was a retrospective study of MIMIC-IV data, making it a post-hoc analysis. Evidences were insufficient to validate the association between the RDW and the 30-day mortality in ICU patients, and more high-quality prospective studies are needed in the future.
In conclusion, our findings indicate that an increase in the RDW is associated with higher 30-day mortality in ICU patients after cardiac surgery and that the RDW can be an efficient predictor. As a risk factor for a poor prognosis, the RDW is a simple and cost-effective laboratory test that can enable early postoperative screening. A high RDW value was associated with a poor prognosis, and the RDW could be an important criterion for early and aggressive intervention and better individualized patient management. Consequently, the RDW might be clinically significant and could become widely used in the future.
XML Download