Patient selection algorithm. ACE-I indicates angiotensin-converting enzyme inhibitors; LV, left ventricular.
Life-table analysis comparing survival in angiotensin-converting enzyme inhibitors (ACE-I)–treated and –untreated patients.
Havranek EP, Abrams F, Stevens E, Parker K. Determinants of Mortality in Elderly Patients With Heart FailureThe Role of Angiotensin-Converting Enzyme Inhibitors. Arch Intern Med. 1998;158(18):2024-2028. doi:10.1001/archinte.158.18.2024
Despite randomized controlled trials demonstrating mortality reduction, many studies have documented persistent low rates of prescription of angiotensin-converting enzyme inhibitors (ACE-I) in patients with heart failures; the reasons for this pattern remain poorly defined. In addition, some authors have argued that the results of carefully controlled clinical trials do not translate well into the uncontrolled world of clinical practice, and the mortality benefits of ACE-I may not extend into special populations such as the elderly.
To understand the reasons for failure to prescribe ACE-I to Medicare patients with heart failure and to assess the impact of this failure on mortality.
We obtained data by reviewing charts of Medicare patients discharged from 7 Colorado hospitals with a diagnosis of heart failure during 1994.
We identified a diagnosis of heart failure in 1016 patients without a contraindication to ACE-I. Three hundred seventy-eight of these patients were receiving ACE-I at the time of admission. Of the 638 remaining, 257 had their left ventricular systolic function assessed and 92 had diminished function. Of these 92, 50 (54.3%) were discharged on a regimen of ACE-I. The only significant difference in baseline comorbidity or demographic variables between those given and those not given ACE-I was that patients not prescribed ACE-I were older. Using multivariate analysis, younger age and cardiology consultation predicted ACE-I prescription (P=.02). By life-table analysis, mortality at 1 year following discharge from the index hospitalization was lower in those treated with ACE-I than in those not so treated (P=.03). The Deyo index of comorbidity, prescription of an ACE-I, site of treatment, and presence or absence of cardiology consultation were significantly associated with 1-year mortality by multivariate analysis (P<.001).
Underinvestigation and undertreatment of chronic heart failure persists. Failure to treat elderly patients with ACE-I is associated with a mortality that appears to be greater than that seen in the placebo arms of large clinical trials of ACE-I therapy. Within the population studied, older patients are less likely to be treated. Failure of age to significantly add to prediction of mortality implies that the apparent bias against treating older patients with chronic heart failure with ACE-I is not justified. Because mortality is dependent on provider and site of treatment, further reductions in mortality from chronic heart failure may require intensive and selective local efforts, or development of regional heart failure centers.
CHRONIC HEART failure (CHF) affects between 1 and 2 million Americans1 and is responsible for large and growing annual health care expenditures.2,3 As its incidence increases, it has become the most common discharge diagnosis among Medicare beneficiaries.4 Guidelines for care of patients with CHF have been proposed by at least 2 expert groups.5,6 The major recommendations contained in these guidelines are basically consistent, and review criteria based on these guidelines have been proposed7 that emphasize identifying patients with left ventricular systolic dysfunction and using angiotensin-converting enzyme inhibitors (ACE-I).
Several surveys have been published on the frequency with which physicians prescribe ACE-I to patients with CHF. Rajfer8 noted that in 1991, self-reported rates of ACE-I prescription in patients with moderate CHF varied from 60% for general or family practitioners to 80% for cardiologists. Since then such results seem to have remained stable. Philbin et al9 noted that 47% of patients treated at 2 hospitals for CHF during 1992 were discharged on a regimen of ACE-I. McDermott et al10 reviewed charts of patients admitted with CHF between August 1992 and November 1993; they found that 69% of patients were discharged on a regimen of ACE-I. In a survey performed in 1995, Edep and colleagues11 found that self-reports of measurement of left ventricular function in patients with heart failure varied from 92% for cardiologists to 61% for family physicians, and that prescription of ACE-I for patients with mild to moderate heart failure varied from 80% for cardiologists to 59% for family physicians.
Based on these observations, we hypothesized that we could identify underdiagnosis and undertreatment of CHF due to systolic dysfunction, and that undertreatment would be associated with adverse outcome. We also sought to identify potentially remediable factors associated with these adverse outcomes.
We conducted a chart review based on discharges from 7 Colorado hospitals, which selected for the study by stratified sampling. Using cluster analysis, we created groups of Colorado hospitals based on total patient discharges and bed-days per year, Medicare discharges and bed-days per year, case-mix index, and cost data. This analysis resulted in 6 groups ranging in size from 3 to 28 hospitals. One hospital from each group was selected; all selected hospitals agreed to participate. The seventh hospital was added at that institution's request. The hospitals ranged in size from 26 to 420 licensed beds (Table 1), and included rural and urban and teaching and nonteaching facilities.
