Anemia and Renal Insufficiency Are Independent Risk Factors for Death among Patients with Congestive Heart Failure Admitted to Community Hospitals: A Population-Based Study

Study Design

This is a retrospective cohort study of randomly selected Medicare beneficiaries discharged from community hospitals with a primary diagnosis of heart failure.

Study Population

All Medicare beneficiaries admitted to any general hospital in a single southeastern state between July 1, 1998 and December 31, 1998, who were alive at discharge with a principal diagnosis of heart failure (ICD9 codes 402.01, 402.11, 402.91, 404.01, 404.11, 404.91, and 428.xx) were eligible for inclusion in the study. We excluded patients with a history of renal disease identified as a history of chronic dialysis in the medical record (n = 9). We also excluded patients who died before discharge (n = 26) and patients for whom we could find no record of their clinical status in the state Medicare status file (n = 50).

Patient Sample

Records for eligible patients were identified by the Medicare Medical Information System, and a random sample of eligible discharges was chosen.

Data Abstraction

Trained abstractors used a standardized form to abstract patient charts. Data abstracted for each patient included date of admission and discharge, demographic information (age, race, and gender), and degree of mobility on discharge. Abstracted comorbid conditions as recorded by the clinician included a history of stroke, ESRD (defined as chronic dialysis), myocardial infarction, heart failure, coronary artery disease, angina pectoris, diabetes mellitus, and hypertension. We used the ICD-9 diagnoses reported on the hospital discharge record to calculate a Charlson comorbidity score for each patient (17). The Charlson comorbidity score is the sum of weighted values for ICD-9 codes assigned during a hospitalization and is predictive of subsequent risk of hospitalization and death. Patients given an angiotensin-converting enzyme inhibitor (ACEI) on discharge were also identified.

Reports of left ventricular ejection fractions were identified and recorded. If a previous and a current ejection fraction were both present in the record, we used the lower value for the purposes of analysis. We categorized individuals with an ejection fraction of ≤40% as having left ventricular systolic dysfunction (LVSD), those with a ejection fraction >40% as having left ventricular diastolic dysfunction (LVDD), and those without a recorded value as undetermined left ventricular function (18).

The pulse rate, systolic and diastolic BP, and the presence of peripheral edema on admission were recorded. Radiographic changes of pulmonary edema or heart failure were recorded. Laboratory values for the first serum creatinine and hematocrit recorded in the chart were abstracted. We defined CKD as a serum creatinine >1.4 mg/dl for women and >1.5 mg/dl for men. We also calculated the GFR using the equation used in the Modification of Diet in Renal Disease (MDRD) study, as modified for age, gender, and race (19), and substituted the resulting rate for creatinine levels in some models. In some analyses, anemia was defined as a hematocrit of <30%.

Follow-up

Follow-up began on the date of hospital discharge for the index admission and continued until March, 1, 2001. Patients were characterized as dead if a date of death had been recorded for them in the Medicare database and as missing if neither a date of death nor a current Medicare eligibility record was found. We were unable to identify either a death date or evidence for current eligibility for 18 (2.6%) of the patients, and these patients were excluded from mortality analyses.

Statistical Analyses

We calculated descriptive statistics and the 12-mo risk of death for selected categories of key covariates. We described the crude associations between level of hematocrit and the presence of CKD with mortality by Kaplan-Meier plots. We used risk ratios and associated confidence intervals to measure the association between categorical variables and risk of death. To account for confounding and to assess the joint effects of renal insufficiency and anemia, we used Cox regression models. In most models, we used serum creatinine and hematocrit as continuous variables along with age, ejection fraction, and indicators for sex, race, comorbid history (stroke, myocardial infarction, coronary artery disease, and angina pectoris), hypertension, diabetes mellitus, a previous history of heart failure, and discharge functional status. We also treated creatinine (CKD yes/no) and hematocrit (anemia yes/no) as dichotomous in some models. Hematocrit was dichotomized at <30% and ≥30% for the purposes of these models.

We assessed the independent association between anemia and mortality for type of left ventricular function in two ways. First, we examined the association between anemia (hematocrit <30%) and mortality for the three levels of left ventricular function. Second, we included interaction terms in the multivariate model between left ventricular function and anemia.

We assessed the proportional hazards assumption using plots of log-log survival and interaction terms with time. We screened for potential collinearity using correlation coefficients. We further assessed model fit by including interaction terms between creatinine and hematocrit, with other independent variables. To determine the importance of these factors on mortality soon after hospital discharge, we repeated these survival analyses by limiting the follow-up to the first 90 d, 180 d, 270 d, and to 1 yr. All analyses were conducted using SAS Version 8.1 (SAS Institute, Cary, NC).

