Cost-Effectiveness of Fixed-Dose Combination of Isosorbide Dinitrate and Hydralazine Therapy for Blacks With Heart Failure
Background— Fixed-dose combination of isosorbide dinitrate/hydralazine (ISDN/HYD) improved clinical outcomes in the African-American Heart Failure Trial (A-HeFT). We assessed the resource use, costs of care, and cost-effectiveness of ISDN/HYD therapy in the A-HeFT trial population.
Methods and Results— We obtained resource use data from A-HeFT, assigning costs through the use of US federal sources. Excluding indirect costs, we summarized the within-trial experience and modeled cost-effectiveness over extended time horizons, including a US societal lifetime reference case. During the mean trial follow-up of 12.8 months, the ISDN/HYD group incurred fewer heart failure–related hospitalizations (0.33 versus 0.47 per subject; P=0.002) and shorter mean hospital stays (6.7 versus 7.9 days; P=0.006). When study drug costs were excluded, both heart failure–related and total healthcare costs were lower in the ISDN/HYD group (mean per-subject heart failure–related costs, $5997 versus $9144; P=0.04; mean per-subject total healthcare costs, $15 384 versus $19 728; P=0.03). With an average daily drug cost of $6.38, ISDN/HYD therapy was dominant (reduced costs and improved outcomes) over the trial duration. Assuming that no additional benefits accrue beyond the trial, we project the cost-effectiveness of ISDN/HYD therapy using heart failure–related costs to be $16 600/life-year at 2 years after enrollment, $37 100/life-year at 5 years, and $41 800/life-year over lifetime (reference case).
Conclusions— ISDN/HYD therapy, previously shown to improve clinical outcomes, also reduced resource use and costs in A-HeFT, primarily because of a large reduction in hospitalizations. Long-term use of ISDN/HYD therapy should be associated with a favorable cost-effectiveness profile in this population.
Received May 23, 2005; revision received September 14, 2005; accepted September 16, 2005.
Despite advances in care such as β-blockade and renin-angiotensin inhibition, chronic heart failure remains a major health problem, shortening survival, impairing quality of life, and requiring frequent hospitalizations. In 2005, the costs of caring for Americans with heart failure were estimated at $28 billion.1 Several studies suggest that the prevalence, severity, and outcome of heart failure are worse in blacks, partly because of differences in etiology and response to therapy.2–5 Recently, the African-American Heart Failure Trial (A-HeFT) demonstrated that fixed-dose combination of isosorbide dinitrate (ISDN) and hydralazine (HYD) improved clinical outcomes in self-identified blacks with moderate to severe heart failure who were already receiving current standard of care.6 The purpose of the present study was to describe healthcare resource use in A-HeFT and to determine the incremental costs and cost-effectiveness of ISDN/HYD fixed-dose combination therapy.
We based our cost and cost-effectiveness analyses primarily on clinical and resource use data collected in A-HeFT, capitalizing on the benefits of randomization. These data were supplemented with cost weights from external sources. We also generated a long-term reference case based on lifetime estimates of costs and effects as per the guidelines of the US Public Health Service Panel on Cost-Effectiveness in Health and Medicine7 under a set of conservative assumptions about duration of survival and treatment effect. We conducted our analysis from the US societal perspective, including direct healthcare costs and excluding indirect and nonhealthcare costs. We expressed costs in 2004 US dollars, discounting future costs and benefits at 3% per annum.