From January 1, 1994, to December 31, 1994, we identified all discharges with a principal or secondary diagnosis of CHF. Those who died or were transferred to another facility within 24 hours, were unable to take oral medications, or had an absolute contraindication to use of an ACE-I (bilateral renal artery stenosis, history of laryngeal angioedema, or severe aortic or mitral valvular or subvalvular obstruction) were excluded from further analysis. Next we divided patients into those who were or were not taking an ACE-I at the time of admission. Charts of patients not receiving ACE-I were reviewed for clinical evidence, confirming a diagnosis of CHF and demonstrating that heart failure was significant to the hospital stay. These criteria were modified from those used in the Framingham study12: (1) chest radiograph showing CHF or pulmonary edema, (2) 1 major criterion (paroxysmal nocturnal dyspnea, orthopnea, or S3 gallop) and 1 minor criterion (edema, dyspnea, jugular venous distention, rales, or displaced point of maximal impulse), or (3) a history of CHF with 1 major criterion. We then reviewed charts for evidence of measurement of left ventricular systolic function by echocardiography, radionuclide ventriculography, or contrast ventriculography and for ACE-I prescription either during hospitalization or at the time of discharge. Reduced systolic function was defined as an ejection fraction of less than 40% when explicitly measured by echocardiography or ventriculography, or when qualitative assessment by echocardiography described left ventricular function as being at least moderately diminished.
Comorbidity was assessed using Deyo modification of the Charlson index.13 This index maps diagnoses according to the International Classification of Diseases, Ninth Edition (ICD-9)14 coding, and has been shown to predict mortality for hospitalized patients.
For patients with confirmed left ventricular systolic dysfunction considered candidates for ACE-I, we assessed vital status for 365 days following discharge as our primary outcome measure using Health Care Financing Administration data files.
Analysis was carried out using STATA software, release 4.0 (Stata Corporation, College Station, Tex). Demographic variables were compared using the Wilcoxon rank sum test for continuous variables and χ2 for categorical variables. Variables were included in multivariate models when P<.25 on univariate testing. The multivariate model was constructed with logistic regression analysis using the comparison of 2 log likelihoods approach.15 Survival was compared using a log-rank test.
The case selection algorithm is shown in Figure 1. We identified 1125 patients with a discharge diagnosis of heart failure. One hundred nine patients were excluded because of death or transfer within 24 hours of admission, inability to take oral medications, missing data, or an absolute contraindication to ACE-I. Of the remaining 1016, 378 were receiving ACE-I at the time of admission. Assessment of left ventricular systolic dysfunction could be confirmed in 257 (40.3%) of the remaining 638 patients. Function was diminished in 92 (35.8%) of these patients. Demographics and comorbidity are summarized in Table 2.
Among the 92 discharged patients with left ventricular systolic dysfunction for whom ACE-I prescription was appropriate, 50 (54.3%) (group 2) and 42 (45.6%) (group 1) were treated and were not treated, respectively.
Patient-specific and provider-specific variables were investigated by univariate analysis for their contribution to predicting which patients were given ACE-I (Table 3). Based on these results, hospital in which the patient was treated, age, specialty of the attending physician, and presence of cardiology consultation were chosen for inclusion in the multivariate model. Using stepwise multiple logistic regression analysis, the multivariate model included age and presence of cardiology consultation, and significantly predicted ACE-I prescription (P=.02). The patient-specific variables of sex, race, diabetes, serum creatinine level higher or lower than 265 µmol/L (≥3.0 or <3.0 mg/dL), and comorbidity, and the provider-specific variables of specialty of the physician-of-record and location of treatment did not contribute to the model.
We evaluated the fit of the model by assessing the false-positive and false-negative classification rates for this model by creating a 2 × 2 table with .50 as the cutoff point (if predicted probability of ACE-I was ≥.50, the prediction was ACE-I at discharge; if the predicted probability of ACE-I was <.50, the prediction was no ACE-I at discharge). The false-negative rate was 26% and the false-positive rate was 43%.
There was a 2:1 preponderance of men in the study sample of patients with demonstrated systolic dysfunction. Because this finding is unexpected in an elderly population, we investigated the possibility of selection bias. In the original study sample of 1016 patients, there were 423 men and 593 women. More men than women had an assessment of left ventricular systolic function, 46.3% vs 36.1% (P=.001).
Next, we assessed the impact of ACE-I prescription on mortality by comparing survival of patients treated with ACE-I at discharge (including those receiving ACE-I on admission) with those not so treated (Figure 2). The 1-year survival was 76% in treated patients and 64% in untreated patients. This difference was statistically significant (P=.03).
We assessed the relationship between 1-year mortality and patient-specific and treatment-specific variables using univariate analysis (Table 4). Treatment site, presence of cardiology consultation, ACE-I prescription at discharge, and Deyo index were chosen for inclusion in the multivariate analysis. Using stepwise logistic regression analysis, all 4 variables were selected for inclusion in the final model. This model significantly predicted mortality (P<.001).
We evaluated the fit of the model by assessing the false-positive and false-negative classification rates as was done earlier for the model for prediction of ACE-I prescription. The false-positive rate was 25% and the false-negative rate was 46%.
