|Home | About | Journals | Submit | Contact Us | Français|
Mildly low hemoglobin concentration is associated with increased mortality in older adults. However, this relationship has not been well characterized in racial/ethnic minorities. Therefore, this study determined the hemoglobin threshold below which risk of death is significantly increased in older non-Hispanic whites, non-Hispanic blacks, and Mexican Americans. Data on 4,089 participants of the 1988-1994 US National Health and Nutrition Examination Survey who were ≥65 years of age were analyzed with mortality follow-up through December 31, 2000. Mean hemoglobin in non-Hispanic whites (n=2,686) and Mexican Americans (n=663) was 140 g/L, while in non-Hispanic blacks (n=740) the mean was 10 g/L lower. A total of 1,944 (47.5%) participants died. Among non-Hispanic whites and Mexican Americans, age- and sex-adjusted models showed that the hemoglobin thresholds below which mortality risk was significantly increased were 4 g/L and 2 g/L, respectively, above the World Health Organization (WHO) cutoffs for anaemia. In contrast, the threshold for non-Hispanic blacks was 7 g/L below the WHO criteria. Similar threshold effects were observed when analyzing hemoglobin in categories and adjusting for multiple confounders. In conclusion, the hemoglobin threshold below which mortality rises significantly is a full g/dL lower in non-Hispanic blacks than in non-Hispanic whites and Mexican Americans.
Accumulating evidence indicates that anaemia in older adults is a potent risk factor for a variety of adverse outcomes, including cardiovascular events, recurrent falls, hospitalizations, poor physical function, and mortality (Chaves, et al 2004, Culleton, et al 2006, Denny, et al 2006, Dong, et al 2008, Izaks, et al 1999, Patel 2008, Patel, et al 2007, Penninx, et al 2003, Penninx, et al 2004, Penninx, et al 2006, Penninx, et al 2005, Zakai, et al 2005). Prevalence of anaemia in community-dwelling older adults is more than 10% and increases with advancing age, exceeding 20% in those aged 85 and older (Guralnik, et al 2004, Salive, et al 1992). Importantly, the majority of anaemia cases in the community are mild, with less than 3% of the non-institutionalized older population having hemoglobin concentrations below 110 g/L (11.0 g/dL) (Guralnik, et al 2004). However, the hemoglobin threshold at which clinicians should be concerned for their patient's risk for adverse events has not been clearly established (Steensma and Tefferi 2007).
The World Health Organization (WHO) criteria (hemoglobin concentration <120 g/L in women and <130 g/L in men) are often used to define anaemia for clinical and epidemiologic purposes, but this definition has been criticized because it was established in 1968 using limited hemoglobin data (Beutler and Waalen 2006, Spivak 2005). To improve upon this, researchers have recently analyzed large databases to define anaemia by identifying the lower boundary of normal values based on the statistical distribution of hemoglobin after excluding sick individuals (Beutler and Waalen 2006). Others have taken an outcomes-based approach to identify abnormal hemoglobin values (Chaves, et al 2004, Culleton, et al 2006, Denny, et al 2006, Izaks, et al 1999, Patel, et al 2007, Penninx, et al 2006, Zakai, et al 2005). Both methods by and large suggest that the WHO criteria provide a slightly conservative definition of anaemia for the general older population, having cutpoints that are too low.
There are, however, significant racial differences in the distribution of hemoglobin in the United States. Numerous community-based studies have demonstrated that hemoglobin levels are lower in blacks compared to whites, even among children and healthy young adults (Bao, et al 1993, Denny, et al 2006, Johnson-Spear and Yip 1994, Pan and Habicht 1991, Patel, et al 2007, Perry, et al 1992, Reed and Diehl 1991, Zakai, et al 2005). In older adults, the prevalence of WHO-defined anaemia is 2 to 3 times higher in blacks (18-39%) than in whites (7-17%) (Denny, et al 2006, Dong, et al 2008, Guralnik, et al 2004, Patel, et al 2007, Penninx, et al 2006, Zakai, et al 2005). Racial differences in hemoglobin values range from 4 g/L to 14 g/L and persist after accounting for socioeconomic status, health behaviors, nutritional status, and multiple morbidities (Bao, et al 1993, Johnson-Spear and Yip 1994, Pan and Habicht 1991, Perry, et al 1992). As a result, a number of expert groups have proposed different criteria to define anaemia separately by race/ethnicity (Beutler and Waalen 2006, Earl and Woteki 1993, INACG 2002, Robins and Blum 2007, WHO/UNICEF/UNU 2001). However, few outcomes-based studies have carefully evaluated whether race-specific criteria are justified.
