Prognostic Value of Red Blood Cell Distribution Width in Non-Cardiovascular Critically or Acutely Patients: A Systematic Review

Background RDW (red cell distribution width) has been reported to been associated with the prognosis of patients with cardiovascular diseases. However, RDW is often overlooked by clinicians in treating patients with non-cardiovascular diseases, especially in an emergency. The objective of this systematic review is to explore the prognostic value of RDW in non-cardiovascular emergencies. Methods PubMed, EMBASE, and the Cochrane Central Register of Controlled Trials were systematically searched from their inception to December 31, 2015. We included studies examining the relationship between RDW and mortality rate by adjusting important covariables in non-cardiovascular emergencies. All included studies were divided into three groups. Group A: general critically ill patients; Group B: patients with infectious disease; Group C: other conditions. We extracted each study’ characteristics, outcomes, covariables, and other items independently. Results A total of 32 studies were eligible for inclusion in our meta-analysis. Six studies belonged to Group A, 9 studies belonged to Group B and 17 studies belonged to Group C. Among these included studies, RDW was assessed as a continuous variable (per 1% increase) in 16 studies, as a binary variable in 8 studies, and as a categorical variable in 8 studies. In addition, AUCs (area under the receiver operating characteristic curve) of RDW for predicting mortality were reported in 25 studies. All studies were published between 2011–2015. The qualities of included 32 studies were moderate or high. Conclusion The present systematic review indicates that the increased RDW is significantly associated with a higher mortality rate in an non-cardiovascular emergency. The low cost and readily accessible of this laboratory variable may strengthen its usefulness in daily practice in the future.


Introduction
Red blood cell distribution width (RDW) is a measure of erythrocyte size variability and calculated as the (standard deviation) SD in red blood cell (RBC) size divided by the mean corpuscular volume. RBC differ in size, whereas, this difference would get smaller during ageing [1]. In addition, any disorders result in the release of immature erythrocyte or shortening the lifespan of RBC would cause the change of RDW. RDW has traditionally been used for the diagnosis of different type of anemia [2]. In recent years, considerable attention were paid to the prognostic value of RDW [3][4][5][6]. In 2007, Michael Felker and his colleagues reported that RDW was a strong independent predictor of morbidity and mortality in chronic heart failure patients [6]. Subsequently, many other scholars found the similar association between RDW and various clinical conditions, including cardiovascular diseases, community-dwelling older adults and general in-hospital patients [3][4][5][6][7][8].

Materials and Methods
This systematical review was performed according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA, S1 Checklist) statement which was published in 2009 [34].

Literature search and inclusion criteria
PubMed, EMBASE, and the Cochrane library were systematically searched from their inception to December 31, 2015. As RDW is not referenced by the Medical Subject Headings, it was used as a keyword to identify relevant studies only. The bibliographies of relevant reviews or meta-analysis were also screened to identify potential eligible studies.
The inclusion criteria: patients with a diagnosis of non-cardiovascular disease were included and those who were diagnosed with cardiovascular diseases, such as heart failure, myocardial infarction and so on were excluded. In addition, patients with malignant tumor were also excluded; Effect sizes [odds ratios (ORs) or hazard ratios (HRs) or AUC and their 95% confidence intervals (CIs)] were available; Randomized controlled study or observational study; The primary outcome was all-cause mortality.

