Abstract
Background
Red cell distribution width (RDW), a component of an electronic complete blood count, is a measure of heterogeneity in the size of circulating erythrocytes. In patients with symptomatic cardiovascular disease (CVD), RDW is associated with mortality. However, it has not been demonstrated that RDW is a predictor of mortality independent of nutritional deficiencies or in the general population.
Methods
RDW was measured in a national sample of 8175 community-dwelling adults aged 45 and older who participated in the 1988–1994 National Health and Nutrition Examination Survey; mortality follow-up occurred through December 31, 2000. Deaths from all causes, CVD, cancer, and other causes were examined as a function of RDW.
Results
Higher RDW values were strongly associated with an increased risk of death. Compared to the lowest quintile of RDW, the following were adjusted hazard ratios (HR) for all-cause mortality (and 95 percent confidence intervals): second quintile, 1.1 (0.9–1.3); third quintile, 1.2 (1.0–1.4); fourth quintile, 1.4 (1.2–1.8); fifth quintile, 2.1 (1.7–2.6). For every 1 percent increment in RDW, all-cause mortality risk increased by 22% [HR = 1.22 (1.15–1.30); p<0.001]. Even when analyses were restricted to non-anemic participants or to those in the normal range of RDW (11–15%) without iron, folate, or vitamin B12 deficiency, RDW remained strongly associated with mortality. The prognostic effect of RDW was observed in both middle-aged and older adults for multiple causes of death.
Conclusions
RDW is a widely-available test that is a strong predictor of mortality in the general population of adults aged 45 and older.
INTRODUCTION
Red cell distribution width (RDW) is routinely assessed as part of the complete blood count (CBC) to gather information on the heterogeneity in the size of circulating erythrocytes. Computationally, RDW is the coefficient of variation of the mean corpuscular volume (MCV) and therefore higher RDW values reflect greater heterogeneity in MCV (anisocytosis), which is usually caused by perturbation in erythrocyte maturation or degradation. The RDW is used as an auxiliary index to help diagnose different types of anemia, but has also been evaluated as a potential screening marker for colon cancer and celiac disease because of its responsiveness to subtle nutrient deficiency.1, 2 Recently, researchers have reported higher mortality risk associated with higher RDW in patient populations with cardiovascular disease (CVD).3–5 However, none of these prospective studies were able to account for nutritional status or levels of inflammation.
Given that the RDW is routinely reported by clinical laboratories as a component of the CBC and it is available for most patients, understanding its prognostic significance could be very valuable for risk stratification in clinical decision making. However, whether the prognostic effect of RDW is specific to patients with CVD or rather is also valid in the general population is unknown. Therefore, the current study sought to determine whether higher RDW levels are associated with increased risk of death in a large, nationally representative sample of middle-aged and older adults.
MATERIALS AND METHODS
Study Population and Design
The Third National Health and Nutrition Examination Survey, 1988–94 (NHANES III) was designed to provide health status information on a nationally representative sample of the civilian noninstitutionalized US population. Adult participants provided written informed consent before entering the study and then were interviewed in their homes and subsequently underwent a physical examination, including phlebotomy, either in a mobile examination center or at home. Data were collected in two phases (1988–1991 and 1991–1994), each of which provided a nationally representative sample. Both phases of data collection included CBCs, but assays for serum nutrients were performed only in Phase 2 (1991–94). Details of the NHANES III study design and data collection protocols have been published previously.6
Of the 9,787 NHANES III participants aged 45 and older, 8,178 had their blood drawn. Participants missing RDW values were significantly older, more likely to be female than male, less likely to be Mexican American than non-Hispanic white, and more likely to die over the follow-up period compared with those not missing RDW values. Vital status through December 31, 2000 was ascertained for 8,175 participants using the National Death Index (NDI); 3 participants were excluded because insufficient information was available to perform an NDI search. Analyses restricted to Phase 2 were performed on 4,087 adults.
Complete Blood Count
The automated Coulter Counter Model S-Plus Jr (Coulter Electronics, Hialeah, FL) was used to measure RDW, MCV, and hemoglobin concentration. Quality control procedures and calibration were performed daily and verified directly via telephone with Coulter Electronics headquarters. The maximum acceptable coefficients of variation for RDW, MCV, and hemoglobin concentration were 3.2%, 2.0%, and 1.0%, respectively. Laboratory methods used in NHANES III have been described in detail in previous publications.7
Mortality outcomes
A total of 2,428 deaths were identified through the NDI search (median follow-up was 7.9 years; maximum was 12.1 years). Over the follow-up period, coding for the underlying causes of death switched in 1999 from the 9th revision of the International Classification of Diseases (ICD) to the 10th revision. For comparability, the National Center for Health Statistics recoded deaths into 113 groups using ICD-10.8 To examine cause-specific mortality, cardiovascular deaths were identified with ICD-10 codes I00-I78 (n=1,195), while codes C00-97 were used for cancer deaths (n=518). All other causes of deaths were grouped into a third category (n=713).
