Author + information
- Received November 12, 1999
- Revision received November 3, 2000
- Accepted January 12, 2001
- Published online May 1, 2001.
- Michael B Fowler, MB, FRCP∗,* (, )
- Montserrat Vera-Llonch, MD, MPH†,
- Gerry Oster, PhD†,
- Michael R Bristow, MD, PhD, FACC§,
- Jay N Cohn, MD, FACC∥,
- Wilson S Colucci, MD, FACC¶,
- Edward M Gilbert, MD, FACC#,
- Mary Ann Lukas, MD, FACC∗∗,
- Michael J Lacey, MS∗∗,
- Randel Richner, RN∗∗,
- Sarah T Young, PhD∗∗,
- Milton Packer, MD, FACC‡,
- for the U.S. Carvedilol Heart Failure Study Group
- ↵*Reprint requests and correspondence: Dr. Michael B. Fowler, Falk C.V.R.C., Room 295, Stanford University School of Medicine, 300 Pasteur Drive, Stanford, California 94305-5246
Carvedilol reduces disease progression in heart failure, but to our knowledge, its effects on hospitalizations and costs have not been evaluated.
We examined the effects on hospitalization frequency and costs in the U.S. Carvedilol Heart Failure Trials Program. This program consisted of four concurrent, multicenter, double-blind, placebo-controlled studies involving 1,094 patients with New York Heart Association class II to IV heart failure, which treated patients with placebo or carvedilol for up to 15 months (median, 6.5 months).
Detailed resource utilization data were collected for all hospitalizations occurring between randomization and the end of follow-up. In-patient care costs were estimated based on observed levels of resource use.
Compared with placebo, carvedilol reduced the risk of hospitalization for any reason by 29% (p = 0.009), cardiovascular hospitalizations by 28% (p = 0.034) and heart failure hospitalizations by 38% (p = 0.041). Carvedilol also decreased the mean number of hospitalizations per patient (for cardiovascular reasons 30% [p = 0.02], for heart failure 53% [p = 0.03]). Among hospitalized patients, carvedilol reduced severity of illness during hospital admission, as reflected by shorter length of stay and less frequent use of intensive care. For heart failure hospital admissions, carvedilol decreased mean length of stay by 37% (p = 0.03) and mean number of intensive care unit/coronary care unit days by 83% (p = 0.001), with similar effects on cardiovascular admissions. As a result, estimated inpatient care costs with carvedilol were 57% lower for cardiovascular admissions (p = 0.016) and 81% lower for heart failure admissions (p = 0.022).
Carvedilol added to angiotensin-converting enzyme inhibition reduces hospitalization risk as well as severity of illness and resource utilization during admission in patients with chronic heart failure.
Heart failure, one of the most important reasons for hospitalization in the U.S., is the primary or secondary cause of over 2.5 million admissions yearly (1–3). Estimated annual costs exceed $12 billion (4). Heart failure hospitalizations have increased dramatically during the past 20 years, with further increases expected due both to treatment advances for coronary artery disease and to the aging population (5–7). Heart failure is the most common primary admission reason in patients older than 65 years, in whom 80% of such hospitalizations occur (7,8). Hospitalization risk is high in heart failure patients, and once hospitalized the risk increases for recurrent hospitalizations and death (3,8–11).
For these reasons, heart failure treatments are likely to have an important public health impact only if they reduce initial and recurrent hospitalization risk (and accompanying costs). Although both digoxin and angiotensin-converting enzyme (ACE) inhibitors decrease hospitalizations and associated costs in heart failure patients, hospitalization risk and the attendant costs of caring for these patients remain extremely high (12–16).
In view of these challenges, reports that adding beta-adrenergic blockade to ACE inhibition can reduce hospitalization risk have generated considerable interest. In large-scale studies, carvedilol, bisoprolol and metoprolol have been reported to reduce heart failure hospitalizations (17–21). However, little is known about the effect of beta-blockade on inpatient utilization of health care resources and costs. To address these issues, we conducted a detailed analysis of the effect of carvedilol on hospitalizations in the U.S. Carvedilol Trials Program.
