The distribution of hematocrit values in the patient sample.
Crude 1-year mortality according to hematocrit categories (≤24%, 46.4%; >24%-28%, 44.9%; >28%-32%, 44.6%; >32%-36%, 39.0%; >36%-40%, 33.9%; >40%-44%, 30.7%; and >44%, 31.5%) (P < .001 for trend) (A) and risk-standardized 1-year mortality according to hematocrit categories (≤24%, 35.9%; >24%-28%, 35.4%; >28%-32%, 38.4%; >32%-36%, 37.6%; >36%-40%, 36.6%; >40%-44%, 35.3%; and >44%, 35.9%) (P = .08 for trend) (B).
Customize your JAMA Network experience by selecting one or more topics from the list below.
Kosiborod M, Curtis JP, Wang Y, et al. Anemia and Outcomes in Patients With Heart Failure: A Study From the National Heart Care Project. Arch Intern Med. 2005;165(19):2237–2244. doi:10.1001/archinte.165.19.2237
Recent reports have suggested that anemia is associated with adverse outcomes in patients with heart failure (HF), but were unable to adjust for a broad range of comorbid conditions. As a result, whether anemia is a truly independent predictor of risk or a marker of comorbid illness in these patients is unknown.
We analyzed medical records from the Centers for Medicare & Medicaid Services’ National Heart Care Project, a national sample of 50 405 patients 65 years and older admitted to acute care hospitals with a principal discharge diagnosis of HF between April 1, 1998, and March 31, 1999, or between July 1, 2000, and June 30, 2001. Multivariable logistic regression analyses were conducted to test whether hematocrit level was an independent predictor of all-cause mortality and HF-related readmission at 1 year.
In unadjusted analysis, lower hematocrit levels were associated with increased 1-year mortality and readmission for HF. Compared with patients with a hematocrit greater than 40% to 44%, those with a hematocrit of 24% or less had a 51% higher risk of death (relative risk [RR], 1.51; 95% confidence interval [CI], 1.35-1.68; P<.001) and a 17% higher risk of HF-related readmission (RR, 1.17; 95% CI, 1.01-1.34; P = .04). However, after adjustment for multiple comorbidities and other clinical factors, the association between lower hematocrit levels and increased 1-year mortality was markedly attenuated, even in those patients with the most severe anemia (hematocrit, ≤24% vs >40%-44%: RR, 1.02; 95% CI, 0.86-1.19; P = .85). The association between lower hematocrit values and HF-related readmission persisted after multivariable adjustment (hematocrit, ≤24% vs >40%-44%: RR, 1.21; 95% CI, 1.04-1.38; P = .01).
Although anemia is an independent predictor of hospital readmission, its relationship with increased mortality in HF patients is largely explained by the severity of comorbid illness. These findings suggest that anemia may be predominantly a marker rather than a mediator of increased mortality risk in older patients with HF.
Recent reports1-9 indicate that anemia may be a modifiable risk factor in patients with heart failure (HF). Given potential treatment implications for patients with HF, the association between anemia and adverse outcomes has been receiving growing attention from expert groups.10 Professional societies, including the Heart Failure Society of America11 and the European Society of Cardiology,12 have recently held symposia dedicated exclusively to this topic, and efforts are being directed toward clinical trials of erythropoietin in anemic patients with HF.13
However, most prior studies were limited by small numbers of patients,1-3,5-8 single-center experience,2 use of administrative claims data,4 and highly selected patient populations that are not representative of the elderly HF patients with extensive comorbidity typically seen in clinical practice.1,2,5,7-9 Furthermore, the ability of previous investigations to adjust for a broad range of non–cardiac-related comorbidities has been limited.1-9 Because of these limitations, the nature of the relationship between anemia and adverse outcomes remains unclear. The central issue is whether anemia is an independent mediator of poor outcomes or a marker of comorbid illness and HF severity.
To address this issue, we analyzed data from the National Heart Care (NHC) Project, a large nationally representative sample of patients in the United States hospitalized with HF from April 1, 1998, to June 30, 2001. Detailed clinical information about these patients, including the presence and severity of comorbid conditions, provided an ideal opportunity to establish whether anemia is an independent predictor of prognosis in this patient population. We specifically focused on the association between various degrees of anemia and the risk of death and readmission at 1 year.