Other Measures
During the home interview, participants were asked to report their date of birth, race/ethnicity, and highest grade of education completed. Smoking status (never, former, or current) was determined through a standard set of questions. Measured height and weight were used to calculate body mass index (BMI = weight in kilograms/height in meters squared). For medical conditions, participants were asked if a doctor had ever told them that they had any of the following: cancer (non-skin related), congestive heart failure, diabetes, heart attack, pulmonary diseases (asthma, chronic bronchitis, or emphysema), and stroke. In addition, overnight hospitalization stays within the past year were assessed by questionnaire. To assess renal function, serum creatinine values were recalibrated to the Cleveland Clinic Research Laboratory standard by applying the following formula: [standardized creatinine = −0.184 + (0.960 × uncalibrated serum creatinine)].9 Estimated glomerular filtration rate (eGFR) was then calculated using the following formula: eGFR = [175 × (standardized creatinine)−1.154 × (age)−0.203 × (0.742 if the participant is female) × (1.212 if the participant is black)].10 Latex-enhanced nephelometry was used to quantify C-reactive protein (CRP) level (Behring Diagnostics, Somerville, NJ). Fibrinogen was measured in citrated plasma using an automated coagulation autoanalyzer (Organon Teknika, Durham, NC). White blood cell count was obtained as part of the CBC (Coulter Electronics, Hialeah, FL). The bromocresol purple method was used to measure albumin.
In Phase 2 of data collection, the requisite assays for identifying iron, folate, and vitamin B12 deficiency were available.7 As previously defined, iron deficiency was considered present in participants with at least 2 of the 3 following criteria: transferrin saturation < 15%, serum ferritin < 12 ng/mL, and erythrocyte protoporphyrin > 1.24 μM.11 Participants with serum B12 < 147.56 pM were considered to have vitamin B12 deficiency.7 To identify folate deficiency, RBC folate < 232.49 nM was used in participants examined in the mobile examination center, while serum folate < 5.89 nM was used in those examined at home (RBC folate was not measured in blood collected in the home).7
Data analysis
To evaluate the association of RDW with mortality outcomes, RDW values were examined as a continuous variable as well as categorized into quintiles using the following cutoffs: <12.60%, 12.60–12.95%, 13.0–13.40%, 13.45–14.05%, and >14.05%. Participant characteristics were first examined across quintiles of RDW (Table 1). Kaplan-Meier survival curves were then plotted by quintiles of RDW in participants aged 45–64 and in those aged ≥65 (Figure 1). The log-rank test was used to test the equality of survivor functions across RDW quintiles in each age group. Cox proportional hazards models were used to assess the association of RDW with mortality adjusting for multiple risk factors (Table 2). The proportional hazards assumption was confirmed by examining plots of Schoenfeld residuals. Four mortality outcomes were examined: all-cause, cardiovascular, cancer, and other. For each mortality outcome, 3 models were tested. The first model included RDW, age, sex, and ethnicity as predictors (Model 1), while all other risk factors that might confound the RDW-to-mortality association were added in the second model (Model 2). In Phase 2 participants, a third model was tested that included the predictors in Model 2 as well as the nutrient deficiency variables (Model 3). Finally, to determine whether the effect of RDW on mortality occurs in both middle-aged and older adults, Model 2 was stratified by age group for each mortality outcome (Figure 2). Tests of trend were performed by entering RDW as an ordinal variable. Using Stata/SE version 9.2, all of these analyses were weighted and accounted for non-response bias and the complex sampling methods designed to provide estimates for the US population.