Given that anaemia is commonly encountered in geriatric practice and is no longer viewed as an “innocent bystander” (Nissenson, et al 2003), the current study determined the hemoglobin threshold below which risk of death is significantly increased in older non-Hispanic whites, non-Hispanic blacks, and Mexican Americans. In addition, the distributions of common causes or subtypes of anaemia (e.g., anaemia associated with nutrient deficiency, chronic kidney disease, inflammation) were examined by race/ethnicity. Mortality risks associated with the subtypes of anaemia were examined as well.
Data from the Third National Health and Nutrition Examination Survey, 1988-94 (NHANES III) were analyzed. The NHANES III used a complex multistage sampling design to provide a nationally representative sample of the civilian noninstitutionalized US population, ages 2 months and older (i.e., no upper age limit) (NCHS 1994). Children, older adults, African Americans, and Mexican Americans were oversampled to provide reliable estimates of health conditions in these subpopulations. All participants provided written informed consent and then underwent a home interview followed by a physical examination either in a mobile examination center or at home. Examinations included phlebotomy for those aged 1 year and older. There were 2 phases of data collection, each providing a nationally representative sample. Complete blood counts were collected in both phases of data collection, but the requisite tests for determining the subtypes of anaemia were performed in Phase 2 (1991-94).
Among 5,252 NHANES III participants aged 65 and older, 4,199 had hemoglobin values available for analysis. An additional 109 persons were excluded who did not identify their race/ethnicity as either non-Hispanic white, non-Hispanic black, or Mexican American. In one participant, vital status tracing was not possible. Therefore, the analytic sample size for the current study was 4,089 older adults; for analyses restricted to Phase II, 1,954 older adults were available for analysis. Vital status was determined by a search of the National Death Index through December 31, 2000. There were 1,944 deaths over the 12 year follow-up period (median follow-up was 7.3 years). Participants missing hemoglobin 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 hemoglobin values.
The laboratory methods used in NHANES III have been reviewed and described in detail in previous publications (Gunter EW 1996, Guralnik, et al 2004). Briefly, hemoglobin concentration was measured using a Coulter S-Plus Jr electronic counter (Coulter Electronics, Hialeah, FL) that was calibrated daily. Colorimetric methods were used to determine serum iron and total iron-binding capacity (tranferrin; Alpkem RFA analyzer, Clackamas, OR). The Quantimmune Ferritin Kit (Bio-Rad Laboratories, Hercules, CA) was used to measure serum ferritin. Fluorescence extraction was applied to measure free erythrocyte protoporphyrin (Gunter, et al 1989). The Bio-Rad Laboratories Quantaphase Folate radioassay kit was used to determine folate and vitamin B12. Red blood cell (RBC) folate was calculated using the following formula after a 1:22 dilution of the whole blood: RBC folate = [(whole blood folate × 22) − serum folate (1 − hematocrit/100)]/[hematocrit/100]. For participants who underwent a home examination, only serum folate was able to be determined.
Serum creatinine was determined by a kinetic rate Jaffe method using an autoanalyzer (Hitachi model 737, Boehringer Mannheim Diagnostics, Indianapolis, IN). As recommended, 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)] (Selvin, et al 2007). Estimated glomerular filtration rate (eGFR) was calculated using the following abbreviated Modification of Diet in Renal Disease (MDRD) Study formula expressed for standardized creatinine: eGFR = [175 × (standardized creatinine)-1.154 × (age)-0.203 × (0.742 if the participant is female) × (1.212 if the participant is black)] (Levey, et al 2006). C-reactive protein (CRP) level was measured using latex-enhanced nephelometry (Behring Diagnostics, Somerville, NJ).
Anaemia was defined using the WHO criteria of hemoglobin concentration <120 g/L in women and <130 g/L in men as well as by race/ethnicity specific criteria based on survival analyses presented in Figure 2. Participants were considered iron deficient if at least 2 of the 3 following criteria were met: transferrin saturation < 15%, serum ferritin < 12 ng/mL, and erythrocyte protoporphyrin > 1.24 μM (Looker, et al 1997). Vitamin B12 deficiency was defined as serum B12 < 147.56 pM (Gunter EW 1996). Folate deficiency was defined as RBC folate < 232.49 nM among participants examined in the mobile examination center, while serum folate < 5.89 nM was used to identify deficiency in those examined at home (Gunter EW 1996).