The patient cohort was selected from the NHC Project sample. The NHC Project is a Centers for Medicare & Medicaid Services initiative designed to assess and improve the quality of care for Medicare beneficiaries hospitalized with HF.14 More than 96% of Americans 65 years and older are covered by Medicare,15 and 87% of all Medicare beneficiaries are fee-for-service beneficiaries.16 Medicare fee-for-service beneficiaries hospitalized with a principal discharge diagnosis of HF (International Classification of Diseases, Ninth Revision, Clinical Modification codes 402.01, 402.11, 402.91, 404.01, 404.91, or 428) between April 1, 1998, and March 31, 1999, or between July 1, 2000, and June 30, 2001, inclusive, were identified. It was previously shown that International Classification of Diseases, Ninth Revision, Clinical Modification code 428.x has a specificity for the diagnosis of HF of greater than 95%.17 In each sampling period, a systematic random sample was drawn from eligible discharges after they were grouped by state, and sorted by age, sex, race, and treating hospital. The goal was to randomly select 800 discharges from each state. All records were included when fewer than 800 hospitalizations occurred in a state during a sampling period (Alaska, Hawaii, Idaho, Utah, Vermont, and Wyoming in both samples). Records were selected by using a random start and a sampling fraction calculated from the requisite sample size and the number of eligible discharges in the state.
Medical records were obtained from the treating hospital and underwent detailed review by trained data abstractors in central data abstraction centers. The diagnosis of HF was corroborated by at least 1 of the following criteria: (1) a history of HF documented in the medical record or (2) radiographic evidence of HF on the admission chest radiograph. Data quality was ensured through the use of trained reviewers, medical record abstraction software, and random record reabstraction. In a review of 100 medical records by a cardiologist, only 1 patient was not diagnosed as having HF.14 Patients with invalid Social Security numbers were excluded. The NHC Project sample, thus, consisted of 78 882 records, of which 39 477 were from April 1, 1998, to March 31, 1999, and 39 405 were from July 1, 2000, to June 30, 2001.
To ensure examination of a representative cohort of older patients admitted with HF, we subsequently excluded patients younger than 65 years (n = 6558), those without documented HF on admission (n = 5003), those transferred from another hospital (n = 2419), those undergoing long-term hemodialysis (n = 549), and those with extreme hematocrit values of less than 15% or greater than 70% (n = 58). Patients with a history of severe aortic or mitral stenosis (n = 5741) and those with chronic liver disease/cirrhosis, immunologic suppression, human immunodeficiency virus infection, or malignancy (n = 6216) were also excluded because their prognosis and management were likely to be substantially different from those of most elderly patients with HF. In the event of multiple admissions for the same patient within the sampling period, only 1 admission was chosen at random, thus excluding 3732 admission records. Because we were interested in evaluating the association between hematocrit values and adverse outcomes, patients with a missing admission hematocrit value (n = 3548) or a missing death status (n = 969) were also excluded. In total, 28 477 patients met 1 or more of the previously described exclusion criteria, leaving a final cohort of 50 405 patients for the 1-year mortality analysis. In addition, we excluded patients with missing readmission information (n = 5964) from the HF-related readmission analysis, leaving a final cohort of 44 441 patients for the 1-year readmission analysis.
Admission hematocrit was classified into the following categories: 24% or less, greater than 24% to 28%, greater than 28% to 32%, greater than 32% to 36%, greater than 36% to 40%, greater than 40% to 44%, and greater than 44%. Professional abstractors collected information about demographic variables, non–cardiac- and cardiac-related comorbidities, and admission characteristics, including medications, vital signs, physical examination results, laboratory values, and diagnostic imaging data, from medical records. Left ventricular systolic function was defined as normal or mildly, moderately, or severely depressed based on either quantitative criteria (left ventricular ejection fraction [LVEF], ≥50%, 40% to <50%, 30% to <40%, or <30%, respectively) or qualitative assessment. In cases in which patients had more than 1 assessment of LVEF, an evaluation performed during the index hospitalization was given precedence. In cases in which multiple evaluations occurred during the hospitalization, quantitative results took precedence over qualitative results.