Subjects and Trial Procedures
A-HeFT was a placebo-controlled, multicenter, randomized phase III trial of 1050 subjects enrolled at 169 sites in the United States.6 To be enrolled, subjects had to be ≥18 years of age, identify themselves as black, and have New York Heart Association (NYHA) class III or IV heart failure with echocardiographic evidence of impaired left ventricular function (ejection fraction ≤35% or <45% with dilated internal end-diastolic diameter). Subjects were required to receive standard therapy for heart failure such as neurohormonal inhibition with β-blockers, angiotensin-converting enzyme (ACE) inhibitors or angiotensin receptor blockers, aldosterone antagonist, and use of digoxin and diuretics as appropriate. ISDN/HYD therapy was prescribed as tablets containing 20 mg ISDN and 37.5 mg HYD, starting at 1 tablet 3 times daily and titrating to 2 tablets 3 times daily as tolerated. Subjects were followed up for up to 18 months with follow-up visits every 3 months and monthly telephone interviews. The primary end point was a weighted composite of all-cause mortality, quality of life, and hospitalization for heart failure. The study was approved by the institutional review committees of the participating institutions, and subjects gave informed consent. Enrollment ran from June 2001 through July 2004, when the large treatment effect triggered the independent data monitoring board’s recommendation to stop the trial.6
Hospitalization information was collected during the monthly phone interviews and study visits and verified by review of office and hospital records. Dates of admission and discharge were collected, and reason for hospitalization was independently and blindly adjudicated as heart failure related or not on the basis of documentation from the subject’s medical record. Death during hospitalization was determined by comparing date of death with hospital discharge date. A-HeFT also collected data on unscheduled physician office visits and emergency room (ER) visits. The reason for the visit was also adjudicated as related to heart failure or not. Information on all concomitant medications and their changes during the trial was recorded, including name of medication, daily dose, start and stop dates, and the reason for the medication.
Calculation of Resource Use and Costs
We used estimated per-subject Medicare costs as a proxy for societal per-subject direct healthcare costs. Using the actual length of stay (LOS) for each hospitalization recorded in A-HeFT, we assigned each hospitalization the mean cost for a discharge with the same LOS, gender, and discharge disposition (alive or dead) for blacks with heart failure in the 2003 Medicare hospital discharge database.8 We obtained costs from the Medicare database by multiplying recorded charges by the Medicare hospital-specific cost-to-charge ratio. We then adjusted to the year 2004 using the medical consumer price index9 and, on the basis of prior methods,10,11 increased costs by an additional 17% to account for physician fees. We created tables of the cost distributions for hospitalizations as a function of LOS, gender, and disposition for use in bootstrap simulations from the same database.
We assigned a cost for unscheduled office and ER visits from the average Medicare payments for such visits by blacks in the 2003 Medicare part B 5% data set, again adjusting to the year 2004.12 We estimated the cost of each concomitant medication by multiplying duration of therapy, daily dose in milligrams, and the price per milligram from the average wholesale price in the 2004 Red Book.13 We estimated the daily costs of scheduled office visits and other medical care from the average daily payments under Medicare part B for heart failure– and non–heart failure–related care. We used the manufacturer’s (NitroMed Inc) announced price of $1.80 per tablet for the fixed-dose combination of ISDN/HYD, along with the average prescribed dose and compliance from the trial, to generate the study drug cost.
Generating “Within-Trial” Cost-Effectiveness Estimates
We generated within-trial cost-effectiveness ratios by dividing the difference in within-trial costs between ISDN/HYD and placebo groups by the difference in mean survival time between treatment groups. To evaluate uncertainty associated with the cost-effectiveness ratios, we performed 10 000 bootstrap replications using data from the 1050 subjects from A-HeFT. This approach minimizes the effects of outliers and permits assessment of the robustness of our point estimates to random variations in costs and survival.