Table 1.
Baseline characteristics by quintiles of RDW among adults aged 45 and older*
| Characteristics | Quintile 1 (RDW <12.55%) | Quintile 2 (RDW 12.55–12.95%) | Quintile 3 (RDW 13.00–13.40%) | Quintile 4 (RDW 13.45–14.05%) | Quintile 5 (RDW >14.05%) | p value |
|---|---|---|---|---|---|---|
| Number of participants | 1660 | 1609 | 1724 | 1601 | 1581 | |
| Age, years, mean | 58.3 (0.5) | 60.9 (0.3) | 61.6 (0.6) | 63.9 (0.5) | 65.7 (0.5) | <0.001 |
| Sex, % female | 56.2 (1.9) | 51.1 (1.8) | 53.7 (1.8) | 53.1 (1.8) | 55.8 (1.8) | 0.26 |
| Ethnicity, % | ||||||
| Non-Hispanic White | 87.8 (1.3) | 83.9 (1.5) | 83.4 (1.5) | 78.1 (2.0) | 70.4 (2.2) | |
| Non-Hispanic Black | 3.7 (0.5) | 5.4 (0.5) | 7.4 (0.6) | 11.7 (1.1) | 22.3 (1.8) | <0.001 |
| Mexican American | 2.6 (0.3) | 3.4 (0.3) | 3.1 (0.4) | 3.4 (0.4) | 3.3 (0.3) | |
| All others | 5.9 (1.1) | 7.3 (1.4) | 6.1 (1.3) | 6.8 (1.4) | 4.1 (1.3) | |
| Education, % | ||||||
| Less than high school graduate | 24.5 (1.7) | 30.9 (1.9) | 32.4 (2.2) | 43.5 (1.9) | 48.0 (2.3) | |
| High school graduate | 34.4 (1.4) | 35.1 (1.7) | 31.5 (1.8) | 30.3 (1.7) | 30.4 (2.0) | <0.001 |
| Post-secondary | 41.2 (2.3) | 34.0 (1.9) | 36.1 (1.8) | 26.2 (1.9) | 21.7 (1.9) | |
| Smoking history, % | ||||||
| Never | 47.2 (2.0) | 39.9 (1.5) | 41.9 (1.9) | 37.8 (1.6) | 38.0 (1.7) | |
| Former | 37.7 (2.3) | 39.4 (1.8) | 35.9 (1.6) | 33.9 (2.0) | 33.7 (1.6) | <0.001 |
| Current | 15.1 (1.6) | 20.8 (1.5) | 22.2 (1.4) | 28.3 (1.8) | 28.4 (1.4) | |
| Body mass index, kg/m2, mean | 26.7 (0.1) | 27.2 (0.2) | 27.4 (0.2) | 27.8 (0.2) | 28.2 (0.2) | <0.001 |
| Cancer, % | 6.0 (0.7) | 5.9 (0.6) | 6.0 (0.9) | 7.1 (1.0) | 8.2 (1.0) | 0.33 |
| Congestive heart failure, % | 2.3 (0.4) | 3.3 (0.5) | 4.7 (0.7) | 6.1 (0.7) | 10.5 (1.2) | <0.001 |
| Diabetes, % | 7.7 (0.8) | 9.5 (1.0) | 9.5 (0.9) | 10.5 (0.9) | 13.6 (1.0) | <0.001 |
| Heart attack, % | 4.8 (0.6) | 5.8 (0.7) | 8.0 (0.9) | 12.3 (1.4) | 12.0 (1.0) | <0.001 |
| Pulmonary disease, % | 12.4 (1.1) | 14.6 (0.9) | 16.3 (1.2) | 17.6 (1.5) | 19.4 (1.6) | 0.002 |
| Stroke, % | 2.4 (0.5) | 3.1 (0.6) | 4.6 (0.6) | 6.2 (0.9) | 7.3 (0.7) | <0.001 |
| eGFR, mL/min/1.73 m2, mean | 83.1 (0.6) | 81.3 (0.6) | 80.8 (0.7) | 79.0 (1.0) | 77.8 (1.2) | <0.001 |
| Hemoglobin, g/dL, mean | 14.2 (0.05) | 14.3 (0.04) | 14.2 (0.05) | 14.0 (0.04) | 13.2 (0.05) | <0.001 |
| Mean corpuscular volume, fL, mean | 91.6 (0.2) | 91.1 (0.2) | 90.7 (0.2) | 90.4 (0.2) | 88.7 (0.4) | <0.001 |
| CRP ≥ 10.0 mg/L, % | 5.3 (0.7) | 7.3 (0.9) | 8.9 (1.1) | 16.0 (1.0) | 24.2 (2.0) | <0.001 |
| Fibrinogen, mg/dL, mean | 292.2 (2.7) | 304.2 (3.8) | 311.4 (3.6) | 327.4 (5.4) | 347.4 (5.3) | <0.001 |
| White blood cell count, × 109 cells/L, mean | 7.0 (0.1) | 7.2 (0.1) | 7.2 (0.1) | 7.4 (0.1) | 7.7 (0.1) | <0.001 |
| Albumin, g/dL, mean | 4.2 (0.02) | 4.1 (0.02) | 4.1 (0.03) | 4.0 (0.02) | 3.9 (0.03) | <0.001 |
| Hospitalized, % | 10.8 (0.9) | 12.6 (1.3) | 12.0 (1.3) | 18.0 (1.5) | 25.9 (1.6) | <0.001 |
| Nutritional status, %† | ||||||
| Iron deficient† | 1.2 (0.6) | 1.4 (0.4) | 3.0 (0.7) | 5.7 (1.4) | 21.8 (3.0) | <0.001 |
| Folate deficient† | 7.2 (1.6) | 6.3 (0.9) | 10.3 (1.2) | 12.7 (1.9) | 13.7 (2.0) | 0.003 |
| B12 deficient† | 1.8 (0.5) | 2.7 (0.9) | 7.9 (2.1) | 3.4 (0.7) | 7.7 (1.3) | <0.001 |
Results shown are weighted % (SE) or mean (SE) SI conversion factors: To convert hemoglobin and albumin from g/dL to g/L, multiply by 10; CRP from mg/L to nmol/L, multiply by 9.524; fibrinogen from mg/dL to μmol/L, multiply by 0.0294
Nutrient levels were determined in Phase 2 of NHANES III (1991–1994) where there were 783, 777, 859, 811, and 747 participants in RDW quintiles 1 through 5, respectively
Figure 1.
Kaplan-Meier survival curves by quintiles of RDW according to age group
Table 2.
