Introduction
Red cell distribution width (RDW) is a measure of erythrocyte size variability and has been shown to be a prognostic marker for mortality, mainly in patients with cardiovascular disease and in community-dwelling patients, as well as in general in-hospital patients [
1‐
16]. Although the mechanisms linking RDW to adverse patient outcomes remain incompletely understood, potential pathways include chronic inflammation [
17,
18], malnutrition [
9‐
11], and anemia of different etiologies [
19,
20], among others. The prognostic potential of RDW is of particular interest because it is routinely included in the automated complete blood count (CBC) analyses in hospitalized patients and thus available at no additional cost for clinicians. Recent studies have found RDW to be a prognostic marker for short- and long-term mortality in critically ill patients [
21‐
23]. The first study in critical illness was conducted in a cohort of 602 patients in China and found that RDW is associated with ICU mortality [
22]. Recently, a large 10-year retrospective study from two US centers validated these findings and found RDW to be a robust predictor of the risk of all-cause patient mortality and bloodstream infection in the critically ill [
21]. Finally, one report found RDW to be a strong outcome predictor in patients with pneumonia [
23]. However, from these studies, it remains unclear whether RDW may improve state-of-the-art risk prediction in unselected critically ill patients.
We therefore aimed to investigate whether adding RDW has the potential to improve the prognostic performance of the Simplified Acute Physiology Score (SAPS) to predict short- and long-term mortality in an independent, large, and unselected population of ICU patients.
Discussion
In this large cohort of ICU patients, RDW strongly and independently predicted hospital mortality, ICU mortality, and 1-year mortality. As evidenced by improvement in both AUC and net reclassification index, RDW materially improved the prediction of mortality over and above traditional severity of illness, as measured with SAPS. This finding was robust across subgroups (for example, different age and hematocrit categories and comorbidities such as anemia or renal failure, factors that are known to be positively associated with RDW values [
12,
19,
20]).
These results are in accordance with previous research showing an association of RDW and long-term mortality risks in different patient populations, such as patients with cardiovascular disease, chronic heart failure, and acute kidney injury treated with continuous renal-replacement therapy [
2‐
16,
30]. Cohort studies using nationally representative samples of the US population (Third National Health and Nutrition Survey (NHANES III)) also reported an association of RDW with long-term outcomes in community-dwelling patients [
9‐
11]. In addition, recent studies have focused on critically ill patients and found RDW to be a prognostic in this setting as well [
21‐
23]. Importantly, our results based on a broad US patient population admitted to different types of ICUs, including medical, cardiac, surgical, and cardiothoracic surgery confirm and expand these previous findings and demonstrate that RDW is a strong outcome predictor and thereby improves risk prediction based on the well-established SAPS score. We found the association of RDW and outcome to be robust when focusing on hospital and 1-year mortality overall and in different subgroups, such as patients with anemia.
Great interest is expressed in finding new prognostic markers in critically ill patients in the ICU setting. Early prognostication potentially helps to improve triage decisions in terms of early discharge of patients, and also with regard to therapeutic interventions. A prognostic marker should be evaluated based on the information it provides, beyond information already available based on clinical information, such as the information incorporated in SAPS. The SAPS system was developed to measure severity of illness in ICU patients, and subsequent research has validated its use to predict outcome in ICU patients [
31]. Within our large sample, SAPS showed reasonable discrimination with AUCs between 0.75 and 0.80. Still, adding RDW to the SAPS further improved its discriminatory ability for both hospital mortality and ICU mortality. This was further confirmed when calculating reclassification statistics. For in-hospital mortality, this was mainly due to better classification of nonsurvivors into lower-risk classes, whereas for ICU mortality, RDW improved classification of survivors and nonsurvivors. Importantly, from a cost perspective, it should be noted that RDW is routinely measured in most patients as part of the admission CBC analysis and thus is free of additional costs. It will important for future studies to assess whether more-advanced severity of illness scores (that is, APACHE IV, SAPS III) can still be improved by the addition of RDW.
Our results provide an important target for future translational research. Why is RDW associated with increased mortality? The underlying mechanism is still unclear and represents an important avenue for future "bedside-to-bench" research. RDW can be increased in anemia or hemoglobinopathies, hemolysis, or after blood transfusions [
19,
20]. Yet, in our study and in the literature, the association of RDW and mortality was independent of anemia, and no evidence was found for effect modification by anemia and hematocrit levels [
9‐
11,
13]. Another possible pathophysiologic explanation is that RDW is a surrogate of inflammation, which is known to increase RDW. Several studies found RDW to be associated with blood markers of inflammation, such as interleukin-6, C-reactive protein (CRP) [
11], as well as impaired iron mobilization [
17]. Also, oxidative stress has been shown to increase anisocytosis by disrupting erythropoiesis, and to alter blood cell membrane deformability and red blood cell circulation half-life, ultimately leading to increased RDW [
9,
11,
19]. Again, previous research found that adjusting for inflammation (for example, CRP) did not substantially reduce the prognostic value of RDW, demonstrating its effect beyond these factors [
9‐
11,
22]. In our cohort, we were not able to adjust the analysis for inflammatory markers, as these are not routinely measured on admission.
RDW may also reflect the patient's degree of physiological reserve, one of three main determinants of clinical outcome, as suggested by Bion [
32]. The physiological reserve represents cellular response to acute stress and the resultant tissue hypoxia. Ischemia activates cellular systems that would reduce oxygen demand and physiologic processes that would improve tissue oxygen delivery, such as increased production and release of mature red cells into the peripheral bloodstream. How well this process of reactive erythropoiesis is carried out under oxidative stress may represent the patient's ability to handle acute physiological insult. Release of large immature red cells with poor oxygen-binding capacity, which results in an increased RDW, implies suboptimal response to oxidative stress. This may explain why the association between RDW and clinical outcome is independent of the severity of acute illness as well as the degree of inflammation. It may well represent the genetic factors that determine how well the body withstands physiological stress, whether from a severe infection, multitrauma, burns, or acute pancreatitis.
Strengths of this study are the large number of patients across different types of ICUs and the availability of a rich annotation of the physiological and clinical context, which allowed rigorous adjustment for severity of acute illness and subgroup analyses with high statistical power. We also included 1-year survival data on all the patients.
This study has limitations; as it was a retrospective, observational study, we were not able to assess causes for mortality and/or for elevated RDW. Also, our data provide no additional information about inflammation and oxidative stress. These pieces of information would be necessary to understand better the pathophysiology underlying the observed effects. Also, only admission RDW levels were considered, and it remains unclear whether changes over time of RDW may provide additional prognostic information. Importantly, we used SAPS, and more recent severity-of-illness scores (such as SAPS III or APACHE IV) have shown superior accuracy for outcome prediction; thus, future studies should address whether RDW has the potential also to improve these severity-of-illness scores. We calculated multivariable regression models, including important demographics and different comorbidities, to investigate the independent association of RDW and outcome; yet, co-linearity and missing clinical information on some important patient-history items (that is, transfusion) may limit the interpretation of the models.
Competing interests
The authors declare that they have no competing interests.
Authors' contributions
SH, LAC, JL, and MH conceived and designed the study and wrote the study protocol. JL and LAC extracted data. The statistical analyses and the first draft of manuscript were performed by SH. All authors amended and commented on the manuscript and approved the final version.