Generating Lifetime Cost-Effectiveness Estimates
Within-trial estimates were extended beyond the period of follow-up to calculate incremental cost-effectiveness ratios over longer time horizons. In the absence of longer-term data, we conservatively assumed that any differences in survival and resource use were accrued only during the observation period of the trial. Thus, for survival beyond the end of follow-up, we assumed that both the ISDN/HYD and placebo arm survival curves would decay at the same rate. We generated the survival curve by relying on the observed long-term survival from a contemporary multicenter US cohort of subjects with NYHA class III heart failure.14 This cohort had a 5-year survival near 50% with a linear decay, which we extrapolated to zero survival at 10 years. We also assumed that the rate at which subjects consumed resources beyond A-HeFT would be the same in both arms and equal to that observed in the placebo group during A-HeFT. Thus, the main consequences beyond study follow-up would be that all subjects in the ISDN/HYD arm would receive the additional costs of drug therapy until death, without benefit, and that the additional survivors noted at end of follow-up would gradually die at the same rate as all other survivors. We generated the reference case using only heart failure–related costs as per the Panel on Cost-Effectiveness in Health and Medicine guidelines.15
To test the sensitivity of the within-trial cost-effectiveness estimates to assumed values, we varied our estimates of hospital costs and the cost of ISDN/HYD by ±50% and the costs of concomitant medications, unscheduled office visits, ER visits, and other usual medical care from +100% to −50%. Because ISDN/HYD therapy was dominant over a wide range of prices, the result that was explored was the savings in incremental costs with ISDN/HYD. Sensitivity of the within-trial effect is displayed through the distribution of bootstrap simulations. We also explored whether there were differences in within-trial cost-effectiveness across a number of the prospectively defined subgroups in A-HeFT: Age <65 or ≥65 years, heart failure cause, left ventricular ejection fraction (<20%, 20% to 30%, >30%), baseline ACE inhibitor use, baseline β-blocker use, and diabetes mellitus. We also explored the cost-effectiveness of ISDN/HYD in the subgroup receiving intense concomitant therapy (ie, combined β-blocker, ACE inhibitor or angiotensin receptor blocker, digoxin, and spironolactone treatment). We explored the sensitivity of our lifetime cost-effectiveness estimates to continuation of treatment effect by assuming that the benefit of ISDN/HYD therapy on costs and outcomes continued for 1 year beyond the duration of follow-up in A-HeFT, at which point the treatment group would assume the resource use and survival of the placebo group. We further explored the robustness of the cost-effectiveness to other measures of hospital cost by using the average total cost unadjusted for gender, LOS, and discharge status for blacks in the Medicare population with a principal diagnosis of heart failure and adding 17% for physician-billed care,10,11 which yielded an average cost per hospitalization of $10 050.
We used the Wilcoxon rank-sum test to compare continuous event data and the χ2 test to compare categorical data as appropriate. We analyzed survival by Kaplan-Meier method, testing differences by treatment group in survival times with the log-rank test. The distributions around the within-trial costs, survival, and cost-effectiveness for the entire cohort and for each subgroup were estimated by generating 10 000 bootstrap replications.16,17 Variance in hospitalization cost estimates was accounted for in a 2-stage process. First, variation in resource use with patient characteristics was accounted for in the sampling of subjects in each bootstrap simulation. Second, variation in hospital costs associated with such resource use was incorporated by randomly sampling the Medicare heart failure cost distributions by LOS, gender, and survival for each hospitalization incurred by the subject. These costs were then summed across all hospitalizations. We generated 95% confidence intervals (CIs) and probability of incremental cost savings from the bootstrap output. We expressed the output of the analysis as summaries of the incremental costs and effectiveness and by scatterplot of the simulations with Fieller’s 25%, 50%, and 75% confidence ellipses.18 We managed data with Excel and FoxPro (Microsoft Corp), and we conducted statistical analyses and modeling in Datadesk (Data Description).
Subject characteristics were reported previously.6 Briefly, the study randomized 518 subjects to ISDN/HYD and 532 to placebo. The average age was 56.8 years; 40.0% were women; 94.9% had NYHA class III heart failure; 93% were on ACE inhibitors or angiotensin receptor blockers; and 87% were on β-blockers. Baseline characteristics were similar across arms. The average prescribed dose was 4.2 tablets per day, with 68% of the treatment arm taking the full dose of 6 tablets at some time in the trial. Adherence was high and similar for ISDN/HYD and placebo (84.6% versus 85.2%; P=0.45). Mean follow-up was 12.8 months. The ISDN/HYD group had lower mortality than the placebo group (6.2% versus 10.2%; P=0.016) and corresponding longer survival time (403 versus 380 days; P=0.01).
