Author + information
- Received September 23, 2002
- Revision received February 6, 2003
- Accepted February 20, 2003
- Published online June 4, 2003.
- ↵*Reprint requests and correspondence:
Dr. Dariush Mozaffarian, VA Puget Sound Health Care System, 1660 South Columbian Way, MS 152, Seattle, Washington 98108, USA.
Objectives Our aim was to examine the relationships between serum hematocrit (Hct) and risk of all-cause mortality among patients with severe heart failure (HF).
Background Anemia occurs with increased frequency in severe HF. However, few studies have examined the impact of anemia on mortality in this population.
Methods Using a prospective cohort design, we evaluated the relationships between baseline serum Hct and mortality among 1,130 patients with left ventricular EF <30% and New York Heart Association functional class IIIB or IV HF treated with angiotensin-converting enzyme inhibitors, diuretics, and digitalis. Mortality was ascertained by centralized adjudication.
Results The mean Hct was 41.8% (range 25.4% to 58.8%). Over 15 months of mean follow-up, there were 407 deaths (29 per 100 person-years). After adjustment for potential confounders, those in the lowest quintile of Hct (range 25.4% to 37.5%) had a 52% higher risk of death (hazard ratio 1.52, 95% confidence interval 1.11 to 2.10), compared with the highest quintile (range 46.1% to 58.8%). Within the lowest quintile of Hct, each 1% decrease in Hct was associated with an 11% higher risk of death (p < 0.01), whereas within the four higher quintiles of Hct, Hct was not associated with total mortality. Evaluation of different causes of death indicated that a lower Hct was strongly associated with death from progressive HF, rather than sudden death or other deaths.
Conclusions Among patients with severe HF, anemia is a significant independent risk factor for death, with a progressively higher risk with increasing severity of anemia. Further investigation of the etiologies, prevention, and treatment of anemia in severe HF is warranted.
While mortality from coronary artery disease and stroke has been declining, the incidence and health burden of congestive heart failure (HF) continue to rise (1). In the U.S., nearly 5 million people have HF and more than half a million new cases are diagnosed each year; HF is the most common cause of hospitalization among persons age ≥65 years; and health care costs for HF exceed $20 billion annually (1,2). Despite advances in medical treatment, HF patients have high mortality rates, and HF deaths have increased 145% over the last two decades (1). People with severely reduced left ventricular systolic function and severe HF symptoms are at particular risk, with mortality rates approaching 30% per year. These trends, mirrored in other industrialized nations (3–5), accentuate the importance of identifying and treating novel risk factors for poor outcomes in HF.
Accumulating evidence suggests that anemia may be an independent risk factor for mortality among patients with HF (6–9). Additionally, clinical studies among small numbers of HF patients (n = 26 and n = 32 ) indicate that treatment of anemia with erythropoietin and iron improves symptoms and the ejection fraction (EF) and decreases hospitalizations and the need for diuretics, suggesting that anemia may be a modifiable risk factor in HF with a causal role in clinical outcomes. However, few previous studies have characterized the impact of anemia on HF mortality, such as the magnitude of risk, threshold of risk, or associations with different causes of death, and only one previous study (9)has examined patients with severe HF, the population at highest risk for both anemia and death.
We therefore investigated, using a prospective cohort design, the relationships between baseline serum hematocrit (Hct) and mortality among 1,130 subjects enrolled in the Prospective Randomized Amlodipine Survival Evaluation (PRAISE), a randomized trial of amlodipine versus placebo among patients with severe HF. Our primary hypothesis was that a lower Hct would be associated with a higher risk of all-cause mortality among these patients with severe HF.
Study design and population
We used baseline information and follow-up data from PRAISE, a multicenter, randomized clinical trial of amlodipine versus placebo among 1,153 men and women with left ventricular EF <30% and New York Heart Association (NYHA) functional class IIIB or IV HF symptoms treated with angiotensin-converting enzyme inhibitors, diuretics, and digitalis. Patients with both ischemic and nonischemic HF were enrolled. The design, inclusion and exclusion criteria, and primary results have been previously described (12). The protocol was approved by the institutional review boards of all 105 participating institutions, and informed written consent was obtained from all subjects. Laboratory analyses, including Hct, were performed at a central laboratory (Hct normal range = 41% to 50% in men and 33% to 46% in women). We excluded 23 participants with missing baseline Hct measurements, resulting in 1,130 participants included in this analysis.