The main outcome was all-cause 1-year mortality from the date of admission. The secondary outcome was 1-year HF-related readmission from the date of discharge.
Baseline demographic and clinical characteristics were compared between patients in different ordinal hematocrit groups using the Pearson χ2 test for categorical variables and the F test for continuous variables. The unadjusted association between worsening degree of anemia and outcomes was tested using the Pearson χ2 test.
Multivariable logistic regression models were subsequently constructed to test whether successively lower hematocrit levels were an independent predictor of 1-year mortality and HF-related readmission. Hematocrit categories were compared using dummy variables, with a hematocrit of greater than 40% to 44% as the reference group. Covariates included sociodemographic factors (age, sex, race, and nursing home residence), non–cardiac-related comorbidities (immobility, urinary incontinence, dementia, stroke, chronic obstructive pulmonary disease, creatinine and serum urea nitrogen levels, and diabetes mellitus), cardiac-related comorbidities (coronary artery disease, hypertension, coronary artery bypass graft surgery, percutaneous transluminal coronary angioplasty, myocardial infarction, severity of left ventricular systolic dysfunction, left bundle branch block, and atrial fibrillation), HF severity indicators (peripheral edema, pulmonary edema, hyponatremia, and history of HF), vital signs on admission (systolic and diastolic blood pressure, heart rate, and respiratory rate), other laboratory values on admission (albumin, potassium, and glucose levels and white blood cell count), and admission medications (angiotensin-converting enzyme inhibitors, angiotensin receptor blockers, β-blockers, diuretics, digoxin, calcium channel blockers, and iron supplements); these were identified in bivariate analyses or in previous studies,18,19 or clinically considered as important predictors of 1-year mortality.
To maximize the ability to adjust for the severity of comorbidities, all continuous variables were parameterized before entry into the multivariable model. We analyzed individual bivariate relationships between each variable and 1-year mortality. Subsequently, based on the nature of the relationship, variables were entered into the multivariable models as continuous variables, ordinal variables, or a spline function.
Continuous variables, for which less than 5% of data were missing, had the missing values replaced with the cohort’s median values. Those continuous variables for which between 5% and 10% of data were missing were assigned the cohort’s median values and a dummy binary variable denoting the missing value. A similar procedure was followed for binary (yes or no) variables, with missing data imputed as no. All continuous and binary variables in our analyses, except for albumin level, had less than 10% missing data. For categorical variables with missing values, multiple dichotomous variables (dummy variables) were created, corresponding to each of the categories (including unknown) and then incorporated into the models, with one of the categories being a reference group. All categorical variables in the models had less than 10% missing data, with the exception of LVEF, immobility, and urinary incontinence; these variables were retained in the model given their importance as predictors of outcome. Odds ratios in the final models were converted to risk ratios.20 Risk-standardized mortality rates were calculated as a multiplication product of overall mortality rates and risk-standardized ratio, using an indirect standardization technique.21 Multivariable models assessing the association between admission hematocrit and 1-year mortality were repeated, including admission hematocrit × LVEF interaction terms to assess whether hematocrit-associated mortality risk differed between patients with an LVEF of less than 40%, an LVEF of 40% or more, and unknown. We performed a bootstrap analysis by sampling with replacement for 1000 iterations to internally validate the stability of the results from multivariable models. We also performed a subsidiary Cox proportional hazards regression analysis for 1-year mortality and HF-related readmissions to confirm the results of the multivariable logistic regression method.
Analyses incorporated probability weights based on the inverse sampling fraction for each state. All models were adjusted for clustering by hospital. Analyses were conducted using Stata, version 8.0 (Stata Corp, College Station, Tex). The use of the Centers for Medicare & Medicaid Services’ NHC Project database was approved by the Yale University School of Medicine Human Investigation Committee.