Baseline Characteristics

Our sample included 755 patients admitted to community hospitals with heart failure. After excluding duplicate records, patients with a history of renal disease, patients who died before discharge, and patients lost to follow-up, data from 665 patients were analyzed. Records of patients from 120 hospitals were sampled, yielding a mean (SD) of 5.5 (5.3) patients per hospital and a 25^th to 75^th intraquartile range from 2 to 7 patients. The range of patients contributed to the sample by an individual hospital was from 1 to 2.

The mean (SD) age was 75.7 (10.9) yr, with a range from 29 to 100 yr; 60% were women, and 71% were white. A history of heart failure was present for 75% of the patients. A history of stroke was reported for 20%, myocardial infarction for 26%, coronary artery disease for 51%, and angina pectoris for 15% of patients. Hypertension was reported for 66% and diabetes mellitus for 44% (Table 1).

A report describing previous or current left ventricular ejection fraction was found in 60.0% of the patient’s charts. The mean (SD) ejection fraction was 38.4% (16.9), with a range from 10 to 85%. Among the 399 patients with a recorded ejection fraction, 60% had LVSD with a mean (SD) ejection fraction of 26.5% (8.1). The mean ejection fraction among individuals with LVDD was significantly higher at 56.4% (9.0%) (P < 0.0001).

Peripheral edema was noted on admission in 70% of patients, and a chest x-ray with findings of either pulmonary edema or heart failure was present for 76%. There was no difference in either the proportion of patients with peripheral edema (P = 0.082) or positive findings on chest x-ray (P = 0.54) among those individuals with undetermined left ventricular function, LVSD and LVDD. An ACEI was prescribed at discharge for 54% of the cohort. ACEI use was less common among individuals with CKD (49%) compared with those without CKD (58%) (P = 0.018).

Prevalence of CKD

A serum creatinine value was reported on admission for 646 members (97%) of the cohort. The mean (SD) value was 1.46 (0.78) mg/dl, with a range from 0.5 to 7.5 mg/dl and a 25^th to 75^th intraquartile range from 1.0 to 1.7 mg/dl. CKD, defined by a serum creatinine of ≥1.4 mg/dl for women and ≥1.5 mg/dl for men, was present in 38% of the cohort. Men (46%) were more likely than were women (33%) to have CKD, (odds ratio [OR], 1.70; 95% confidence interval [CI], 1.23 to 2.36). The mean (SD) serum creatinine among individuals with CKD was 2.16 (0.86) mg/dl; among those without CKD, it was 1.02 (0.21) mg/dl (P < 0.0001).

Prevalence of Anemia

A hematocrit was recorded for 633 members (95%) of the cohort. The mean (SD) hematocrit was 36.6% (6.3), with a 25^th to 75^th intraquartile range from 32.6% to 40.9%. The proportion of patients with a hematocrit ≥40% was 30.3%; 22.9% had a hematocrit between 36% and 39%, 33.2% between 30% and 35%, and 13.6% had a hematocrit of <30%.

The proportion of patients with CKD increased with increasing anemia (Table 2). A third (33%) of individuals with a serum creatinine between 2 and 3 mg/dl and half (50%) with a creatinine >3 mg/dl had a hematocrit <30%. Of note was that 25.1% of nonanemic CHF patients had CKD, and 40.0% of severely anemic patients had normal serum creatinine (Table 2). The mean (SD) serum creatinine was 1.26 (0.56) mg/dl among patients with no anemia, 1.35 mg/dl (0.55) in patients with mild anemia, 1.48 mg/dl (0.75) for moderate anemia, and 2.01 mg/dl (1.23) for severe anemia (P < 0.0001).

Mortality

Mortality rates were high; 59% of the cohort (n = 393) died during follow-up, and mortality at 1 yr was 36.8%. The presence of CKD was associated with an increased risk of death (Figure 1); 67% of patients with and 54% of those without CKD died during follow-up (relative risk [RR], 1.26; 95% CI, 1.11 to 1.42). One-year mortality rates among individuals with and without CKD were 44.9% and 31.4% respectively (RR, 1.43; [95% CI, 1.17 to 1.75).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1. Survival among heart failure patients with and without CKD adjusted for anemia, age, sex, race, stroke, heart attack, coronary artery disease, angina pectoris, hypertension and diabetes mellitus status, a previous history of heart failure, ejection fraction and discharge functional status.