Ascertainment and classification of mortality
All deaths were reviewed and classified by a centralized adjudication committee based on information from hospital and emergency room records, electrocardiograms, chest X-rays, autopsy reports, death certificates, and statements from witnesses (12,13). Cardiac deaths were categorized as sudden death, pump failure death, or fatal myocardial infarction (MI). Sudden death was defined as death from cardiac or unknown causes that was unexpected and occurred after an abrupt loss of consciousness in a previously stable patient. Pump failure death was defined as progressive HF culminating in death not due to acute MI or observed or suspected life-threatening arrhythmia; heart transplantations (n = 8) were included as pump failure deaths. Deaths >60 min from the onset of symptoms for hospitalized acute MI were classified as fatal MI. The primary outcome of this analysis was all-cause mortality. We also evaluated different causes of death, such as pump failure death. No patients were lost to follow-up.
Hematocrit was evaluated as an indicator variable (dummy variable) in quintiles and deciles, and as a continuous variable. Baseline characteristics were compared across quintiles of Hct using regression, with the characteristic as the dependent variable and Hct quintile as the independent variable. Kaplan-Meier estimates were used to evaluate survival over time, with differences evaluated using the log-rank test for equality of survivor functions. Cox proportional hazards models were used to evaluate risk, with multivariate Cox models evaluated to assess for confounders. Age, gender, HF etiology, and NYHA functional class were included in an initial multivariate Cox model. Other characteristics were added, both individually and in groups, and retained in the final model based on clinical interest or an appreciable change (±5%) in the hazard ratio (HR) associated with Hct. In addition to the characteristics in the final model (age, gender, diabetes, smoking, HF etiology, EF, NYHA functional class, systolic blood pressure, white blood cell count, and creatinine), we also evaluated for potential confounding by race, heart rate, diastolic blood pressure, cardiothoracic ratio, furosemide use and dose, metolazone use and dose, potassium-sparing diuretic use, treatment assignment (amlodipine or placebo), allopurinol use, uric acid, serum cholesterol, triglycerides, estimated creatinine clearance, blood urea nitrogen, sodium, and lymphocyte count. Likelihood ratio testing was used to assess for the effect modification by age, gender, diabetes, smoking, NYHA functional class, and etiology of HF. All p values were two-tailed. Analyses were performed using Stata version 6.0 (Stata Corp., College Station, Texas).
Baseline characteristics overall and according to quintiles of Hct are presented in Table 1. The mean Hct was 41.8% (range 25.4% to 58.8%). The average age of the participants was 65 years at baseline. Three-quarters of participants were male, and more than one-third were diabetic. The mean EF was 21%, with 81% of participants with NYHA functional class IIIB and 19% with class IV symptoms. A higher Hct value was associated with younger age, male gender, more prevalent smoking, slightly lower EF, and higher blood pressure. Hematocrit was also positively associated with the white blood cell count, mean corpuscular volume, indexes of liver function, and serum cholesterol and inversely associated with blood urea nitrogen and creatinine.
Over 15 months of mean follow-up, there were 407 deaths (29 deaths per 100 person-years). Cardiac deaths accounted for the vast majority (87%) of fatalities, including 182 sudden deaths, 162 pump failure deaths, and 12 fatal MIs. In Kaplan-Meier survival analyses, those in the lowest quintile of Hct had significantly poorer survival, whereas survival in the four higher quintiles was relatively similar (Fig. 1). After adjustment for age, gender, diabetes, smoking, HF etiology, EF, NYHA functional class, systolic blood pressure, white blood cell count, and creatinine, those in the lowest quintile of Hct (range 25.4% to 37.5%) had a 52% higher risk of death (HR 1.52, 95% confidence interval [CI] 1.11 to 2.10), compared with the highest quintile of Hct (range 46.1% to 58.8%) (Table 2). In contrast, subjects in the second, third, and fourth quintiles of Hct did not have significantly different mortality compared with those in the highest quintile (Table 2). When Hct was evaluated in deciles, there was significantly higher mortality only in the lowest decile (Hct <35.4%) and no significant associations in the second to ninth deciles, compared with the highest decile (Hct >48.3%) as the reference group (Fig. 2).
There also appeared to be a threshold of risk when Hct was evaluated continuously. After adjustment for potential confounders (Table 2, footnote), each 1% decrease in Hct was associated with a 3% higher risk of death (HR 1.03, 95% CI 1.01 to 1.05, p < 0.01) in the overall cohort. However, this association was entirely due to a higher risk in the lowest quintile of Hct (range 25.4% to 37.5%), within which each 1% decrease in Hct was associated with an 11% higher risk of death (HR 1.11, 95% CI 1.02 to 1.20, p < 0.01). In contrast, there was no association between each 1% decrease in Hct and total mortality (HR 1.01, 95% CI 0.98 to 1.04, p = 0.58) within the four higher quintiles (Hct >37.5%).