Data extraction and quality assessment
Data extraction was performed independently by two authors. The following data were extracted using a standard form: characteristics of each study (publication year, the first author, study design, the primary endpoint and the type of population), characteristics of all included patients (the mean age, male/female, and number of included patients), unadjusted and adjusted size effects (ORs or HRs or AUCs and their CIs) and important confounders (APACHEⅡ, age, hematocrit, hemoglobin, mean corpuscular volume, mean corpuscular hemoglobin, mean corpuscular hemoglobin concentration, C-reactive protein, sepsis, mechanical ventilation, admission type, leukocyte count and so on). Newcastle-Ottawa Scale was used to assess the methodological quality of included observational studies (http://www. ohri.ca/programs/clinical_epidemiology/oxford.asp). This scale consists of three domains: Selection, Comparability, and Outcome. Selection is evaluated through four items: representativeness of the exposed cohort, selection of the non-exposed cohort, ascertainment of exposure to implants and demonstration that outcome of interest was not present at start of study. Comparability is evaluated through one item: study controls for confounders. Outcome is evaluated through three items: assessment of outcome, duration of follow up and completeness of follow up. A study can be awarded with a maximum of one star for each numbered item within selection and outcome categories. A maximum of two stars can be given, as for comparability.

Data analysis
RDW was reported in different forms [dichotomous variable (i.e. normal vs. abnormal); continuous variable (i.e. per 1% increase); and ordinal categorical variable (i.e. tertiles, quartiles)]. The AUC of RDW for mortality prediction was also regarded as an additional effect size. All included studies were divided into three groups: Group A: general critically ill patients; Group B: patients with infectious disease; Group C: other conditions. If applicable, the extracted effect sizes were used for quantitative analysis. ORs and HRs could be combined because of similar magnitude [35]. The method of "generic inverse variance" was applied in this model (http:// www.cochrane.org/handbook). The data required for the generic inverse variance method are an estimate for the relative effect and its standard error (SE) for each of the studies. Each study is given a weight which is equal to the inverse of the variance of the effect estimate (i.e. one divided by the standard error squared). Then, we need to enter the napierian logarithm (ln) of the effect size and the standard error of the ln (effect size). If these two values are entered as the effect estimate and standard error, from them the software will calculate the effect size and 95%CI (this was not on the log scale). In our study, we needed to use the 95% CI to work backwards and calculated the SE of the ln (effect size). Heterogeneity between studies was assessed by chi-squared test and I 2 statistic, and P<0.1 or I 2 >50% indicated the presence of significant heterogeneity. Due to the wide clinical and methodological variability between studies, we chose random effects model to pool the results. Publication biases were assessed using Egger's tests. All statistical analysis was performed on STATA 12.0 (SERIAL NO.40120519635), and RevMan 5.2.10 (http://tech.cochrane.org/revman/download). derived from the same study periods [2,12] and the risk of data duplication was inevitable, thus the study by Purtle et al was excluded [12]. A similar situation also occurred in the studies by Braun et al [26,28], and his early study was excluded [26]. There was no additional studies were obtained through reviewing the bibliographies of relevant reviews or systematic reviews and there was no discrepancy between two reviewers.
The adjusted OR were in the range of 1.04 to 1.55. And 4 studies provided the AUCs of RDW for mortality prediction and its value from 0.62 to 0.73 [9][10][11]38]. Only one study [2] assessed RDW as a categorical variable and along with the increase of RDW, the mortality rate of critically ill patients would greatly increase.

Group B: Patients with infectious diseases
RDW was reported as a continuous variable in 4 studies [23,29,31,32]. The adjusted OR rose from1.12 to 1.40. RDW was reported as a binary variable in 2 studies [28,33], and we found that the increased RDW levels were associated with a higher risk of death (OR, 2.12; 95%CI, 1.82-2.46) with low heterogeneity (I 2 = 0, P = 0.49). And 4 studies provided data on AUC from 0.63 to 0.76 of RDW in prediction mortality [23,25,31,33]. Finally, three studies reported RDW as a categorical variable [25,27,29], and a rank correlation was found between the RDW levels and mortality in these patients.