The distribution of hematocrit values is shown in Figure 1. The median hematocrit was 37%; 61.2% of men and 52.1% of women were anemic (defined as a hematocrit of <40% for men and <37% for women), and 42.2% of patients had a hematocrit of 36% or less. When compared with patients who had a higher admission hematocrit, those with a lower admission hematocrit were more likely to be women and nonwhite, and more likely to have non–cardiac-related comorbidities, such as urinary incontinence, stroke, dementia, immobility, and diabetes mellitus. Patients with a lower hematocrit were also more likely to have a history of HF, a lower systolic and diastolic blood pressure, peripheral edema and pulmonary edema on admission, higher creatinine and serum urea nitrogen levels, and lower albumin and sodium levels (Table 1).
In unadjusted analysis, lower hematocrit values were associated with higher mortality rates, with statistically and clinically significant differences present between the groups (P < .001 for trend; (Figure 2A). In multivariable analysis, adjustment for demographic factors did not affect the higher relative risk (RR) of death for patients with lower hematocrit values compared with patients whose hematocrit was greater than 40% to 44%. However, adjustment for non–cardiac-related comorbidities markedly attenuated the association between lower hematocrit level and risk of death. Adjustments for other clinical factors, including HF severity, lowered the risk ratios even further (Table 2).
In the final multivariable model, patients in the 2 groups with the lowest hematocrit were not at increased risk of death at 1 year compared with patients who had a hematocrit of greater than 40% to 44% (for hematocrit ≤24%: RR, 1.02, 95% confidence interval [CI], 0.86-1.19; for hematocrit >24%-28%: RR, 0.99, 95% CI, 0.89-1.09; Table 2). The RR of death for patients in groups with a hematocrit greater than 28% to 32% and greater than 32% to 36% was slightly higher than that of patients with a normal hematocrit (RR, 1.14, 95% CI, 1.06-1.22; and RR, 1.09, 95% CI, 1.03-1.16, respectively, vs those with a hematocrit greater than 40% to 44%). This higher RR corresponded to only a modest increase in the absolute risk-adjusted mortality of 3.1% for patients with a hematocrit greater than 28% to 32% and 2.3% for those with a hematocrit of greater than 32% to 36% (vs those with a hematocrit >40% to 44%; (Figure 2B). More important, the graded relationship between lower hematocrit values and higher mortality initially seen in crude analysis was no longer observed in the multivariable model (P = .08 for trend; Figure 2B).
The effect of hematocrit levels on 1-year mortality was similar across the patient subgroups with an LVEF less than 40%, 40% or more, and unknown, with no significant hematocrit × LVEF interaction (P = .33 for the interaction).
In unadjusted analysis, lower hematocrit values were associated with higher risk of 1-year HF-related readmissions (Table 3). This association persisted in multivariable analysis after adjustments for demographic and clinical factors. Patients in all groups with a lower hematocrit had a significantly higher risk of 1-year HF-related readmission compared with patients who had a hematocrit greater than 40% to 44% (Table 3).
Bootstrap analysis validated the results obtained from multivariable logistic regression models (Table 2 and Table 3). Results from the Cox proportional hazards regression analysis of 1-year mortality and HF-related readmissions were similar to those obtained by using multivariable logistic regression (eg, 1-year mortality for patients with a hematocrit ≤24% vs those with a hematocrit >40%-44%: unadjusted hazard ratio, 1.71 [95% CI, 1.47-1.98]; and fully adjusted hazard ratio, 0.98 [95% CI, 0.84-1.15]).
This large nationally representative study of older patients in the United States hospitalized with HF demonstrated no graded relationship between lower hematocrit values and increased mortality. Furthermore, hematocrit values were not independently associated with higher risk of mortality in patients with the most severe degrees of anemia. The effect of anemia on mortality observed in crude analysis was markedly attenuated by adjustments for the presence and severity of non–cardiac-related comorbidities. However, the independent graded relationship between lower hematocrit values and hospital readmission for HF was not affected by adjustments for patient characteristics. Our findings suggest that although anemia is an independent predictor of hospital readmission, it may be predominantly a marker of increased mortality rather than a direct mediator of this risk.