Because women normally have lower Hct values than men, we also evaluated associations with quintiles of Hct derived separately for men versus women (Table 3). After adjustment for potential confounders, both men and women in the lowest quintile of Hct had an approximately 60% higher risk of mortality, compared with those in the highest quintile, although the power was limited to confirm this finding among women because of fewer numbers of female participants.
When we evaluated different causes of death, there was a strong inverse association between Hct and pump failure death, with a significantly higher risk as Hct decreased across quintiles (p for trend <0.001) (Table 4, Fig. 3). In similar analyses with Hct evaluated continuously, each 1% decrease in Hct was associated with an 8% higher risk of pump failure death (HR 1.08, 95% CI 1.05 to 1.12, p < 0.001). In contrast, there was a suggestion of a U-shaped relationship with other causes of death (e.g., sudden death and other deaths), with trends toward a higher risk among those in the lowest and highest quintiles of Hct (Table 4, Fig. 3). We evaluated this possible U-shaped relationship by adding a squared Hct term to the model: both the term (p = 0.03) and the likelihood ratio test (p = 0.04) were significant. In continuous analyses, the risk of these non-pump failure deaths was greatest for each 1% decrease in Hct below 36% (HR 1.18, 95% CI 1.01 to 1.37, p = 0.04; n = 138) and each 1% increase in Hct above 47% (HR 1.12, 95% CI 1.00 to 1.26, p = 0.06; n = 180); there was no relationship with non-pump failure deaths when Hct was between 36% and 47% (HR 1.00, 95% CI 0.94 to 1.05, p = 0.88; n = 812). Evaluation of sudden deaths and other deaths separately did not greatly alter these results, although the CIs were broader due to fewer total events.
There was little evidence that the associations between Hct and mortality varied according to age, gender, diabetes, smoking, NYHA functional class, or etiology of HF (p > 0.05 by the likelihood ratio test for each interaction). All findings were similar if hemoglobin was used instead of Hct.
Among these patients with severe HF, we observed a significant and independent association between anemia and total mortality, with progressively higher mortality with increasing severity of anemia. Hematocrit was not associated with total mortality when values were within or near normal ranges, suggesting a potential threshold of risk rather than a linear or graded relationship. The results were similar whether Hct was evaluated in quintiles, in deciles, or continuously. The magnitude of the increased risk is striking, with a 52% higher risk of death in the lowest quintile of Hct compared with the highest quintile. This increased risk is more than twice the risk associated with diabetes, smoking, a decade of age, or a 10% difference in EF in this population.
There are several possible explanations for these findings. Although the associations persisted after adjustment for a wide variety of demographic, clinical, and laboratory characteristics, a lower Hct value may be a marker for some other factor that increases mortality among patients with severe HF. For example, anemia may be a marker for poor renal function or volume overload. However, there were only modest differences in blood urea nitrogen and serum creatinine associated with Hct, and no significant differences in NYHA functional class, diuretic use, or serum sodium. Moreover, the higher risk persisted after adjustment for these factors, suggesting that baseline differences in these characteristics did not entirely mediate the relationship between anemia and mortality.
Anemia may also be a marker for higher circulating cytokines and chemokines, which are associated with both anemia of chronic disease and higher mortality in HF (14–16). Lower Hct levels were associated with lower white blood cell counts, suggesting possible generalized bone marrow hypoproduction; decreased production may also extend to other organs, as suggested by liver function indexes. Systemic inflammation and immune activation in HF may be related to bacterial endotoxin translocation resulting from bowel wall edema and altered gut permeability (17); interestingly, from our clinical experience, HF patients with suspected bowel edema often show evidence of poor iron absorption, which may further exacerbate anemia.
There are also plausible biologic mechanisms for a causal relationship between anemia and mortality in HF. Myocardial ischemia may result from a reduced oxygen-carrying capacity combined with a low EF and increased wall stress. Among elderly patients with acute MI, blood transfusion is associated with lower mortality when the initial Hct is ≤33%, but not when the initial Hct is higher (18), consistent with a relative threshold of ischemia risk. Another plausible biologic mechanism is chronically increased myocardial work and adrenergic stimulation due to a diminished oxygen-carrying capacity of anemic blood, resulting in progressive HF. This hypothesis is supported by the strong inverse association between Hct and death due to progressive HF, rather than sudden death or other deaths. The resting cardiac output begins to increase with hemoglobin <10 g/dl (≈ Hct <30 to 33%) (19), and this threshold is likely higher in the setting of abnormal states such as severe HF. Anemia independently predicts the development of HF among both renal failure patients and renal transplant recipients, independent of ischemic heart disease (20,21). Additionally, anemia is associated with elevated plasma catecholamines and alpha2-receptor densities among renal failure patients, proportional to the degree of anemia, and correction of anemia with erythropoietin reduces plasma noradrenaline and alpha2-receptor density (22). Correction of anemia reduces the cardiac output among renal failure patients (23,24), and improves the EF and reduces symptoms and hospitalizations among HF patients (10,11), further supporting the hemodynamic significance of anemia.