Anemic participants who were not deficient in iron, folate, or B12 were evaluated for other causes of anaemia. Estimated GFR < 60 mL/min/1.73m2 was used to identify those with anaemia associated with chronic kidney disease (Astor, et al 2002). Anaemia of chronic inflammation, characterized by reduced circulating serum iron levels but adequate iron storage, was defined as serum iron < 10.74 μM and no evidence of iron deficiency. If participants with anaemia were not classified with nutrient deficiency, chronic kidney disease, or anaemia of chronic inflammation, then they were considered to have unexplained anaemia.
A wide range of demographic and health status variables were collected during the home interview. Participants were asked to self-identify their race and ethnicity. Socioeconomic position was assessed in terms of the highest grade of education completed and by the poverty income ratio, defined as household income divided by the US poverty level according to household size and area of residence. A standard set of questions were used to classify participants as never, former, or current smokers. Body mass index (BMI) was calculated using measured height and weight (BMI = weight in kilograms / height in meters squared). Medical conditions were assessed by asking participants if a doctor has 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), rheumatoid arthritis, and stroke. In addition, participants were asked if they had difficulty walking a quarter of a mile or walking up to 10 stairs without resting. If participants reported much difficulty or were unable to do these activities, then they were considered to have mobility limitations. Finally, overnight hospitalization stays within the past year were also assessed through self-report.
Descriptive statistics (means and proportions) were used to examine the baseline characteristics of the study sample by race/ethnicity (Table 1). Hemoglobin concentration was analyzed both as a continuous and categorical variable. Six hemoglobin categories were specified a priori, 3 below the WHO cutoff (hemoglobin > 10.0 g/L, 5.1-10.0 g/L, and 0.1-5.0 g/L below the WHO cutoff) and 3 above the WHO cutoff (hemoglobin 0.0-9.9 g/L, 10.0-19.9 g/L, and ≥20.0 g/L above the WHO cutoff). For all survival analyses, the length of follow-up was calculated from the date of examination when blood was collected to the date of death or the end of follow-up on December 31, 2000. All analyses were stratified by race/ethnicity because there were significant interactions of race (non-Hispanic black vs. non-Hispanic white) with WHO-defined anaemia as well as with the categories of hemoglobin in predicting mortality; however, there were no statistically significant differences between Mexican Americans and non-Hispanic whites or Mexican Americans and non-Hispanic blacks in mortality risk according to WHO-defined anaemia or categories of hemoglobin concentration.
Kaplan-Meier survival curves were plotted by categories of hemoglobin concentration for each racial/ethnic group (Figure 1). Cox proportional hazard models were then used to assess the association of hemoglobin level with mortality (Table 2). The first model adjusted for age and sex, while the second adjusted for all potential confounding factors. Plots of Schoenfeld residuals were examined to confirm the proportional hazards assumption. To examine the non-linear effects of hemoglobin concentration and to identify the threshold below which mortality risk is increased, proportional hazard models were fitted with a penalized spline for hemoglobin concentration (centered on the WHO criteria cutoffs) adjusting for age and sex using R statistical software (Figure 2) (Team 2007). Finally, among Phase II participants, the distribution of common causes or subtypes of anaemia were compared by race/ethnicity using chi square statistics in those with anaemia defined by the WHO criteria and by race/ethnicity specific criteria identified by mortality risks (Table 3). Proportional hazard models were also used to determine whether each of the underlying causes or subtypes of anaemia is associated with an increased risk of death (Table 4). Using indicator variables to compare non-anaemia to each of the anaemia subtypes, a basic model was first fitted adjusting for age, sex, and race/ethnicity and then a fully adjusted model was estimated. Unless noted otherwise, all analyses were weighted and accounted for the complex sampling methods and non-response bias to provide estimates for the US population using Stata version 10.
The average age of the study sample was approximately 73 years and slightly more than half were women (Table 1). Non-Hispanic blacks and Mexican Americans had substantially lower levels of education and income compared to non-Hispanic whites. Hemoglobin concentration was on average 10 g/L lower in non-Hispanic blacks than in non-Hispanic whites and Mexican Americans. The prevalence of current smoking and the mean BMI was higher in non-Hispanic blacks, although non-Hispanic whites and Mexican Americans were more likely to be former smokers. Congestive heart failure, diabetes, mobility limitations, and overnight hospitalizations were more common in non-Hispanic blacks and Mexican Americans than in non-Hispanic whites; however, this pattern was reversed for cancer and pulmonary diseases.