Patients and procedures
Details of the design and conduct of the U.S. Carvedilol Trials Program, four concurrent trials that evaluated the effects of carvedilol in heart failure, have been previously published (17). All patients had systolic dysfunction (left ventricular ejection fraction [LVEF] ≤0.35) and were in New York Heart Association (NYHA) class II to IV despite therapy (diuretics, and usually, digoxin and ACE inhibition). Patients tolerating open-label carvedilol (6.25 twice daily) for 2 weeks underwent double-blind randomization to treatment with placebo or carvedilol, then were uptitrated and maintained on regimens of target doses (up to 25 to 50 mg twice daily) for up to 15 months.
Data on the occurrence of all hospitalizations were prospectively obtained throughout double-blind therapy. For all hospital admissions deemed cardiovascular in nature, information was prospectively collected on primary admission cause, admission and discharge dates and use of selected cardiac procedures (coronary angiography, permanent or temporary pacemaker implantation, direct current cardioversion, defibrillation, coronary artery bypass grafting [CABG], percutaneous transluminal coronary angioplasty [PTCA], cardiac catheterization or heart transplantation). Data on special-care unit use (i.e., intensive care unit or coronary care unit [ICU/CCU]) were collected retrospectively for all such admissions.
Primary hospital admission cause (worsening congestive heart failure, myocardial infarction, worsening or new angina, other cardiovascular reason or noncardiovascular) was investigator-determined. For hospitalizations with more than one recorded primary cardiovascular cause of admission, a single cause was assigned based on rank ordering of the assumed seriousness of diagnoses as follows: acute myocardial infarction, worsening congestive heart failure, worsening or new angina and “other” cardiovascular reasons.
For hospitalizations during the open-label phase, only the primary hospitalization cause was recorded. Among 1,197 patients entering this phase, 57 hospitalizations were reported (of which 41 were classified as cardiovascular, including 26 for worsening heart failure).
Measures of interest included hospitalization frequency, numbers of hospital days in special-care units (e.g., ICU/CCU) and overall, utilization of selected cardiovascular procedures, and costs of cardiovascular-related in-patient care. A cost was assigned to each cardiovascular hospitalization using cost-prediction models that were estimated using a large database containing admission-level information from 153 U.S. short-term, acute-care hospitals (HCIA, Inc.) (22). Data available for each hospital discharge included all International Classification of Diseases, Ninth Revision (ICD-9-CM) diagnoses and procedure codes, time spent in various units (e.g., ICU/CCU), services used, and total hospitalization cost. Four distinct models were developed, corresponding to the four cardiovascular primary admission causes designated on the study case-report form.
To select a master sample of hospital admissions for model estimation, all hospital stays in the HCIA database with a listed principalor secondarydiagnosis of congestive heart failure (hypertensive heart disease, benign, with congestive heart failure, ICD-9-CMdiagnosis code 402.11; hypertensive heart disease, unspecified, with congestive heart failure, 402.91; congestive heart failure, 428.0; left heart failure, 428.1; or heart failure, unspecified, 428.9) and discharge date between July 1, 1993 and June 30, 1994 were identified. These were then grouped into mutually exclusive subsamples corresponding to each of the four primary admission causes noted on the study case-report form, according to the listed principal diagnosis.
Costs were estimated using admissions with listed principal diagnoses of: 1) congestive heart failure (ICD-9-CMdiagnosis codes 402.11, 402.91, 428.0, 428.1, 428.9) for hospitalizations for “worsening CHF;” 2) myocardial infarction (ICD-9-CMdiagnosis code 410) for hospitalizations for “myocardial infarction;” and 3) intermediate coronary syndrome (411.1), angina decubitus (413.0), Prinzmetal angina (413.1), or other and unspecified angina pectoris (413.9) for hospitalizations for “worsening or new angina.” Admissions for cardiovascular conditions other than those listed above were used to estimate costs of hospitalizations for “other” cardiovascular reasons.