Shape of risk
We observed a threshold of risk for total mortality, due to a graded inverse relationship with pump failure death and a U-shaped relationship with other causes of death (largely sudden death). Among patients referred for heart transplant evaluation, Hct was inversely associated with total mortality, largely due to the risk of progressive HF (9). The relationships with non-pump failure deaths did not appear to be U-shaped (G. C. Fonarow, personal communication, September 2002), as suggested by our study. Further investigation is necessary to determine whether there is a threshold of risk, a graded relationship, or a U-shaped relationship between Hct and mortality in HF, taking into consideration that this relationship may differ depending on the cause of death, as suggested by our results.
Our analysis has several strengths. Patients were enrolled from multiple centers, rather than from a single HF or transplant clinic, which increases the generalizability. All patients were receiving similar background medical therapy, including angiotensin-converting enzyme inhibitors, diuretics, and digitalis. Demographic, clinical, and laboratory characteristics were well characterized using standardized techniques, increasing the capacity to adjust for confounders. The predominant mortality end point was death (98%), rather than transplantation (2%). Comprehensive follow-up, review of events, and centralized adjudication minimized the potential for missed or misclassified outcomes.
There are also potential limitations to our findings. Hematocrit was assessed at baseline and may have changed over time; such misclassification might cause underestimation of the associations between Hct and mortality. Also, although we adjusted for a wide variety of participant characteristics, residual confounding due to unmeasured or incompletely measured factors cannot be excluded. These patients were participants in a randomized clinical trial, so the results may not be generalizable to all patients with severe HF. We did not have information on levels of iron, erythropoietin, or cytokines and chemokines, so we could not differentiate etiologies of anemia. As this was a completed study and no further enrollment was possible, we did not calculate the power a priori; post hoc calculations showed >80% power to detect a 33% increase in mortality risk across the quintiles of Hct.
Our findings indicate that anemia is a significant independent risk factor for death among patients with severe HF, particularly death due to progressive pump failure. If this association is causal, normalization of Hct in this population would be expected to reduce mortality by approximately 33% based on the magnitude of risk observed in our study, or one less annual death for every eight patients successfully treated. These results support the need for further investigation of the etiologies, prevention, and treatment of anemia in HF, including appropriately powered randomized clinical trials to determine whether prevention or treatment of anemia reduces mortality in severe HF.
☆ Dr. Mozaffarian was supported by a VA Health Services Research and Development fellowship at the VA Puget Sound Health Care System. The data used in this analysis were collected by the PRAISE Study Group and provided by Pfizer PGRD; no funding support was provided by Pfizer PGRD for this analysis or manuscript.
- confidence interval
- ejection fraction
- heart failure
- hazard ratio
- myocardial infarction
- New York Heart Association
- Prospective Randomized Amlodipine Survival Evaluation
- Received September 23, 2002.
- Revision received February 6, 2003.
- Accepted February 20, 2003.
- American College of Cardiology Foundation
- American Heart Association
- Health Care Financing Administration
- McMurray J.,
- McDonagh T.,
- Morrison C.E.,
- Dargie H.J.
- Reitsma J.B.,
- Mosterd A.,
- de Craen A.J.,
- et al.
- Rodriguez-Artalejo F.,
- Guallar-Castillon P.,
- Banegas Banegas J.R.,
- del Rey Calero J.
- Kannel W.B.
- Al-Ahmad A.,
- Rand W.M.,
- Manjunath G.,
- et al.
- McClellan W.M.,
- Flanders W.D.,
- Langston R.D.,
- Jurkovitz C.,
- Presley R.
- Horwich T.B.,
- Fonarow G.C.,
- Hamilton M.A.,
- MacLellan W.R.,
- Borenstein J.
- Silverberg D.S.,
- Wexler D.,
- Blum M.,
- et al.
- Silverberg D.S.,
- Wexler D.,
- Sheps D.,
- et al.
- Metivier F.,
- Marchais S.J.,
- Guerin A.P.,
- Pannier B.,
- London G.M.
- Rigatto C.,
- Parfrey P.,
- Foley R.,
- Negrijn C.,
- Tribula C.,
- Jeffery J.