Our results contrast with those of several reports.1-9 An association between anemia and increased mortality rates in a cohort of elderly patients hospitalized with HF in Connecticut was previously reported.6 A study by Horwich and colleagues2 also linked anemia with higher mortality, HF severity, and worse functional status in a single-center cohort of patients with severe HF. Similar observations were made in a small community-based investigation3 and in a larger population-based study of HF patients in Alberta, Canada.4 Furthermore, retrospective analyses of randomized controlled trials showed an association between anemia and increased mortality in patients with systolic left ventricular dysfunction.1,5,7-9 Although these prior studies suggest a 2% to 5% increase in the RR of death for each 1% decrease in hematocrit,1-3,5,6,8 our data do not reveal a linear relationship between the degree of anemia and adjusted 1-year mortality. In fact, after multivariable adjustment, the RR of 1-year mortality in our sample was similar in patients with the most severe anemia (hematocrit, ≤24% and >24%-28%) and in those with normal hematocrit values. While the RR of death was slightly higher (9%-14% increase) in patients with moderate anemia (hematocrit, >28%-32% and >32%-36%), the magnitude of this risk was markedly lower than the 25% to 50% RR increase observed by previous investigators.6,7 The overall pattern suggests the absence of an important independent effect of hematocrit on mortality in this population.
There are several possible explanations why our findings differ from those of prior studies. As previously observed in other patient populations, insufficient adjustment for confounding by the burden of comorbidities can lead to an overestimation of risk.22 Recent data have demonstrated that non–cardiac-related comorbidities are highly prevalent in elderly patients with HF, and strongly associated with adverse outcomes.23 However, many of the prior population-based studies did not consider (because of either exclusion or unavailable data) non–cardiac-related comorbidities such as dementia, immobility, urinary incontinence,1-9 cerebrovascular disease,2,5,7-9 and chronic obstructive pulmonary disease,1,2,5,7,8 many of which have been prognostically important in patients with HF.18 Although several studies did attempt to adjust for this broader range of non–cardiac-related comorbidities,3,4,6 most used administrative claims data4 (or indexes derived from claims data),3,6 a method that is less precise than medical record review.22 The present study provided the opportunity to adjust for a broad scope of cardiac- and non–cardiac-related comorbidities. Adjustment for non–cardiac-related comorbidities, which included chronic conditions such as dementia, incontinence, and immobility in addition to other potential risk factors such as renal insufficiency, chronic obstructive pulmonary disease, diabetes mellitus, and stroke, markedly attenuated the initially observed association between anemia and mortality.
A second potential explanation is the limited generalizability of prior studies. Most investigations analyzed relatively small numbers of patients and had geographic limitations,2,3,6 with some studies being conducted in single centers.2 In addition, randomized clinical trials, which provided the data for some previous analyses, typically used multiple exclusion criteria, thus selecting younger mostly male patients1,5,7,8 with relatively few comorbidities1,5,7-9 and focusing exclusively on patients with left ventricular systolic dysfunction.1,5,7-9 As a result, these studies analyzed patient populations that are different from ours, and that might not have been representative of elderly HF patients typically seen in clinical practice, many of whom have extensive non–cardiac-related comorbidities and/or diastolic dysfunction.14,24 To our knowledge, we examined not only the largest patient sample to date but also one that represents the full spectrum of elderly patients hospitalized for HF in the United States.
Our findings are consistent with those of a recent study25 that failed to demonstrate an association between lower hemoglobin concentrations and increased mortality in patients with acute HF. The researchers25 suggested that this lack of association could have resulted from the acute nature of HF in their small patient sample. While this may be true, our results confirm this finding in a much larger and more diverse patient population.