Risk of death according to RDW among adults aged 45 and older
| RDW quintiles | Unweighted number of deaths | 12-year mortality % | Mortality rate per 1,000 person years | Model 1 HR (95% CI)* (n=8175) | Model 2 HR (95% CI)* (n=7862) | Model 3 HR (95% CI)* (n=3927) |
|---|---|---|---|---|---|---|
| All deaths | ||||||
| Quintile 1 | 287 | 17.3 | 14.3 | 1.0 | 1.0 | 1.0 |
| Quintile 2 | 363 | 22.6 | 18.7 | 1.1 (0.9–1.3) | 1.1 (0.9–1.3) | 0.8 (0.6–1.2) |
| Quintile 3 | 449 | 26.0 | 22.2 | 1.2 (1.0–1.5) | 1.2 (1.0–1.4) | 1.0 (0.7–1.5) |
| Quintile 4 | 538 | 33.6 | 35.6 | 1.6 (1.3–2.0) | 1.4 (1.2–1.8) | 1.5 (1.1–2.1) |
| Quintile 5 | 791 | 50.0 | 63.4 | 2.7 (2.2–3.3) | 2.1 (1.7–2.6) | 2.0 (1.4–2.8) |
| Test for trend | p < 0.01 | p < 0.01 | p < 0.01 | |||
| Cardiovascular deaths | ||||||
| Quintile 1 | 127 | 7.7 | 5.6 | 1.0 | 1.0 | 1.0 |
| Quintile 2 | 180 | 11.2 | 9.6 | 1.3 (1.0–1.8) | 1.4 (1.0–1.9) | 0.9 (0.5–1.5) |
| Quintile 3 | 220 | 12.8 | 10.0 | 1.3 (1.0–1.8) | 1.2 (0.9–1.7) | 1.0 (0.6–1.6) |
| Quintile 4 | 284 | 17.7 | 17.8 | 1.9 (1.4–2.6) | 1.7 (1.2–2.3) | 1.7 (1.1–2.7) |
| Quintile 5 | 384 | 24.3 | 30.6 | 2.9 (2.2–4.0) | 2.5 (1.8–3.3) | 2.1 (1.3–3.4) |
| Test for trend | p < 0.01 | p < 0.01 | p < 0.01 | |||
| Cancer deaths | ||||||
| Quintile 1 | 72 | 4.3 | 3.9 | 1.0 | 1.0 | 1.0 |
| Quintile 2 | 78 | 4.9 | 4.3 | 0.9 (0.6–1.5) | 0.9 (0.6–1.4) | 0.9 (0.4–1.8) |
| Quintile 3 | 88 | 5.1 | 5.5 | 1.2 (0.8–1.8) | 1.1 (0.7–1.7) | 0.9 (0.4–1.8) |
| Quintile 4 | 106 | 6.6 | 8.2 | 1.6 (1.1–2.4) | 1.3 (0.9–2.0) | 1.5 (0.8–2.8) |
| Quintile 5 | 174 | 11.0 | 15.1 | 2.7 (1.8–4.1) | 2.0 (1.3–3.0) | 1.7 (0.8–3.5) |
| Test for trend | p < 0.01 | p < 0.01 | p = 0.07 | |||
| Other deaths | ||||||
| Quintile 1 | 88 | 5.3 | 4.8 | 1.0 | 1.0 | 1.0 |
| Quintile 2 | 105 | 6.5 | 4.8 | 0.8 (0.6–1.2) | 0.8 (0.6–1.2) | 0.7 (0.3–1.3) |
| Quintile 3 | 140 | 8.1 | 6.7 | 1.1 (0.8–1.6) | 1.1 (0.8–1.6) | 1.2 (0.7–2.3) |
| Quintile 4 | 148 | 9.2 | 9.5 | 1.4 (0.9–2.0) | 1.2 (0.8–1.7) | 1.3 (0.7–2.5) |
| Quintile 5 | 232 | 14.7 | 17.6 | 2.4 (1.7–3.4) | 1.8 (1.3–2.6) | 2.0 (1.1–3.6) |
| Test for trend | p < 0.01 | p < 0.01 | p < 0.01 | |||
Model 1 adjusted for age, sex, and ethnicity
Model 2 adjusted for Model 1 covariates and education, BMI, smoking status, cancer, congestive heart failure, diabetes, heart attack, pulmonary disease, stroke, overnight hospitalization, estimated GFR, hemoglobin concentration, mean corpuscular volume, and CRP level.
Model 3 adjusted for Model 2 covariates and iron, folate, and B12 deficiency using data collected in Phase 2 of NHANES III (1991–1994)
Figure 2.
Risk of death by RDW according to age group and cause of death*.
The added ability of RDW to predict all-cause mortality was evaluated using methods recommended by Cook and Pencina et al (Table 3).12–14 To compare the global fit of models with and without RDW as a continuous variable predicting death, the −2 log likelihood and Bayes Information Criterion (BIC) were examined. Lower values of these sensitive measures indicate better model fit and the difference between −2 log likelihoods of nested models can be compared using a χ2 distribution, known as the Likelihood Ratio test. The Hosmer-Lemeshow statistic was used to assess model calibration by comparing predicted and observed probabilities of death across categories of 5% increments in predicted risk. To assess discrimination, the area under the receiver operating characteristic curve (AUC) as well as the C statistic, which is similar to the AUC but allows for time to event analysis, were examined (higher values indicate better discrimination). In addition, the integrated discrimination improvement (IDI) statistic was calculated, which compares improvement in the integral of sensitivity to any change in the intergral of the false positive rate (1-specificity) when a new marker (RDW) is added.14 Higher IDI values indicate better discrimination in risk prediction. Unweighted Cox models were used to obtain −2 log likelihood, BIC, and C statistics, while the Hosmer-Lemeshow, AUC, and IDI were obtained with unweighted logistic regression models that examined deaths through 6 years of follow-up, which was available for all study participants.