Group C: Other conditions
In this group, there was one study was obtained respectively focusing on out-of-hospital cardiac arrest victims, patients with AKI who were treated with CRRT, patients with acute dyspnea who visited the emergency department, patients after organophosphate insecticide poisoning, prevalent kidney transplant recipients, and patients in the adult internal medicine ward. In addition, 3 studies focusing on patients with acute pancreatitis, 3 studies focusing on adult trauma patients, 2 studies focusing on patients with acute mesenteric ischemia, and 3 studies focusing on patients with acute pulmonary embolism were included. Nine studies reported RDW as a continuous variable [16,17,21,22,36,37,40,43], and the adjusted OR rose from 1.06 to 1.60. Six studies reported RDW as a binary variable [14,16,18,19,37,40] and the adjusted OR rose from 1.21 to 15.47. Ten studies provided data on AUC [14-20, 37-39, 41], and its value between 0.61 and 0.82. Finally, four studies reported RDW as a categorical variable [13,17,22,42]. We also found the RDW levels were significantly associated with morality with a rank correlation.

Publication bias
The quantity of included studies in each subgroup was insufficient to evaluate publication bias. However, as the populations in our study were determined by two senior doctors, and several studies were excluded due to insufficient data available, therefore, we could not rule out the possibility of the existence of publication bias, although publication bias was considered to be absent by Egger's test (P>0.05).

Discussion
This systematic review attempted to quantify the association between RDW and mortality in non-cardiovascular critically or acutely ill conditions. The results indicated that in each group (general critically ill patients, patients with inflammation and other critically or acutely ill patients), no matter as a continuous variable or binary variable or categorical variable, higher RDW levels were associated with a higher mortality rate. To our best knowledge, this systematic review firstly explores the association between RDW and mortality in non-cardiovascular conditions.
Early diagnosis and timely risk stratification are particularly important to critically or acutely ill patients. It is helpful in allocation of limited ICU (intensive care unit) resources (especially in developing countries, such as China) [44], judging the severity of critical illnesses, deciding curative effects, comparing performance between different centers, and clinical decision-making [9,10]. Physicians in the emergency department also need a simple and accurate tool to identify the patients who will benefit from hospitalization, especially those who should be admitted to ICU. In addition, an accurate prognostic model may be used as a quality improve tool through the construct of observed/ expected ratio. In recent years, a great deal of studies have attempted to find new prognostic factors [45,46], however, due to various reasons (not readily available, expensive etc.), no predictor was widely acceptable by clinicians.
RDW is a coefficient of variation of circulating red blood cells (RBC) and is a part of the complete cell count (CBC) panel. Increased RDW reflects greater heterogeneity in red cell volume. Disorders related to ineffective RBC production (e.g. iron deficiency anemia, vitaminB12 and folic acid deficiency, bone marrow suppression and hemoglobinopathies) or increased RBC destruction (hemolysis) or blood transfusion will cause higher RDW [2]. In addition, the change of RDW is affected by many factors, such as liver or renal dysfunction, malnutrition, cancer, thyroid disease, acute or chronic inflammatory response, use of some medications, renin-angiotensin system activation, and ethnicity [1,2,11,47,48]. Until now, RDW is mainly used for the differential diagnosis of anemia (especially iron-deficiency anemia). If the possibility of anemia has been ruled out, the RDW is often completely overlooked by clinicians. However, Felker et al. in 2007 and subsequent authors have demonstrated RDW is one of the strongest independent predictors of morbidity and mortality in various cardiovascular conditions including coronary heart disease, pulmonary hypertension, acute heart failure, peripheral artery disease, stroke, or pulmonary embolism and is independent of hemoglobin level [3][4][5][6][7][8].
In addition, recent studies also found that RDW is also associated with mortality in non-cardiovascular conditions, especially in critically or acutely patients . And RDW is not only associated with short-and long-term mortality [2], but also is associated with the incidence of Mucsi et al. [36] bloodstream infection [2], the length of hospital stay [9,30], the incidence of recurrent hospital admissions [33], and the use of vasopressors [25]. Although the association between RDW and mortality was reported repeatedly, the exact mechanism underling the RDW-mortality association remains unclear. However, several reasons were proposed in previous reports. Firstly, bone marrow function and iron metabolism are always influenced by systemic inflammation response [49,50]. Meanwhile, the erythropoietin-induced erythrocyte maturation and proliferation are restrained by pro-inflammatory cytokines [51]. Additionally, pro-inflammatory cytokines usually down-regulate the expression of erythropoietin receptor which is important in the process of erythrocyte maturation [51]. This will result in the release of larger, immature reticulocytes into the circulation. It has been reported that RDW is significantly influenced by inflammatory biomarker like C-reactive protein, erythrocyte sedimentation rate (ESR), white blood count (WBC), and IL-6 [16,23,52,53]. Another explanation for the RDW-mortality relationship may be oxidative stress. Lots of reactive oxygen species generated by activated leukocytes are released in the process of acutely or critically illness. And the change of RDW could be mediated by oxidative stress through increasing the fragility of RBCs, shortening the life-span of RBCs [54]. In addition, malnutrition may be another contributing factor. Recent studies have shown that the level of total cholesterol and albumin which are markers of nutrition are significantly associated with RDW [16,30]. Other possible factors also include renal or hepatic dysfunction, chronic hyperglycemia and so on [55]. Based on the above analysis, it is reasonable to believe that RDW is an integrative marker of multiple harmful pathologic processes. However, these factors may not completely account for the relationship between RDW and mortality, because RDW is independently associated with mortality even after adjustment for total cholesterol, albumin, and SOFA score, anemia, transfusion and so on [9-11, 16, 22, 25, 29]. As for anemia, the relationship between RDW and mortality was reported to be more significant in non-anemic patients, compared with anemic patients [56]. Therefore, further studies are needed to address the association between RDW and adverse outcomes.
The clinical value of RDW remains a question. Although, the adding of RDW to established prediction models such as APACHEⅡ, SOFA, SAPS significantly improves the discriminative power [1,11,25], the ability of RDW in distinguishing survivors from non-survivors is suboptimal, the pooled AUC in our study ranged from 0.68 to 0.69. Secondly, Farid and his colleagues have reported that the discriminative power of RDW is better than both APACHEⅡscore and SOFA score in patients with severe sepsis and septic shock. The low cost and readily available of RDW may strengthen its value in daily practice in the future. However, the optimal combination of RDW and other parameters or predictive model remains unclear [47,48]. Another inconclusive issue is the value of the change of RDW over time. Chan and his colleagues reported that the increase of the level of RDW from the baseline during the first 72h after hospitalization was also associated with a higher mortality rate in patients with severe sepsis and septic shock [30]. However, this relationship was not observed in another two studies [10,31].