Anemia on admission to the hospital was associated with an increased risk of death (Table 2 and Figure 2). Compared with individuals with a hematocrit ≥40%, the RR (95% CI) for death for individuals with a hematocrit between 36 and 39% was 1.05 (0.86 to 1.28); 1.19 (1.00 to 1.41) for a hematocrit between 30 and 35%; and 1.36 (1.12 to 1.65) for a hematocrit <30% (χ² for trend 10.0, 1 df; P = 0.016). Corresponding 1-yr RR were 1.08 (0.79 to 1.47); 1.17 (0.89 to 1.54); and 1.60 (1.19 to 2.16); χ² for trend was 7.37; P = 0.007.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2. Survival among heart failure patients with varying degrees of anemia adjusted for renal function, age, sex, race, stroke, heart attack, coronary artery disease, angina pectoris, hypertension and diabetes mellitus status, a previous history of heart failure, ejection fraction and discharge functional status.

Individuals with left ventricular diastolic and systolic dysfunction, as well as those with undetermined ventricular function, had comparable increased risk of death during the first year. Compared with individuals with a hematocrit ≥30%, the RR (95% CI) for death within 1 yr for anemic patients with left ventricular diastolic dysfunction was 1.74 (0.7 to 4.33); left ventricular systolic dysfunction, 1.56 (0.65 to 3.23); and undetermined left ventricular function, 2.03 (1.27 to 3.23). The estimates of RR did not differ significantly from one another by the Breslow-Day test for homogeneity (P = 0.61).

Multivariate Analysis

Both hematocrit and serum creatinine were independently associated with increased mortality during follow-up (Table 3) after controlling for age, gender, race, comorbidity, hypertension, and diabetes mellitus, a previous history of heart failure, lowest ejection fraction, discharge functional status, Charlson comorbidity index, and status of ACEI use at discharge. For each 1% increase in hematocrit, the mortality rate declined 1.6% (P = 0.09); for each one mg/dl increase in serum creatinine, the rate increased 16% (P = 0.025).

During the first year of follow-up, there was a 2.4% decrease in the hazard ratio associated with a 1% increase in hematocrit (P = 0.046) and a 15% increase in the hazard ratio associated with a one mg/dl increase in serum creatinine (P = 0.08) (Table 3). After controlling for all other risk factors, the hazard ratio associated with either the presence of anemia, 1.07 (0.59 to 1.9), or CKD, 1.24 (0.93 to 1.7), alone at 1 yr were comparable. The hazard ratio associated with the presence of both conditions was 2.17 (1.4 to 3.3), somewhat more than would be expected under a multiplicative model, but interaction terms in the statistical analysis were NS. The interaction terms for anemia-left ventricular function were NS.

We also examined the hazard ratio for both serum creatinine and hematocrit at 90 d, 180 d, and 270 d to determine the relative importance of the two measures immediately after hospital discharge. We retained the same covariates in these models that were used in the all-years mortality and 1-yr mortality analyses. The association between the hematocrit and risk of death was stable during the first year of follow-up and became significant at 270 d. The RR (95% CI) for a 1% increase in hematocrit at 90 d was 0.98 (0.95 to 1.02); at 180 d, 0.98 (0.95 to 1.01); and at 270 d, 0.97 (0.95 to 0.999).

The RR (95% CI) for a 1 mg/dl increase in serum creatinine was slightly less pronounced during the first year of follow-up than it was for all years combined. During the first 90 d of follow-up, the RR of death for each 1 mg/dl increase in serum creatinine was 1.05 (0.81 to 1.35); at 180 d, 1.07 (0.88 to 1.31); and at 270 d, 1.07 (0.9 to 1.27).

We treated both CKD and anemia as dichotomous exposures by grouping hematocrit as <30% (severe anemia) and those with a hematocrit ≥≥30%. When we retained the same covariates in the Cox models, the mortality rate during 1 yr of follow-up associated with severe anemia was 1.48 (1.04 to 2.12) (P = 0.029) and the mortality rate at 1 yr associated with CKD was 1.43 (1.06 to 1.92) (P = 0.0182).

Finally, we examined both hematocrit and creatinine clearance using the equation for calculating GFR as mentioned earlier. The creatinine clearance variable replaced creatinine, and all other covariates were retained. In examining total mortality, for each 1% increase in hematocrit, the mortality rate declined 1.6% (P = 0.087); for each 1 ml/min increase in creatinine clearance, the rate decreased 0.6% (P = 0.011). For one-year mortality, for each 1% increase in hematocrit the mortality rate declined 2.5% (P = 0.038); for each 1 ml/min increase in creatinine clearance, the rate decreased 0.6% (P = 0.083).