Table 3.
Summary statistics to compare risk prediction of models with and without RDW entered as a predictor of all-cause mortality, by age group*
| < 65 years | ≥ 65 years | |||
|---|---|---|---|---|
| Measures of Model Prediction | With RDW | Without RDW | With RDW | Without RDW |
| Calibration/Global Fit | ||||
| −2 Log Likelihood | 6365.3 | 6408.5 | 27904.3 | 27985.4 |
| Likelihood Ratio Test | p < 0.001 | p < 0.001 | ||
| Bayes Information Criterion | 6530.4 | 6565.3 | 28070.3 | 28143.1 |
| Hosmer-Lemeshow Goodness of Fit | p = 0.99 | p = 0.61 | p = 0.75 | p = 0.62 |
| Discrimination | ||||
| C Statistic | 0.7529 | 0.7397 | 0.7387 | 0.7305 |
| Area Under Receiver Operating Characteristic Curve | 0.7786 | 0.7639 | 0.7897 | 0.7799 |
| Integrated Discrimination Improvement, % | 1.27 | 1.55 | ||
| p <0.001 | p<0.001 | |||
All models included the following predictors: age, sex, ethnicity, education, BMI, smoking status, cancer, congestive heart failure, diabetes, heart attack, pulmonary disease, stroke, overnight hospitalization, estimated GFR, and CRP level.
RESULTS
Characteristics associated with RDW
RDW values ranged from 11.0% to 30.6% (median = 13.15%; interquartile range = 12.65–13.85%). Baseline participant characteristics are shown by quintiles of RDW in Table 1. Participants with higher RDW values were more likely to be older, less educated, currently smoking, and had higher BMI than those with lower RDW values. The proportion of non-Hispanic Blacks increased from the lowest to highest quintiles of RDW. With the exception of cancer, the prevalence of age-associated diseases and hospitalization in the past year increased with higher RDW quintiles. In addition, estimated GFR, hemoglobin concentration, and MCV decreased with higher RDW, while CRP, fibrinogen, and white blood cell count increased with higher RDW. As expected, higher prevalence of nutrient deficiencies was observed in participants with higher RDW, although a clear gradient across RDW quintiles was not seen with vitamin B12 deficiency.
RDW and all-cause mortality
Figure 1 graphically displays the probability of survival over the follow-up period by quintiles of RDW for middle-aged and older adults. Middle-aged adults in the highest quintile of RDW had poorer survival relative to those with lower RDW. Although the number of deaths was relatively low among middle-aged participants in the lower quintiles of RDW, an intermediate survival pattern was observed in those with RDW of 13.45–14.05% (4th quintile). In older adults, there was a clear survival gradient across RDW quintiles sustained over the 12 years of follow-up. Higher RDW was associated with poorer survival among older participants, particularly for those in the 4th and 5th quintiles of RDW.
After adjusting for age, sex, and ethnicity, there remained a stepwise, graded association between RDW levels and all-cause mortality in adults aged 45 and older (Model 1, Table 2). Participants in the 4th and 5th quintiles of RDW were 1.6 and 2.7 times more likely to die, respectively, compared to those in the lowest quintile of RDW. Further adjustment for major age-associated diseases as well as education, BMI, smoking status, hospitalizations, renal function, hemoglobin concentration, MCV, and CRP level only partially attenuated the effect of RDW on all-cause mortality (Model 2). Even when the study population was restricted to those with serum markers of nutrition (participants in Phase 2), higher RDW remained strongly associated with increased risk of death while adjusting for all potential confounding factors including iron, folate, and vitamin B12 deficiency (Model 3). When RDW was examined as a continuous variable, mortality risk increased by 22% for every 1 percent increment in RDW adjusting for Model 2 covariates [HR = 1.22 (95%CI: 1.15–1.30)]. The association of RDW with mortality did not vary significantly by sex (p=0.45 for interaction term).
RDW and cause-specific mortality
In addition to overall mortality, RDW was a particularly strong predictor of cardiovascular deaths (Table 2). The risk of dying from cardiovascular diseases was nearly 2-and 3-fold higher in participants with RDW values in the 4th and 5th quintiles, respectively, compared to those in the bottom quintile (Model 1). Adjustment for other risk factors did not substantially diminish this association. For cancer mortality, there was also a significant risk gradient across RDW quintiles, although the hazard ratios (HR) for the upper quintiles were no longer statistically significant when adjusting for nutrient deficiencies in Phase 2 participants (Model 3). Mortality due to other causes was also significantly associated with RDW. Participants in the highest quintile had a 2-fold increased risk of death from non-cardiovascular and non-cancer causes compared those in the lowest quintile of RDW.
Age-stratified analyses
Figure 2 illustrates that the association of RDW with mortality adjusted for multiple risk factors occurs in both middle-aged and older adults. There was a significant trend of increased risk of death from all-causes with higher RDW in both age groups. Although the number of events was much smaller in participants aged 45–64 than in those who were 65 and older, risk of death from cardiovascular diseases, cancer, and other causes was significantly higher in participants in the highest RDW quintile compared to those in the lowest quintile for both age groups.
Mortality risk prediction by RDW
The added prognostic value of RDW to mortality risk prediction is demonstrated by the summary statistics of model performance shown in Table 3. In both middle-aged and older adults, the addition of RDW significantly improved model fit as indicated by lower −2 log likelihood and BIC values as well as by the significant Likelihood Ratio test results. The Hosmer-Lemeshow statistic comparing predicted to observed mortality risk indicated excellent calibration when RDW was included in the model. In addition, the C statistic and AUC values improved in models with RDW, reflecting better discrimination in risk prediction. Further, the significant IDI value of 1.27% in middle-aged adults resulted from a significant increase in the average sensitivity of 1.19% (p<0.001) and a decrease in the average false positive rate of 0.08% (p = 0.06) when RDW was included as a predictor of mortality. In older adults, the average sensitivity increased by 1.07% (p < 0.001), while the average false positive rate decreased by 0.48% (p < 0.001).
Sensitivity analyses
To further evaluate the association of RDW with mortality, a series of sensitivity analyses were performed. When participants with RDW greater than 15.0% were excluded, RDW remained significantly associated with all-cause mortality [comparing the highest RDW quintile to lowest, multivariable adjusted HR=2.0 (95%CI:1.6–2.4); p for trend <0.01)]. In fact, Figure 3 illustrates that there was a continuous rise in mortality risk among participants with RDW values of 11.0–15.0%. For every percentage point increase in RDW, risk of death increased by 37% adjusting for demographic, behavioral, and biomedical risk factors [HR=1.37 (95%CI:1.24–1.51)]. The actual observed proportion of deaths for each 0.5% interval of RDW (individual points) closely matched the modeled mortality risk represented by the solid line in Figure 3, supporting the validity of the risk modeling. Further exclusion of participants with iron, folate, or vitamin B12 deficiency did not eliminate the association [comparing the highest RDW quintile to lowest, multivariable adjusted HR=2.2 (95%CI:1.5–3.2); p for trend <0.01)]. Additionally, RDW was a significant predictor of all-cause mortality among non-anemic participants (baseline hemoglobin >13.0 g/dL in women and >14.0 g/dL in men) [comparing the highest RDW quintile to lowest, multivariable adjusted HR=2.1 (95%CI:1.6–2.8); p for trend <0.01)]. Finally, the effect of RDW on survival was only slightly reduced and remained statistically significant after further adjusting for white blood cell count, fibrinogen, and serum albumin.
Figure 3.
Probability of death by RDW levels in the normal range among adults aged 45 and older
DISCUSSION
In this nationally representative study, RDW was a strong predictor of mortality in both middle-aged and older adults. Mortality rates were graded across the entire distribution of RDW and were particularly elevated in participants with RDW greater than 13.4%. Deaths from CVD, cancer, and other causes were all associated with RDW, although the effect was stronger for CVD mortality. Associations were independent of multiple potential confounding factors. Even when participants with high RDW values or nutritional deficiencies were excluded, RDW remained strongly associated with mortality. Importantly, RDW predicted mortality among participants who were clearly non-anemic. Summary measures of global model fit, model calibration, and model discrimination further showed that RDW significantly improved mortality risk prediction. The magnitude and robustness of these associations indicate that RDW is an age-associated biomarker that is prognostic in adults aged 45 and older.
Although the exact physiologic mechanisms that underlie the association of RDW with survival are unknown, systemic factors that alter erythrocyte homeostasis, such as inflammation and oxidative stress, likely play a role. Inflammation might contribute to increased RDW levels by not only impairing iron metabolism but also by inhibiting the production of or response to erythropoietin or by shortening red blood cell survival.15, 16 Using an index comprised of multiple proinflammatory cytokines, Ferrucci and colleagues showed that higher levels of inflammation were associated with higher erythropoietin concentration among non-anemic older adults, while an inverse association was observed in anemic persons.17 This suggests that in a pro-inflammatory state the increase in erythropoietin is a compensatory mechanism for maintaining normal hemoglobin concentration and that anemia occurs when the compensatory increment in erythropoietin production is unsustainable. Indeed, a number of studies have shown that proinflammatory cytokines suppress erythropoietin gene expression, inhibit proliferation of erythroid progenitor cells, down-regulate erythropoietin receptor expression, and reduce erythrocyte life-span.16 Importantly, Ershler and colleagues demonstrated in a longitudinal cohort that erythropoietin increased with aging, which again suggests that increased erythropoietin production is a compensatory mechanism for decreasing bone marrow response and/or red cell survival.18 In the current study, older age as well as elevated CRP, fibrinogen, and white blood cell count were strongly associated with higher RDW levels (Table 1). Even though RDW predicted mortality in participants who were non-anemic, it is possible that there are fluctuations in erythropoiesis caused by inflammation that lead to greater heterogeneity in erythrocyte size as younger erythrocytes are usually larger and more variable in size than older ones.
In addition to inflammation, oxidative stress might also contribute importantly to anisocytosis. While erythrocytes have tremendous antioxidant capacity and serve as the primary “oxidative sink”, they are prone to oxidative damage that reduces cell survival.19 In a population-based study, higher RDW values were independently associated with poorer pulmonary function, a condition associated with oxidative stress.20 Among persons with Down syndrome, a condition characterized by accelerated aging, levels of oxidative stress were significantly higher than in healthy controls.21 The Down syndrome group in that study had higher RDW values than controls (14.5% vs. 12.8%), but this difference was not viewed as important because both means were in the normal range.21 In light of our findings, however, these differences in RDW are not trivial in terms of mortality risk. Hemodialysis patients also experience oxidative stress and inflammation because of blood contact with dialysis membranes. A small randomized trial of hemodialysis patients showed that a variety of outcomes, including RDW values, improved over 1 year in patients treated with a vitamin E-bonded cellulose membrane hemodialyzer compared to those treated with a standard cellulose membrane.22 While more mechanistic research is needed, these studies indicate that variation in erythrocyte size is associated with oxidative stress, possibly through increased red cell turnover.
Findings from the current study are consistent with other recently published studies. In a randomized controlled trial of heart failure patients, Felker and colleagues showed that patients with RDW >15.8% had nearly a 2-fold increased risk of CVD death or hospitalization as well as death from any cause compared to those with RDW <13.3%. As a validation, Felker et al. also showed a 2-fold increased mortality risk comparing the highest RDW quintile (>15.5%) to the lowest (<13.0%) in a separate clinical cohort of heart failure patients. Similarly, Tonnelli et al. analyzed data from a randomized controlled trial of patients with coronary artery disease (CAD) and reported that death was twice as likely in patients with RDW >13.8% versus those with RDW <12.6%. Finally, Anderson et al. reported that the risk of death over 1 year was 3 times higher in patients undergoing cardiac catheterization with RDW >14.0% relative to those with RDW <12.7%. All of these studies were able to adjust for major CVD risk factors, but data on nutritional status or inflammation were not available.
As with other studies, a limitation of the NHANES III is that laboratory assessments were made on a single occasion and therefore fluctuations in the CBC could not be evaluated. In addition, erythropoietin, reticulocyte count, markers of oxidative stress, and more sensitive assays of CRP and other proinflammatory cytokines were not available, but could provide valuable clues as to the pathophysiology underlying anisocytosis (exploratory analyses showed that serum antioxidants, such as carotenoids, selenium, and vitamin E, were strongly associated with RDW, but practically did not change the effect of RDW on mortality). Furthermore, the sampling occurred in an era that preceded folate food supplementation, which could lower RDW. However, the relationship of RDW with mortality remained after adjusting for folate deficiency and exclusion of participants with nutritional deficiencies, so it is likely that this relationship would persist in the post-supplementation era. The current study has several strengths including the large, community-based sample that is representative of nearly 73 million US adults aged 45 and older. The relatively long follow-up period provided a sufficient number of deaths to examine cause-specific mortality. Finally, unlike previous studies, the current analysis adjusted for a comprehensive set of potential confounders, including iron, folate, and vitamin B12 deficiency as well as CRP levels.
Red cell distribution width is a widely available and inexpensive test that is prognostic in the general population of adults aged 45 and older. Given that RDW rises with age and strongly predicts mortality, it is conceivable that anisocytosis might reflect impairment of multiple physiologic systems related to the aging process or is caused by inflammation and age-associated diseases. While further research is needed to elucidate the mechanisms, RDW provides prognostic information that can be used to improve risk stratification. Future studies should further evaluate the clinical utility of RDW for risk prediction as well as characterize change in RDW over time and define associations with other age-associated outcomes, such as the onset of anemia, physical disability, and cognitive impairment.
Acknowledgments
This study was supported by the Intramural Research Program of the US National Institute on Aging, National Institutes of Health. Dr. Patel had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
References
- 1.Sategna Guidetti C, Scaglione N, Martini S. Red cell distribution width as a marker of coeliac disease: a prospective study. Eur J Gastroenterol Hepatol. 2002;14:177–81. doi: 10.1097/00042737-200202000-00012. [DOI] [PubMed] [Google Scholar]
- 2.Spell DW, Jones DV, Jr, Harper WF, Bessman JD. The value of a complete blood count in predicting cancer of the colon. Cancer Detect Prev. 2004;28:37–42. doi: 10.1016/j.cdp.2003.10.002. [DOI] [PubMed] [Google Scholar]
- 3.Anderson JL, Ronnow BS, Horne BD, et al. Usefulness of a complete blood count-derived risk score to predict incident mortality in patients with suspected cardiovascular disease. Am J Cardiol. 2007;99:169–74. doi: 10.1016/j.amjcard.2006.08.015. [DOI] [PubMed] [Google Scholar]
- 4.Felker GM, Allen LA, Pocock SJ, et al. Red cell distribution width as a novel prognostic marker in heart failure: data from the CHARM Program and the Duke Databank. J Am Coll Cardiol. 2007;50:40–7. doi: 10.1016/j.jacc.2007.02.067. [DOI] [PubMed] [Google Scholar]
- 5.Tonelli M, Sacks F, Arnold M, Moye L, Davis B, Pfeffer M. Relation between red blood cell distribution width and cardiovascular event rate in people with coronary disease. Circulation. 2008;117:163–168. doi: 10.1161/CIRCULATIONAHA.107.727545. [DOI] [PubMed] [Google Scholar]
- 6.Plan and operation of the Third National Health and Nutrition Examination Survey, 1988–94. Series 1: programs and collection procedures. Vital Health Stat 1. 1994:1–407. [PubMed] [Google Scholar]
- 7.Gunter EWLB, Koncikowski SM. Laboratory procedures used for the Third National Health and Nutrition Examination Survey (NHANES III), 1988–1994. Hyattsville, MD: National Center for Health Statistics, Centers for Disease Control and Prevention; 1996. [Google Scholar]
- 8.Anderson RN, Minino AM, Hoyert DL, Rosenberg HM. Comparability of causes of death between ICD-9 and ICD-10: Preliminary estimates. National Vital Statistics Reports. 2001;49:1–32. [PubMed] [Google Scholar]
- 9.Selvin E, Manzi J, Stevens LA, et al. Calibration of serum creatinine in the National Health and Nutrition Examination Surveys (NHANES) 1988–1994, 1999–2004. Am J Kidney Dis. 2007;50:918–26. doi: 10.1053/j.ajkd.2007.08.020. [DOI] [PubMed] [Google Scholar]
- 10.Levey AS, Coresh J, Greene T, et al. Using standardized serum creatinine values in the modification of diet in renal disease study equation for estimating glomerular filtration rate. Ann Intern Med. 2006;145:247–54. doi: 10.7326/0003-4819-145-4-200608150-00004. [DOI] [PubMed] [Google Scholar]
- 11.Looker AC, Dallman PR, Carroll MD, Gunter EW, Johnson CL. Prevalence of iron deficiency in the United States. JAMA. 1997;277:973–6. doi: 10.1001/jama.1997.03540360041028. [DOI] [PubMed] [Google Scholar]
- 12.Cook NR. Use and misuse of the receiver operating characteristic curve in risk prediction. Circulation. 2007;115:928–35. doi: 10.1161/CIRCULATIONAHA.106.672402. [DOI] [PubMed] [Google Scholar]
- 13.Cook NR. Statistical evaluation of prognostic versus diagnostic models: beyond the ROC curve. Clin Chem. 2008;54:17–23. doi: 10.1373/clinchem.2007.096529. [DOI] [PubMed] [Google Scholar]
- 14.Pencina MJ, D’Agostino RB, Sr, D’Agostino RB, Jr, Vasan RS. Evaluating the added predictive ability of a new marker: from area under the ROC curve to reclassification and beyond. Stat Med. 2008;27:157–72. doi: 10.1002/sim.2929. [DOI] [PubMed] [Google Scholar]
- 15.Douglas SW, Adamson JW. The anemia of chronic disorders: studies of marrow regulation and iron metabolism. Blood. 1975;45:55–65. [PubMed] [Google Scholar]
- 16.Weiss G, Goodnough LT. Anemia of chronic disease. N Engl J Med. 2005;352:1011–23. doi: 10.1056/NEJMra041809. [DOI] [PubMed] [Google Scholar]
- 17.Ferrucci L, Guralnik JM, Woodman RC, et al. Proinflammatory state and circulating erythropoietin in persons with and without anemia. Am J Med. 2005;118:1288. doi: 10.1016/j.amjmed.2005.06.039. [DOI] [PubMed] [Google Scholar]
- 18.Ershler WB, Sheng S, McKelvey J, et al. Serum erythropoietin and aging: a longitudinal analysis. J Am Geriatr Soc. 2005;53:1360–5. doi: 10.1111/j.1532-5415.2005.53416.x. [DOI] [PubMed] [Google Scholar]
- 19.Kiefer CR, Snyder LM. Oxidation and erythrocyte senescence. Curr Opin Hematol. 2000;7:113–6. doi: 10.1097/00062752-200003000-00007. [DOI] [PubMed] [Google Scholar]
- 20.Grant BJ, Kudalkar DP, Muti P, et al. Relation between lung function and RBC distribution width in a population-based study. Chest. 2003;124:494–500. doi: 10.1378/chest.124.2.494. [DOI] [PubMed] [Google Scholar]
- 21.Garcez ME, Peres W, Salvador M. Oxidative stress and hematologic and biochemical parameters in individuals with Down syndrome. Mayo Clin Proc. 2005;80:1607–11. doi: 10.4065/80.12.1607. [DOI] [PubMed] [Google Scholar]
- 22.Kobayashi S, Moriya H, Aso K, Ohtake T. Vitamin E-bonded hemodialyzer improves atherosclerosis associated with a rheological improvement of circulating red blood cells. Kidney Int. 2003;63:1881–7. doi: 10.1046/j.1523-1755.2003.00920.x. [DOI] [PubMed] [Google Scholar]