Limitations
Our study was mainly limited by the heterogeneity of the included studies. The design of study, categorization of RDW and endpoints were different between studies. And all included studies were non-random, and the inherently bias was inevitable. The change of RDW was affected by many factors, despite multiple important factors affecting RDW levels were adjusted in most studies, there was the possibility that other residual confounding factors not included in the analysis. For example, in most studies, the authors did not adjust for the levels of vitamin B12 and folic acid which may significantly affect RDW levels. However, Patel et al reported a similar strong RDW-mortality association even after adjustment for these covariables [7]. The elapsed time between the blood sampling and RDW measuring was not defined in almost all studies, this time interval may significantly alter RDW levels [57]. In addition, the blood analyzer and reagents used for measurement are different between studies. Study populations varied widely. Therefore, we do not have enough studies to conduct subgroup analysis to explore the association between RDW and mortality in each kind of disease. However, the results presented in this review are a summary of the best evidence currently available in noncardiovascular critically or acutely ill patients.
Implications for research 1. The exact mechanism underlying this relationship; 2. Future research should focus on subgroups of patients; 3. The value of the combination of RDW and other indexes or predictive models; 4. The value of the dynamic change of RDW levels.

Conclusion
The increased RDW is significantly associated with a higher mortality rate and is helpful for risk stratification in non-cardiovascular critically or acutely ill patients.