Using multiple regression analysis, the four cost-prediction models were then estimated, and each was used to establish the relationship between inpatient treatment costs for a specific admission cause and selected explanatory variables, including the duration of routine- and special-care (i.e., ICU/CCU) unit stay, and use of the selected cardiac procedures described earlier. For each admission, the days spent in routine-care units were calculated by subtracting days spent in special-care units from the total of inpatient days. The models included two alternative sets of procedure-use variables. The first consisted of the total number of selected cardiac procedures, along with a binary variable indicating whether CABG, PTCA or heart transplantation was performed. The second included binary variables for each of these procedures, as well as a count of all other selected cardiac procedures. Hospitalization costs were adjusted to 1994 price levels using the Consumer Price Index for Medical Care (unpublished data, U.S. Bureau of Labor Statistics, 1995).
A variety of functional specifications were explored, including linear, quadratic, inverse, log-linear and log-log. The final models selected, based on R2values and visual inspection of residual plots, are shown in the Appendix. For all the selected models, >90% of the standardized residuals fell between −1.64 and 1.64, indicating approximately normal distribution. The estimated coefficients from these models were then combined with data from the carvedilol trials to estimate the cost of each cardiovascular hospitalization. For patients with multiple hospital admissions, costs were summed to yield a total cost of cardiovascular in-patient care.
All randomized patients were included in the analyses. Relative risk ratios and 95% confidence intervals (CIs) were estimated for differences between treatments in the occurrence of hospitalization at least once for any reason, for cardiovascular causes and for worsening heart failure. Differences in hospitalization risk were tested by the log-rank statistic using a Cox proportional hazards regression model (23). Kaplan-Meier methods were used to construct a life-table plot of death or hospitalization for any cause (24), and the difference between the curves was tested for significance by the method described above. Differences between treatments in the number of patients with no admissions, at least one admission, and two or more admissions were tested for significance using logistic regression. The relationships between hospitalization and subsequent death were assessed with the chi-square test according to Cochran-Mantel-Haenszel.
The mean numbers of cardiovascular admissions, hospital days, special-care unit days, and selected cardiovascular procedures, and the mean cost of cardiovascular hospitalization by treatment group were calculated by dividing the total for each measure by the number of randomized patients in each group. To estimate the average cost per stay among patients with one or more hospitalizations, an average was calculated across all admissions for each patient and the resulting mean was used as a single observation for that patient (i.e., each subject contributed a single value to the analysis, irrespective of number of hospital admissions).
Differences between treatments for resource-use outcomes were tested using the Wilcoxon rank-sum test (25); information on utilization of selected cardiovascular procedures was reported on a descriptive basis only. The extent to which a reduction in hospital costs was due to fewer admissions, or to reductions in cost per stay, was addressed using cost-variance analyses (26). Data on health care utilization measures (mean number of hospitalizations per patient, inpatient days and costs) are reported using univariate statistics.
The influence of baseline characteristics (i.e., age, gender, NYHA class, heart failure etiology and LVEF) on hospitalization risk was evaluated by specific interaction tests (27). Descriptive statistics on inpatient care costs were generated for these subgroups.
Randomization into the four trials began in April 1993 and stopped early in February 1995 because of a significant effect of carvedilol on survival. A total of 1,197 patients entered the open-label, run-in period, and 1,094 were randomized to double-blind treatment (398 to placebo, 696 to carvedilol). Pretreatment characteristics were similar in the two groups (17). Placebo-treated patients achieved target doses as frequently as carvedilol-treated patients (80% vs. 78%), but were more likely to discontinue treatment with double-blind medication for an adverse clinical event or outcome (13% for placebo vs. 8% for carvedilol, p = 0.009). Duration of therapy ranged from 1 day to 15.1 months (median, 6.5 months); 83% of placebo-treated patients and 89% of carvedilol-treated patients continued treatment with study medication until the end of the trial.