Resource Use and Costs
Measured resource use is provided in Table 1. Of the 994 hospitalizations recorded during the trial, 43% were considered to be related to heart failure by the blinded adjudication process. Subjects in the ISDN/HYD group had 30% fewer heart failure–related hospitalizations than the placebo group and a 1-day reduction in the average LOS for each heart failure–related hospitalization, yielding a 41% decrease in heart failure–related hospital days and a number needed to treat of 7.6 to avoid 1 heart failure–related hospitalization. Nearly three quarters (73%) of the reduction in hospital days was due to the reduction in admissions; the remaining 27% was due to the reduction in average LOS. Of note, the decreased frequency of heart failure–related hospitalizations was a broad effect across the ISDN/HYD group, with more subjects having no hospitalizations and fewer subjects having multiple hospitalizations compared with placebo. The number of all-cause hospitalizations was not significantly different between groups, but the average LOS for all-cause hospitalizations was shorter in the ISDN/HYD group. Subjects in the ISDN/HYD group incurred fewer hospitalizations throughout the study follow-up period, as demonstrated by the widening differences over time in the cumulative distribution of hospitalizations shown in Figure 1. Of the 549 unscheduled office and ER visits, 13.7% were heart failure related, with no difference between groups. There were also no differences in the use of concomitant medications.
Estimated within-trial costs of care are provided in Table 2. Combining across all resource categories, hospital costs were three quarters of all heart failure–related costs and two thirds of all healthcare costs. When study drug costs were excluded, heart failure–related costs were 34% lower ($3147 cost savings; 95% CI, $5650 cost savings to $275 cost outlay; P=0.04), and total healthcare costs were 22% lower ($4344 cost savings; 95% CI, $7,250 cost savings to $150 cost outlay; P=0.03) in the ISDN/HYD group compared with placebo.
The average prescribed daily dose of ISDN/HYD was 4.19 tablets per day, and average compliance to prescribed therapy was similar for fixed-dose ISDN/HYD and placebo (84.6% versus 85.2%, respectively). Using the announced price per tablet of $1.80 and the average prescribed dose and compliance observed in the trial, we found the average cost per day for treatment drug to be $6.38. When study drug costs were included, heart failure–related costs were 6% lower ($533 cost savings; 95% CI, $2422 cost savings to $2241 cost outlay; P=0.36), and total healthcare costs were 9% lower ($1730 cost savings; 95% CI, $4107 cost savings to $1751 cost outlay; P=0.19) in the ISDN/HYD group compared with placebo.
Figure 2 shows scatterplots of the bootstrap sampling of the within-trial cost-effectiveness. Assuming an ISDN/HYD cost of $6.38 per day, ISDN/HYD therapy was dominant (ie, improved survival and saved costs) using both heart failure–related and total direct healthcare costs. The average added survival was 22.9 days per subject (95% CI, 0.0 to 43.6). For heart failure–related costs, 49% of simulations were dominant, 66% were better than $10 000/life-year gained, and 92% were better than $50 000/life-year gained. For all healthcare costs, 71% of simulations were dominant, 82% were better than $10 000/life-year gained, and 95% were better than $50 000/life-year gained.
The inputs and outputs for the extended time horizon model are shown in Table 3 through Table 5⇓⇓. The average modeled survival times were 5.33 and 5.07 years for the ISDN/HYD and placebo groups. With the assumption that there are no additional benefits of ISDN/HYD therapy beyond the duration of A-HeFT, the cost-effectiveness ratios gradually deteriorate because of the ongoing costs of the drug, with a lifetime reference case estimate of $41 800/life-year. If ISDN/HYD therapy has benefit beyond the duration of A-HeFT, the cost-effectiveness improves. For example, with 1 additional year of benefit beyond the trial time horizon, the reference case estimate is $22 900/life-year.