The present study replicates the finding of underevaluation and undertreatment of patients with heart failure. The likelihood that patients would be treated with ACE-I decreased with advancing age. Consultation with a cardiologist made it more likely that an ACE-I would be prescribed. Failure to treat with ACE-I was associated with higher 1-year mortality. Multivariate analysis demonstrated that comorbidity, site of treatment, and cardiology consultation also were significant predictors of survival.
Variability in outcome is the result of random variation, variation resulting from patient factors, and variation from delivery system factors. In our study all these factors were present, with comorbidity being the major patient factor and ACE-I prescription, site of hospitalization, and consultation by a cardiologist as significant delivery system factors.
In the Studies of Left Ventricular Dysfunction (SOLVD),16 mortality at 12 months was 12.3% in patients treated with enalapril and 15.5% in placebo-treated patients; corresponding 12-month mortality in the current study was approximately 24% and 36%, respectively, for patients treated and not treated with ACE-I. The higher overall mortality in the current study almost certainly reflects the greater comorbidity and age of our cohort compared with patients enrolled in clinical trials. Despite this finding, the impact of ACE-I on mortality in the elderly in clinical practice appears to be greater than that seen in clinical trials. A meta-analysis of the trials of ACE-I in CHF,17 which included results of the SOLVD study, estimates that risk of mortality is reduced by 23% by this class of drug, which is smaller than the apparent risk reduction in our study. There are several possible reasons for the discrepancy between the results of clinical trials and the results in clinical practice.
One explanation is selection bias. It is reasonable to believe that the baseline characteristics of the patients in our study and those in the randomized trials are not strictly comparable. If, however, "real world" patients had poorer baseline prognosis than randomized trial patients, we would expect the mortality in the ACE-I–treated patients in our study to also be much greater than the mortality in treated patients in randomized trials.
Another possible explanation is that failure to use ACE-I in patients with CHF is a marker for poor overall care and not the sole factor in different care. It is possible that other treatment modalities, besides use of ACE-I, have an impact on mortality. Stevenson et al18 have documented other changes in medical care that may reduce mortality in patients with CHF such as diminished use of type I antiarrhythmic agents and greater use of anticoagulants. Patients not treated with ACE-I in the current study may have had their care differ in ways not measured, the net result of which is a greater reduction in mortality over those not treated with ACE-I.
The factors associated with failure to prescribe ACE-I and with survival are consistent with most data in the literature. Although age was a factor in whether patients had ACE-I prescribed, it did not predict outcome; this finding is consistent with the findings of Krumholz et al.4 The fact that outcome varied with the hospital in which the patient was treated should not be surprising. Regional variation in health care practices are well documented.19
Explaining failure of the other factors to predict use of ACE-I has a more variable relationship to precedents in the literature. The relationship between quality of CHF care and specialty of the physician-of-record is not clear. Surveys8,11 have demonstrated cardiologists to be more knowledgable about the indications for ACE-I therapy, but differences in practice have not been demonstrated. In a previous study,20 we found cardiologists and primary care practitioners to be equally likely to prescribe ACE-I. Data on readmission rates for CHF by specialty of provider (cardiologists vs generalists) have yielded conflicting results.21,22 A strength of our study is that we examined care in a wide range of settings that may be more reflective of wider trends. A higher baseline creatinine level has been a factor in other studies9 but not in the current study, possibly due to relatively small numbers of patients in our study with creatinine levels higher than 221 µmol/L (2.5 mg/dL).
The retrospective design of our study precluded assessment of important baseline characteristics, possibly confounding the mortality analysis. In particular, it would be important to assess functional status in an elderly cohort; a prospective study design would be necessary to accomplish this with reasonable accuracy. As noted earlier, the selective nature of the population studied, especially the restricted age range, may render our results not applicable to other populations. The sample size was relatively small. Some of the variability in ACE-I prescription and in survival remains unexplained; it is unlikely that random variation accounts for all of this. Unmeasured factors are likely playing a role, and confirmation of the current results with a larger sample may uncover some of these factors. In particular, the effect of sex bias in measurement of systolic function needs to be assessed.
Low rates of adequate treatment of patients with CHF documented in our study are similar to rates described in previous studies. Undertreatment is associated with a mortality rate that appears to be much higher than that seen in the placebo arms of randomized trials. Variation in the way care is provided, including site of treatment and consultation with a cardiologist, also affects outcome.
Although findings such as these typically result in calls for increased physician education, new approaches may be needed in the treatment of patients with CHF. Concurrent review of patients during hospitalizations for heart failure in facilities in which practice varies widely from the median or regionalization of care are examples of approaches that should be tested.
Accepted for publication March 26, 1998.
The analyses on which this publication is based were performed under contract No. 500-96P611, titled "Utilization and Quality Control Peer Review Organization for the State of Colorado," sponsored by the Health Care Financing Administration, Department of Health and Human Services. The content of this publication does not reflect the views or policies of the US 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.
Presented at the American Heart Association 70th Scientific Sessions, Orlando, Fla, November 12, 1997.
We acknowledge the contributions of Rosemary Wilkinson and Lane Craddock, MD, to the design of the study.
Reprints: Edward P. Havranek, MD, Denver Health Medical Center No. 0960, 777 Bannock St, Denver, CO 80220-4507.