The association of hemoglobin level with mortality is shown in Figure 1. Among older non-Hispanic whites, the survival curves clearly separate with lower hemoglobin concentration associated with poorer survival that is sustained over time. In contrast, a clear survival gradient across hemoglobin levels was not seen in older non-Hispanic blacks, but rather poor survival was only observed in those with hemoglobin >10.0 g/L below the WHO cutoff. Importantly, non-Hispanic blacks with hemoglobin 0.1-5.0 g/L below the WHO cutoff had a survival pattern similar to those with hemoglobin levels above the WHO cutoff. Although the survival curves were less stable in older Mexican Americans because of smaller sample sizes and lower event rates, survival was worse in those with lower hemoglobin concentration.
After adjusting for age and sex, there was a stepwise increased risk of death associated with lower hemoglobin concentration in non-Hispanic whites relative to those in the reference category of hemoglobin 10.0-19.9 g/L above the WHO cutoff (Table 2). There was even a significant 27% increased mortality risk in those with a hemoglobin 0.0-9.9 g/L above the WHO cutoff. For non-Hispanic blacks, mortality risk was flat until hemoglobin was down to 5.1-10.0 g/L below the WHO cutoff, but only became clearly significant for those >10.0 g/L below the WHO cutoff compared to the reference group. Mexican Americans had a stepwise increased risk for mortality associated with lower hemoglobin level that was similar to the pattern observed in non-Hispanic whites, although mortality risk was nearly 5-fold higher in Mexican Americans with hemoglobin >10.0 below the WHO cutoff relative to the reference group. Further adjustment for socioeconomic factors, health behavior, and health status did not substantively change the patterns of association (Table 2). Consistent with results in Table 2, the non-linear association between hemoglobin and mortality adjusted for age and sex is shown in Figure 2. The hemoglobin threshold below which risk of death increased significantly was 4 g/L and 2 g/L above the WHO cutoff in non-Hispanic whites and Mexican Americans, respectively. In non-Hispanic blacks, however, the threshold was 7 g/L below the WHO cutoff.
Table 3 shows the distribution of the subtypes of anaemia by race/ethnicity. Among those with WHO defined anaemia, nutrient deficiency ranged between 33% and 41% across the 3 racial/ethnic groups. Iron deficiency anaemia was higher among Mexican Americans but not statistically different (p=0.12); whereas folate deficiency was significantly more common in non-Hispanic blacks (p<0.01). The proportion of anemic non-Hispanic whites and Mexican Americans with chronic kidney disease (eGFR<60 mL/min/1.73m2) alone was twice as high as compared to non-Hispanic blacks, but these differences were not statistically different (p=0.10). In contrast, the proportion with anaemia of chronic inflammation alone was more than doubled in non-Hispanic blacks relative to non-Hispanic whites and Mexican Americans (p<0.01). Unexplained anaemia did not vary by race/ethnicity (p=0.81), but was less common in Mexican Americans than in non-Hispanic whites or non-Hispanic blacks. Interestingly, when ethnicity-specific hemoglobin cutoffs (identified from survival analyses in Figure 2) were used to define anaemia instead of the WHO criteria (Table 3), the distribution of anaemia subtypes did not substantially change even though the number of anaemia cases increased in non-Hispanic whites (from 138 to 210) and decreased in non-Hispanic blacks (from 86 to 45).
The association of anaemia subtypes with mortality is shown in Table 4. Relative to those who were non-anemic according to WHO criteria, anaemia in participants with nutrient deficiency and ACI was associated with a significantly increased risk of death adjusting for age, sex, and race/ethnicity. In contrast, when ethnicity-specific criteria were applied to define anaemia, each subtype was a significant predictor of mortality except for unexplained anaemia (p = 0.06) adjusting for age, sex, and race/ethnicity. Further adjustment for potential confounding factors generally reduced the strengths of association.
In a nationally representative cohort of community-dwelling older adults, the current study finds that the hemoglobin threshold below which mortality risk is significantly increased varies by race/ethnicity. Among non-Hispanic whites and Mexican Americans, the hemoglobin thresholds were 4 g/L and 2 g/L, respectively, above the WHO cutoffs for defining anaemia, while in non-Hispanic blacks the threshold was 7 g/L below the WHO criteria. These differences are remarkably consistent with known racial/ethnic differences in the distribution of hemoglobin concentration. In the NHANES III, hemoglobin was, on average, 10 g/L lower in older non-Hispanic blacks than in non-Hispanic whites, which was nearly the same magnitude of difference in hemoglobin thresholds for increased risk of death. Importantly, the distribution of common causes of anaemia generally did not differ by race/ethnicity and each anaemia subtype was associated with higher mortality compared to non-anaemic persons. Taken together, these findings indicate that racial differences in the hemoglobin threshold for mortality are not explained by differences in the distribution of common causes of anaemia in older adults. Thus, these findings underscore the need to clinically address mild cases of anaemia in older adults and provide outcomes-based evidence that a revised definition of anaemia is needed that takes race into account.
Racial differences in hemoglobin distribution are believed to result, in part, from strong genetic pressure exerted by endemic malaria in various regions of Africa. For instance, certain hemoglobinopathies, such as α-thalassemia, that reduce the burden of malaria infection, occur more frequently in blacks than in whites. Indeed, Beutler and West (Beutler and West 2005) showed that as much as one-third of the racial difference in hemoglobin concentration might be explained by α-thalassemia. Given that hemoglobin concentration is heritable (heritability estimates range 0.37 to 0.59) (Evans, et al 1999, Garner, et al 2000, Sala, et al 2008), it is conceivable that other genetic adaptations to environmental demand inside and outside of Africa contribute to racial differences in hemoglobin set point.
Few studies have prospectively assessed the effect of hemoglobin or anaemia on adverse outcomes separately by race or ethnicity. In a cohort of older adults living in North Carolina, Denny and colleagues (2006) showed that WHO-defined anaemia was associated with an increased risk of death in both black and white participants, but hemoglobin thresholds for predicting mortality were not evaluated by race. In contrast, a study of relatively healthy and well-functioning older adults showed that WHO-defined anaemia significantly predicted mortality over a 6-year period in whites, but not in blacks (Patel, et al 2007). Although there was some indication that mortality risk increased at a lower hemoglobin threshold in black men than in white men, there were not enough deaths in the study to determine stable threshold effects by race and sex (Patel, et al 2007). More recently, a lower hemoglobin threshold for mortality risk was observed in community-dwelling older blacks than in whites living in Chicago (Dong, et al 2008).
A limitation of the current study and others in the literature is that hemoglobin was assessed on a single occasion. While it is likely that low hemoglobin in some participants would have resolved 3 to 6 months later, this is unlikely to explain the survival gradients shown in Figure 1. A second limitation is that there were not enough racial/ethnic minority participants available to further stratify the analyses by sex, although all analyses were adjusted for sex and there were no significant sex by hemoglobin interactions. The major strength of the study is that the sample is representative of community-dwelling older adults in the United States. In addition, the NHANES III cohort was followed-up for mortality over a long time period and the study participants were well characterized to permit adjustment for multiple potential confounding factors. Finally, unlike any previous large, population-based study of anaemia in older adults with long-term mortality follow-up, the laboratory testing in NHANES III permitted classification of subjects into common subtypes of anaemia. Although more up-to-date laboratory measures, including tests for methylmalonic acid and serum transferrin receptor levels, and a more in-depth clinical hematologic work-up would have improved this classification, the distribution of anaemia subtypes did not vary substantially by race/ethnicity.
In conclusion, anaemia is a multifactorial condition that is associated with a variety of adverse outcomes in older adults. While further research is needed to elucidate the pathophysiology of unexplained anaemia cases in late life (Longo 2005), results from the current study demonstrate that the major subtypes of anaemia are associated with increased risk of death. The majority of anaemia cases in older adults are likely amenable to therapy, but whether correcting mild anaemia improves function and survival will require randomized controlled trials. It is imperative then for clinical and research purposes that a clear case definition of anaemia be established. In view of the NHANES III mortality results as well as the extant literature documenting racial differences in hemoglobin distribution, a revised definition of anaemia that takes race into account should be considered. To establish the hemoglobin cutoffs based on risk for adverse events, there must be consistency across studies and therefore additional epidemiologic research in racially/ethnically diverse cohorts is required.
This study was supported by the Intramural Research Program of the US National Institute on Aging, National Institutes of Health.
Disclosure and funding: The authors declare no commercial or financial conflict of interests in publishing the study results. Funding support for this study was provided by the Intramural Research Program of the US National Institute on Aging, National Institutes of Health.