Effect of carvedilol on hospitalization risk
Compared with placebo, carvedilol reduced the risk of being hospitalized at least once for any reason by 29% (95% CI: 8% to 46%, p = 0.009), for cardiovascular causes by 28% (95% CI: 2% to 46%, p = 0.034) and for heart failure by 38% (95% CI: 2% to 61%, p = 0.041) (Table 1). The cumulative rate of death or hospitalization for any cause is shown in Figure 1(reduction in risk of 35% [95% CI: 16% to 49%, p = 0.0008]).
Carvedilol also reduced the risk of repeated hospitalization following randomization: by 35% for hospitalizations for any reason (95% CI: 14% to 52%, p = 0.003), by 32% for cardiovascular hospitalizations (95% CI: 6% to 50%, p = 0.02), and by 40% for heart failure hospitalizations (95% CI: 5% to 62%, p = 0.029) (Table 1). As a result, carvedilol decreased the mean number of hospitalizations per patient for any reason by 25% (p = 0.003), for cardiovascular reasons by 30% (p = 0.021) and for heart failure by 53% (p = 0.028) (Table 2).
There was no increase in cardiovascular or heart failure hospitalizations during the first three months after randomization in carvedilol-treated patients (Fig. 2). Carvedilol’s effects on the risk of being hospitalized at least once were not influenced by patient gender, age, baseline LVEF, heart failure severity or etiology.
Effect of carvedilol on the severity of illness during hospitalization
Carvedilol reduced the severity of illness during admission, as reflected by shorter length of stay and less need for specialized care. Carvedilol was associated with a reduction in mean number of hospital days per patient for cardiovascular reasons by 49% (1.56 vs. 3.08 days receiving placebo, p = 0.019) and for heart failure by 68% (0.54 vs. 1.67 days, p = 0.025). For each hospital admission, carvedilol was associated with a shorter length of stay (7.4 vs. 10.8 days receiving placebo [p = 0.298] for cardiovascular admissions; 6.8 vs. 10.8 days [p = 0.025] for heart failure admissions) (Table 3).
These findings were particularly marked for special-care unit use. For cardiovascular admissions, carvedilol decreased the mean number of days per patient spent in special-care units by 77% (0.33 vs. 1.46 days receiving placebo, p = 0.011) and for heart failure admissions by 90% (0.07 vs. 0.68 days, p < 0.001). For each admission, carvedilol was associated with shorter special-care unit stay (1.5 vs. 5.6 days receiving placebo for cardiovascular admissions [p = 0.049]; 0.7 vs. 4.3 days for heart failure admissions [p = 0.001]). For the latter, only 13% of hospital days in the carvedilol group were spent in special-care units, as compared with 41% of hospital days in the placebo group. When three placebo-treated patients whose special-care unit stay exceeded 60 days (88, 110 and 116 days, respectively) were excluded from the analysis, the differences between carvedilol and placebo were 1.5 versus 2.5 days (p = 0.09) for cardiovascular admissions and 0.7 versus 3.6 days for heart failure admissions (p = 0.001).
The utilization of selected cardiovascular procedures is reported for each treatment group in Table 4.
Effect of carvedilol on hospitalization costs
Total hospitalization cost was lower among patients receiving carvedilol (Table 3). For cardiovascular admissions, carvedilol reduced the mean per patient cost of hospitalization by 57% ($1,912 vs. $4,463 receiving placebo, p = 0.016) and for heart failure admissions by 81% ($452 vs. $2,338, p = 0.022). The cost for each hospital stay was also reduced: for cardiovascular causes by 43% ($9,318 vs. $16,426 receiving placebo, p = 0.097) and for heart failure by 63% ($5,632 vs. $15,258, p = 0.002). These differences remained significant when the three placebo “outliers” were excluded from the analysis. For cardiovascular hospitalizations, 55% of the savings reflected fewer admissions and 45% reflected reduced cost per stay; for heart failure admissions, these figures were 63% and 37%, respectively.
Effect of carvedilol on prognosis following hospitalization
Sixty patients died during the conduct of the program (inclusive of open-label challenge). Of these, 24 (40%) died following a cardiovascular admission and 36 (60%) without such a preceding admission. Patients having a cardiovascular hospitalization had a 67% higher mortality risk than patients not hospitalized (p < 0.001). Interestingly, 16% (13/80) of patients in the placebo group who had a cardiovascular hospitalization died, compared with 8% (11/131) of such patients in the carvedilol group (p = 0.08).
Our results support earlier work that carvedilol reduces the risk of clinical deterioration in patients with heart failure (17,18,28). Compared with placebo, adding carvedilol to conventional therapy decreased the risk of being hospitalized (once or multiple times) by 30% to 40% regardless of cause. Carvedilol also lessened the severity of illness among those hospitalized, as reflected by shorter length of stay and less need for specialized care. Specifically, carvedilol reduced the duration of heart failure admissions (6.8 days vs. 10.8 receiving placebo), particularly time spent in special-care units (0.7 days vs. 4.3 receiving placebo). Whereas nearly one half of the hospital days related to heart failure in the placebo group were spent in special-care units, this was true for less than one sixth of the hospital days in the carvedilol group. These benefits were likely related to carvedilol’s interference with pathophysiologic factors responsible for heart failure progression (29,30).
In addition, carvedilol also reduced the deleterious consequences of hospitalization, specifically the high cost and adverse impact on prognosis. Carvedilol decreased total in-patient care costs for heart failure by 81%, nearly two thirds of which reflected fewer admissions (which were 53% lower), with the remainder related to reduced costs for each stay. The mean estimated cost per admission was significantly lower in the carvedilol group ($5,632 vs. $15,258 for placebo), primarily due to decreased utilization of special-care units, the most expensive component of in-hospital care. Carvedilol also tended to diminish the prognostic implications of hospitalization for heart failure. Hospitalization occurrence and recurrence have portended a poor outcome for heart failure patients (9), and carvedilol lessened the adverse impact of hospitalization in the present trial. This was likely due to hospitalization representing a less serious event in the carvedilol group than in the placebo group, as reflected by shorter admissions and reduced requirement for intensive care. Since carvedilol reduces disease progression, continued therapy following hospitalization could explain carvedilol’s effect on the prognostic import of being hospitalized.
The benefits of carvedilol were apparent regardless of patient gender, age, LVEF, heart failure severity or etiology. In particular, elderly patients (>65 years) or those with heart failure due to ischemic disease responded as favorably to carvedilol (with respect to hospitalization risk and costs) as did younger patients or those with a nonischemic cardiomyopathy. Such observations are important because age and ischemic heart disease are important risk factors for the occurrence of a first hospitalization in heart failure patients, and indeed, the vast majority of heart failure hospitalizations occur in the elderly (8,31,32). Finally, beta-blockade in patients with ischemic heart disease may not only reduce the risk of heart failure progression, but also diminish the risk of recurrent ischemic events that might require or contribute to hospitalization (17,33). Therefore, it is noteworthy that carvedilol not only decreased the frequency of heart failure admissions but also admissions for any cardiovascular cause and for any reason whatsoever.
Limitations of analysis
Certain aspects of the present study could have led to underestimation or overestimation of carvedilol’s effects. Because hospitalizations were recorded only during double-blind therapy (with a median duration of 6.5 months), the greater withdrawal rate (for death and adverse reactions) for placebo-treated patients would have reduced the number at risk of hospitalization. Such bias was particularly likely since many patients receiving placebo discontinued treatment because of worsening heart failure and thus were at particular risk for subsequent hospitalization. The failure to include such events in the present analysis would have led to an underestimation of the magnitude of treatment effect. Conversely, some (but not all) of the hospitalizations during the open-label phase may have been precipitated by active treatment rather than underlying disease. If so, excluding these patients prior to randomization and thus from the present analysis would have led to an overestimation of treatment effect. However, even if all hospitalizations during the open-label period were attributed to carvedilol (an extremely conservative analysis), its effects on frequency of cardiovascular and heart failure hospitalizations remained statistically significant.
Our cost analyses should be interpreted cautiously for two reasons. First, in the absence of primary cost data, we estimated costs using cost-prediction models in conjunction with actual levels of resource use (see appendix). Therefore, we do not know if our estimates of in-patient cost reduction accurately reflect actual cost savings during the trial period. However, the approach we employed has been reported previously to be an accurate method of estimating in-patient care costs (34). Second, since our analysis focused only on in-patient resource utilization, it does not include the cost of carvedilol or costs that may result from additional clinic visits required to monitor the drug. Indeed, our analysis may have underestimated the potential benefit if, by retarding heart failure progression, carvedilol actually decreased outpatient visits during long-term treatment. However, the long-term effect on costs is complicated by the likelihood that any significant mortality reduction produced by carvedilol would eventually result in more patients at risk who may subsequently consume more health-care resources (35). These factors have been considered in a separate analysis, which showed that the cost-effectiveness of carvedilol compares favorably to that of other generally accepted medical interventions (36).
These results add to the growing evidence that many (if not most) hospitalizations in heart failure patients can be prevented (11). Previous reports have indicated that inadequate prescribing of effective medications by physicians, or lack of patient adherence with such medications, contributes importantly to hospitalization frequency (particularly true for ACE inhibitors) (37). The present analysis indicates that beta-blockade with carvedilol, in conjunction with ACE inhibition, should be added to the list of measures that can diminish hospitalization frequency and its impact on individual and public health.
We thank Sandra Lottes, PharmD, Boll Wu, PhD and Neil Shusterman, MD, for their technical assistance and support, and Ms. Ele Emanuel for her assistance in manuscript preparation.
Summary of cost-prediction models
|legend||Primary Reason for Hospital Admission|
|Worsening Heart Failure||Myocardial Infarction||New or Worsening Angina||Other|
|Model Coefficients (SEs)|
|Intercept||$899 (40)||$94 (208)||$144 (145)||$13 (122)|
|Routine-unit days (n)||552 (4)||654 (30)||607 (26)||749 (12)|
|Routine days squared (n)||—||10 (1)||6 (1)||—|
|Coronary care unit days (n)||2,093 (12)||2,022 (28)||1,866 (38)||2,255 (25)|
|Intensive-care unit days (n)||1,704 (16)||1,846 (30)||1,649 (29)||1,825 (32)|
|PTCA (%)||2,781 (503)||2,543 (366)||2,261 (372)||1,636 (1,778)|
|CABG (%)||13,809 (361)||10,721 (387)||11,726 (323)||12,188 (1,347)|
|Heart transplant (%)||53,222 (948)||32,417 (4,666)||65,267 (3,786)||22,271 (1,571)|
|Other cardiovascular procedures (n)||1,211 (48)||951 (104)||634 (95)||1,177 (103)|
↵legend CABG = coronary artery bypass grafting; PTCA = percutaneous transluminal coronary angioplasty.
☆ Supported by grants from SmithKline Beecham Pharmaceuticals and Roche Laboratories.
- angiotensin-converting enzyme
- coronary artery bypass grafting
- confidence interval
- intensive care unit/coronary care unit
- left ventricular ejection fraction
- New York Heart Association
- percutaneous transluminal coronary angioplasty
- Received November 12, 1999.
- Revision received November 3, 2000.
- Accepted January 12, 2001.
- American College of Cardiology
- Ward R.E,
- Gheorghiade M,
- Young J.B,
- Uretsky B
- CIBIS Investigators and Committees
- Hjalmarson A,
- Goldstein S,
- Fagerberg B,
- Wedel H,
- Waagstein F
- ↵Health Care Investments Analysts Inc. Special Study. Congestive Heart Failure Data Dump. Baltimore, MD: August 1995.
- Cox D.R
- Hollander M,
- Wolfe D
- Cleverley W.O
- Mantel N,
- Haenszel W
- Packer M
- Ruffolo R.R Jr.,
- Feuerstein G.Z
- Wolinsky F.D,
- Overhage J.M,
- Stump T.E,
- Lubitz R.M,
- Smith D.M
- ISAM Study Group,
- Willich S.N,
- Linderer T,
- Wegscheider K,
- et al.
- Oster G,
- Huse D.M,
- Lacey M.J,
- et al.
- Cleland J.G.F