Our findings confirm the association between anemia and repeat hospital admissions for HF,5,6 although the magnitude of this association in our study (20%-30% RR increase in patients with severe anemia) was less pronounced than the previously reported 30% to 49% RR increase.6 The reasons for this discrepancy between the utility of anemia as a predictor of readmission and its relative lack of value as a predictor of mortality in HF patients are unclear. In the absence of well-defined standards for hospital admission, high variation among physicians26 in their perception of HF severity in patients with anemia could result in more frequent admissions in this patient population. Another possibility is that anemia could still be associated with more pronounced HF symptoms and functional impairment,2 through its hemodynamic and neurohormonal effects.27,28 This possible differential effect of anemia on outcomes is supported by previous studies in HF patients with end-stage renal disease that show that treatment of anemia improves the functional status and lessens the severity of HF29 but does not lower mortality.30,31 Finally, the effect of anemia on repeat hospital admissions could also result from residual confounding by unmeasured factors.
Whether treatment of anemia with iron supplementation, erythropoietin, or blood transfusions would have any effect on outcomes remains unknown. Several pilot investigations32,33 have suggested that treatment of anemia with iron supplementation and erythropoietin may improve functional status and LVEF, and reduce the need for repeat hospitalizations in HF patients. However, because this therapy is not risk free,34 its effects need to be clearly demonstrated in well-designed, prospective, randomized trials before any definitive conclusions about its usefulness can be made. Our results suggest that these trials should focus on functional status and readmission as potential outcomes, rather than exclusively on mortality.
We were unable to determine the cause of anemia in the patient population. Previous research35 has indicated that development of anemia in patients with HF is often multifactorial. One recent study36 suggested that HF patients with increased plasma volume and dilutional anemia may have worse outcomes compared with patients who have reduced red blood cell volume and “true” anemia. Although our patient sample likely included many patients with both types of anemia, it is possible that anemia may have a different effect on outcomes in these distinct patient groups. We did not have follow-up laboratory information and could not determine how many patients who were anemic on admission remained anemic during hospitalization. We also did not have information about transfusion therapy for our patients, but adjustments were made for other treatments of anemia. Although it is conceivable that patients in groups with moderate anemia had slightly higher mortality risk because they were treated less aggressively (with blood transfusions, iron supplements, and erythropoietin) than patients with severe anemia, it is unlikely. First, adjustment for differences in iron therapy between the groups did not affect the results. Second, although we did not have information about blood transfusion therapy, there are no data to suggest any therapeutic benefit of blood transfusion therapy in patients with HF. Our study was performed in patients 65 years and older. Even though older patients represent most patients with HF, the results of this investigation might not apply to younger patients with HF.
In conclusion, although anemia is an independent predictor of hospital readmission for HF, its prognostic association with increased mortality in HF patients is largely explained by the severity of comorbid illness. Future studies of prognostic markers in patients with HF should account for the broad range of non–cardiac- and cardiac-related comorbidities.
Correspondence: Harlan M. Krumholz, MD, SM, Yale University School of Medicine, 333 Cedar St, PO Box 208088, New Haven, CT 06520-8088 (email@example.com).
Accepted for Publication: April 26, 2005.
Financial Disclosure: None.
Funding/Support: This study was supported by Research Career Award K08-AG01011 from the National Institutes of Health, Bethesda, Md (Dr Masoudi); and Research Career Award K08-AG20623-01 from the National Institute on Aging, National Institutes of Health (Dr Foody). Dr Foody also received a National Institute on Aging/Hartford Foundation Fellowship in Geriatrics.
Role of the Sponsor: The funding bodies had no role in data extraction and analyses, in the writing of the manuscript, or in the decision to submit the manuscript for publication.
Disclaimer: The analyses upon which this publication is based were performed under contract number 500-02-CO-01, entitled “Utilization and Quality Control Peer Review Organization for the State of Colorado,” sponsored by the Centers for Medicare & Medicaid Services (CMS, formerly the Health Care Financing Administration), Department of Health and Human Services. The content of the publication does not necessarily reflect the views or policies of the Department of Health and Human Services, nor does mention of trade names, commercial products, or organizations imply endorsement by the US government. The authors assume full responsibility for the accuracy and completeness of the ideas presented. This article is a direct result of the Health Care Quality Improvement Program initiated by CMS, which has encouraged identification of quality improvement projects from analysis of patterns of care and, therefore, required no special funding on the part of this contractor. Ideas and contributions to the authors concerning experiences in engaging with issues presented are welcomed.
Create a personal account or sign in to: