Background
Red cell distribution width (RDW), a measure of variation in red blood cell size also known as anisocytosis, is an emerging biomarker of chronic disease morbidity and mortality [
1]. RDW is associated with increased risk of cardiovascular diseases (CVD) including coronary artery disease, heart failure, atherosclerosis, and venous thromboembolism independent of traditional CVD risk factors [
2‐
7]. Elevated RDW was also associated with overall and disease-specific mortality risk [
6,
8‐
11]. Most of these studies, however, were conducted in older individuals, and hence the role of RDW on mortality risk in younger individuals is poorly understood. Although the specific biological mechanisms behind the association between RDW and adverse health outcomes have not been fully explained, it was suggested that it could be mediated through inflammatory cytokines, or oxidative stress [
2,
9,
12]. RDW is influenced by factors such as sex, race, body mass index (BMI), cigarette smoking, markers of systemic inflammation and lipids although results were inconclusive [
6,
7,
12,
13]. Whereas genetic variants in loci including
G6PD,
CD36, and
NOL4L have been linked with RDW level [
14,
15], the interplay between environmental factors and genetic variants on RDW has not yet been investigated.
While the age-adjusted mortality rate in the United States has been declining, there remains a gap in premature mortality and life expectancy among racial/ethnic groups and individuals of various socioeconomic status [
16,
17]. Besides, Case and Deaton recently reported that in the past 10–15 years there has been an alarmingly increasing mortality rate among white American men and women aged 30–54 years [
18]. Established risk factors of mortality include tobacco use, obesity, and clinical markers such as elevated blood pressure and hyperlipidemia [
19]. The growing mortality disparity and widening of life expectancy gap suggest a need to identify additional biomarkers to improve early detection of at-risk individuals to enhance precise and targeted interventions.
This new evidence of sharply declining longevity not only among African Americans (AA) but also young whites has made us reevaluate the role of RDW in younger populations of urban AA and whites since RDW was shown to be a good predictor of mortality for those in middle and old age [
11]. Using prospectively collected data (August 2004–December 2013) from the Healthy Aging in Neighborhoods of Diversity across the Life Span (HANDLS) study, which is a large longitudinal study of socioeconomically diverse community-dwelling AA and white working-age adults, the objectives of this study were (1) to assess the association between RDW and all-cause and cause-specific mortality; (2) to identify lifestyle and environmental determinants of RDW; and (3) to explore the effect of gene × environment interaction on RDW.
Methods
Population and study design
The HANDLS study is a large population-based prospective longitudinal study conducted in Baltimore, MD. Study design and methods of data collection have been described previously [
20]. Briefly, the HANDLS study was designed to establish a single-site study for the investigation of age-related health disparities in socioeconomically diverse adult AAs and whites. The initial wave of data collection recruited 3720 individuals from 2004 to 2009. Participants were men and women aged 30–64 years, above and below 125% of the 2004 US Federal poverty level. At baseline, 2744 participants had complete RDW data. We excluded 18 subjects with sickle cell disease leaving 2726 for the current analysis.
Red cell distribution width and covariates
During the baseline data collection (2004–2009), study participants were interviewed using structured questionnaires, underwent medical examinations, and provided social and medical histories as well as a blood sample. Data included age, sex, race, poverty status, BMI (kg/m
2), waist-hip ratio (WHR), current smoking status, current alcohol use, highest education level attained and reported illicit drug use (e.g., marijuana, cocaine). RDW was measured by automated Coulter DXH 800 hematology analyzer as part of peripheral complete blood count (Beckman Coulter, Brea, CA), and was expressed as coefficient of variation (%) of red blood cell volume distribution. Regular calibration was performed every 3 months on the hematology analyzer and quality control was performed according to the manufacturer’s recommendations. Hypertension was defined as self-reported history of hypertension, use of anti-hypertensive medications, or blood pressure of > 140/90. Diabetes mellitus was defined as self-reported history of Type I or II diabetes mellitus, reported use of diabetes medication, or fasting plasma glucose ≥ 126 mg/dl. Self-reported depressive symptoms were assessed using the Center for Epidemiologic Studies-Depression (CES-D) scale [
21]. Additional covariates included hemoglobin, serum iron, serum vitamin B12, total white blood cell (WBC) count, erythrocyte sedimentation rate (ESR), high-sensitivity C-reactive protein (hsCRP), low density lipoprotein-cholesterol (LDL), and estimated glomerular filtration rate (eGFR) [
22].
Mortality and cause of death assessment
Mortality was ascertained through the National Death Index (NDI) database. Each study participant was matched to the NDI using name, date and state of birth, sex, race, maiden name and social security number [
16,
23]. Participants were followed from the date of study enrolment to date of death or censored date, December 31, 2013. Primary cause of death is based on the International Classification of Diseases 10th edition (ICD-10). Death due to cardiovascular diseases included the ICD codes from I10.0 to I82.9. A total of 2704 participants had complete follow-up and mortality outcome data.
Genotyping
A subset of AA participants was genotyped using Illumina 1 M and Illumina 1 M-Duo genotyping arrays (Illumina, San Diego, CA). Genotype calling and quality control followed standard procedures. Samples with call rate < 95%, sex mismatch, ethnic ancestry outliers and cryptic related individuals were excluded. SNPs with minor allele frequency less than 1%, Hardy–Weinberg equilibrium
p value of less than 1.0 × 10
−7, call rate < 95% were excluded. After sample and genotype quality control, 1024 AAs had complete genome-wide genotype data. Genotype data management and quality control was performed using PLINK (
https://www.cog-genomics.org/plink2) [
24]. Principal components were generated using a set of linkage disequilibrium based pruned independent single nucleotide polymorphisms. This set of single nucleotide polymorphisms were selected under PLINK default settings of pairwise correlation coefficient of 0.2 between single nucleotide polymorphisms in a sliding window of size of 50. The correlation coefficient of 0.2 was used to select near completely independent SNPs to avoid collinearity between them. To increase power, genotypes were imputed using the 1000 Genomes Project phase 1 version 3 reference panel. Genotypes were phased using the MaCH software and imputation was performed using minimac software (
http://genome.sph.umich.edu/wiki/MaCH).
Statistical analysis
Distributions of continuous variables were plotted using histograms and were checked for skewness. To achieve normal distribution, hsCRP, ESR, and WBC count were natural-logarithm transformed. Categorical variables included in the analysis were: sex, race (AA/white), poverty status (above/below), college degree (yes/no), current cigarette smoking (yes/no), current alcohol use (yes/no), hypertension (yes/no), diabetes mellitus (yes/no), CES-D score (< 16/≥ 16), marijuana use (yes/no), current cocaine use (yes/no).
Survival analysis
We fitted unadjusted and multivariable adjusted Cox proportional hazards regression models to estimate hazard ratios and 95% confidence intervals (CI) of all-cause mortality and CVD-specific mortality. Age at study entrance and exit (death or censored date) were used as the measurement of follow-up time [
25]. To allow a balanced comparison of mortality risk across the different strata, RDW was categorized by quartiles (cut-offs: 13.2, 13.8, and 14.6%). We initially adjusted for age, sex, race, and poverty status. The full Cox regression model was also adjusted for established predictors of mortality and potential confounders: current cigarette smoking, BMI, LDL, hypertension and diabetes mellitus. Test for linear trend across the categories of RDW was performed by assigning the median value to each quartile of RDW and included this variable as a continuous term in the full Cox regression model. The proportional hazards assumption was checked by inspection of log–log survival curves, and formally evaluated using the scaled Schoenfeld residuals method yielding p-values > 0.1 suggesting the proportional hazards assumptions were not violated. Although for the sake of parsimony we adjusted the full Cox regression model for the above nine variables, we performed sensitivity analyses by adding the following variables one at time to the full model, and simultaneously adjusting for them: WHR, systolic blood pressure, hemoglobin, serum iron, serum vitamin B12, ESR, hsCRP, WBC count, and eGFR.
To assess the association between sickle cell trait (SCT) and RDW and its role on the association between RDW and mortality risk, we used HBB-rs334 (A > T, p.Glu6Val, minor allele frequency = 6.2%) to determine SCT status among study participants. We compared RDW level between non-carriers and carriers of the sickle cell allele. Main effects of SCT as well as interaction between RDW and SCT on all-cause and CVD-specific mortality risk were estimated.
We explored effect modifications by established predictors of mortality by including a multiplicative interaction term in the full Cox regression model. The variables tested for interaction were age, sex, race, poverty status, current cigarette smoking, BMI, LDL, hypertension, and diabetes mellitus. Evidence of interaction was tested by comparing the models with and without the interaction term using a likelihood ratio test.
To assess the performance of RDW as predictive marker of all-cause and CVD-specific mortality we assessed model calibration of the fully adjusted Cox regression models using the modified Nam-D’Agostino goodness-of-fit (GOF) test for survival data [
26].
Determinants of red cell distribution width
To identify determinants of RDW, we fitted multiple linear regression models to estimate beta coefficients and 95% CIs. The initial model included age, sex, race, and poverty status; and the full model was further adjusted for potential confounders and previously reported determinants of RDW [
5,
6] including smoking status, BMI, hypertension, diabetes mellitus, LDL, eGFR, and hsCRP. Sensitivity analyses were performed by further adjusting the full model for hemoglobin levels.
Gene × environment interaction analysis in RDW
Among AAs (N = 998) who had complete RDW and genotype data, gene × environment interactions were assessed between lifestyle factors and genetic variants by including an interaction term into the regression model adjusted for age, sex and the first five principal components to account for population stratifications. The variants tested were:
TMEM57-
RHD-rs10903129,
NOL4L-rs4911241,
CD36-rs3211938,
LINC01184-
SLC12A2-rs10063647,
LINC01184-
SLC12A2-rs10089,
TRIB1-rs2954029 [
14,
15]. These variants were selected since they were identified by one of the largest genetic association studies and may have biological significance in red cell biology. All statistical tests were two-sided, and p-value < 0.05 was considered significant. Data analyses were performed using the R statistical software version 3.2.3 (
https://www.R-project.org/) [
27].
Discussion
In the present study, using prospectively collected clinical, genetic and mortality data, we assessed the association between RDW and all-cause and disease-specific mortality risk, as well as investigated determinants of RDW. Elevated RDW conferred a 1.73-times increased risk of all-cause mortality and a 2.49-times increased risk of CVD-specific mortality in urban adults. This elevated risk was substantially modified by BMI. Our results confirmed previously reported predictors of RDW such as cigarette smoking, BMI and CRP, and identified novel associations between RDW and illicit drug use: marijuana and current cocaine use, and WHR.
Our findings of positive associations between RDW and premature mortality among younger urban adults are consistent with previous reports of elevated RDW and all-cause and CVD-specific mortality among middle and older age individuals. Two separate studies using the National Health and Nutrition Examination Survey (NHANES) (1988–1994) data reported that the highest quantile of RDW was associated with a 2-fold, and 2.1- to 2.3-fold increased risk of all-cause and CVD-specific mortality, respectively [
9,
10]. In a meta-analysis of RDW and mortality in older individuals, Patel et al. reported a positive association between RDW and all-cause and cause specific mortalities due to CVD, cancer, and other causes [
11]. It should be noted that our cohort represents adults younger (mean age = 48.1 years) than participants included in the NHANES study (mean age = 62.0 years) [
9], and the meta-analysis (mean age ranged 73.6–79.1 years) [
11]. Also, the presents study contains a large proportion of AAs (57%) compared to 16.9% in previous studies of RDW and mortality [
11].
Contrary to the declining mortality rate trends seen in industrialized countries, the recent rise of premature mortality rates among US adults in their prime-age is a serious clinical and public health problem [
18]. Most of these early deaths, measured as years of life lost, were due mainly to preventable factors such as ischemic heart disease, drug and alcohol use disorders which are themselves multifactorial in origin [
18,
19,
28]. While some of these factors (socioeconomic factors, tobacco use, drug and alcohol use, sedentary lifestyle) have long been the focus of clinical and public health prevention strategies, there is a substantial disparity in premature mortality, particularly among young vulnerable and disadvantaged individuals [
16,
17,
29]. To improve targeted and precise interventions and eliminate this disparity gap, identifying readily available and less expensive biomarkers of premature mortality is important. The fact that RDW is simple and routinely measured clinical parameter makes it a useful tool for health prevention intervention strategies.
RDW reflects variations in red blood cell size (anisocytosis). In addition to aiding the clinical evaluation of anemia, RDW is independently associated with chronic disease risk and mortality [
6]. The exact pathophysiological mechanisms behind this association are not clearly understood although inflammatory cytokines, oxidative stress, and neurohumoral factors are implicated as potential mediators [
2,
9,
12,
30]. Inflammatory cytokines such as interleukin-1, tumor necrosis factor-α, and interferon-γ, which are known to be released in chronic inflammatory states, could affect bone marrow red blood cell production, maturation, and could subsequently lead to anisocytosis [
31]. Although we found a positive association between RDW and inflammation markers, it should be noted that the effect of RDW on both all-cause and cardiovascular mortality risk was independent of hsCRP and ESR—two commonly used markers of systemic inflammation. On the other hand, oxidative stress and systemic inflammation have a complex interrelationship and they are associated with morbidity and mortality. A recent study conducted in elderly individuals showed that serum oxidative stress markers (e.g., derivatives of reactive oxygen metabolites and total thiol levels) were significantly associated with all-cause mortality [
32]. However, following adjustment for CRP, the effect of derivatives of reactive oxygen metabolites on mortality was no longer significant. Further, the positive correlation between derivatives of reactive oxygen metabolites and CRP suggested the effect of these non-specific markers of oxidative stress might partly be explained by inflammation [
32]. To date, there are no studies that directly investigated oxidative stress and RDW in relation to mortality risk and thus we cannot rule out that the effect of increased RDW on mortality is not mediated through oxidative stress processes. To understand the role of SCT on the association between RDW and mortality risk, we assessed the confounding effect of SCT and interaction between SCT and RDW. We found no evidence of a main effect, interaction or confounding effect for SCT. Together with the results of model calibration analyses, these data indicate that RDW could be a valuable predictive biomarker of mortality to identify not only older individuals but also younger adults at higher risk of premature mortality. Future studies utilizing red blood cell specific oxidative stress markers such as fluorescent heme degradation products [
33] which are not associated with CRP [
34] could shed light on the causal link between RDW, oxidative stress and mortality risk.
The findings in the present study that sex, race, cigarette smoking, BMI, and hsCRP influence RDW are consistent with previous reports [
6,
7,
12,
13]. In the present study, we provide the first evidence that WHR and illicit drug use as novel independent determinants of RDW. The positive association between WHR and RDW independent of BMI and other factors known to influence RDW suggests that central obesity as novel predictor of RDW, and could shine light as to how RDW is associated with chronic disease morbidity and mortality.
Interestingly, we found inverse associations between marijuana and cocaine use and RDW independent of potential confounders including systemic inflammation markers. There are no experimental data on the link between RDW, cocaine, and ∆
9-tetrahydrocannabinol (THC). In humans, little is known about the effect of illicit drug use on red blood cell physiology. Two small studies that assessed the effect of cocaine on bone marrow function reported conflicting findings. A study by Siegel et al. showed cocaine use resulted in erythrocytosis while another study by Weber et al. found no significant association between cocaine use and red blood cell count as well as reticulocyte count suggesting that cocaine use might not have effect on bone marrow mediated erythropoiesis [
35,
36]. Cocaine use is known to have negative effects including raised blood pressure and oxidative stress damage to cardiac tissues [
37]. Although cannabinoids regulate the production of inflammatory mediators [
38] and sometimes prescribed by physicians for therapeutic benefits, recent reports indicate that marijuana (natural or synthetic toxic street form) is associated with production of reactive oxygen species, and cardiovascular diseases [
39]. These findings imply that, in the context of the current drug use epidemic, the significance of RDW as a marker of mortality requires additional evaluation. Further experimental studies are required to elucidate the molecular mechanism behind illicit drug use and RDW. We did not find a significant gene x environment interaction in our targeted analysis after correcting for multiple testing. This could be due to the smaller sample size with genotype data. Future interaction studies on larger samples using genome-wide sequence variants are required to understand the influence of environmental exposure on genes that affect RDW levels.
The limitations of the present study include the following: first, the lack of genotype data among white participants that precluded gene × environment interaction analysis. Second, the absence of objectively measured data in our cohort on potential confounders such as serum nutrients and antioxidants, although a previous study showed serum antioxidants did not change the effect of RDW on mortality [
9]. Third, the smaller number of subjects with specific causes of death limited further disease-specific mortality analyses.
The strengths of this study are its large sample size, inclusion of participants with diverse race and socioeconomic status, and availability of a wide array of data on epidemiologic, environmental, clinical (e.g., high-sensitivity CRP) and genomic data. Our results should be applicable to similar urban dwelling communities. Compared to the NHANES study which was collected between 1988 and 1994 and included rural and urban communities, our study used recently collected data (2004–2013), and was comprised of younger adults from a single urban study site, minimizing unmeasured confounding by geographic localities. Previous studies of RDW were conducted mostly in the elderly, and individuals of European ancestry [
11]. By focusing on racially diverse younger individuals who are at risk of health disparity, we showed that the adverse effects of elevated RDW could be seen earlier in the life course.
Authors’ contributions
MKE and ABZ acquired funding and are co-principal investigators of the study. SMT conducted all statistical analyses. MAN participated in acquisition of genetic data. All authors contributed to writing and revision of the manuscript, and approved the final version for publication. All authors read and approved the final manuscript.