Incremental costs were most sensitive to hospital costs and ISDN/HYD drug price (Figure 3): Higher hospital costs increased the projected cost savings of ISDN/HYD therapy, and higher ISDN/HYD drug costs necessarily decreased cost savings. Using the Medicare heart failure–related hospitalization cost estimate unadjusted for race, gender, and LOS, we found that ISDN/HYD therapy was no longer dominant but resulted in an incremental cost of $615 per subject for heart failure cost and $63 for overall healthcare costs. These estimates yielded a heart failure–related cost-effectiveness ratio of $10 335/life-year, with 24% of simulations dominant, 49% costing less than $10 000, and 92% costing <$50 000/life-year. Analyses that considered all healthcare costs yielded a ratio of $1546/life-year, with 46% of simulations dominant, 66% costing less than $10 000, and 94% costing <$50 000/life-year. When only heart failure–related costs were considered, ISDN/HYD therapy was dominant up to a daily per-subject cost of $8.05. When all direct healthcare costs were considered, ISDN/HYD therapy was dominant up to daily subject costs of $12.00. Above these thresholds, the within-trial cost-effectiveness ratios increase by $6400/life-year for each $1 increase in daily drug cost. Box plots for the cost-effectiveness ratios in different subgroups are shown in Figure 4. The whiskers were wide because of the smaller sample sizes, but the central estimates were generally similar to each other and to the overall trial estimates.
We found that use of ISDN/HYD therapy in A-HeFT led to fewer hospitalizations, shorter hospitalizations, and consequently lower healthcare costs. Combined with the improved survival, these observations suggest that ISDN/HYD therapy should be a dominant strategy, at least over a short time horizon. We also projected that longer use of this therapy is still likely to be associated with a favorable cost-effectiveness profile, even assuming no long-term clinical benefits. In other words, adoption of this strategy seems prudent for blacks with moderate to severe heart failure.
There are, however, key caveats to the interpretation of our findings. First, our observations arise from a carefully conducted clinical trial. This is an important strength for the internal validity of our findings (ie, the robustness of the treatment effect on resource use in the trial), but it may be a weakness with regard to external validity. In many areas of medicine, the observed effect of an intervention in clinical trials (efficacy) is greater than that observed subsequently when the therapy is disseminated into routine clinical practice (effectiveness). Reasons include differences in patient selection, use of concomitant therapies, and use of the proposed intervention.
With regard to patient selection, most subjects in A-HeFT had NYHA class III heart failure, and we cannot assume that the therapy would yield the same gains if prescribed to subjects with milder or more severe degrees of heart failure. In addition, the decision to limit the trial to individuals who identified themselves as black was based on retrospective subgroup analyses of previous heart failure trials,19,20 coupled with evidence of racial differences in endothelial function and nitric oxide bioavailability,4,5 which may potentially lead to differences in response to alternative pharmacological interventions. However, this line of argument relies on potentially arbitrary definitions of race.21 Thus, other groups of individuals who identify themselves as black may not have the same pharmacological response as those enrolled in A-HeFT, and by extension, clinical benefits may not be restricted to blacks.
With regard to concomitant medications, subjects enrolled in A-HeFT were arguably receiving a high standard of care, with most receiving neurohormonal inhibition. However, we did not ascertain whether concomitant medications were prescribed at target or maximally tolerated doses. In clinical practice, the use of concomitant heart failure medications may differ, which may result in smaller or larger benefits with ISDN/HYD therapy. During the trial, adherence to study drug was high, possibly because of the frequent contact between investigators and subjects or a desire of the subjects and their physicians to be cooperative with study procedures. In the absence of these factors, adherence and benefits may be less.
Another caveat relates to drug price. A-HeFT studied a patented fixed-dose combination of 2 therapies. Using a daily cost of $6.38, we found that the strategy was favorable under a variety of scenarios. Because the 2 components of therapy are available generically at nominal cost, replicating the fixed-dose combination with generic components would produce an even better cost-effectiveness ratio. However, the generic strategy would require twice as many pills, which could adversely affect compliance in patients already receiving a complex medical regimen,22,23 in turn diminishing clinical benefit and cost-effectiveness.
There are also a number of methodological caveats. Although other cost-effectiveness studies in heart failure relied on counts of hospitalizations with external cost estimates from broad unrelated populations, we based cost estimates on collected data on LOS, gender, and disposition. Similar to the work of Reed et al24 on the Valsartan Heart Failure Trial, this approach captures the added cost of terminal hospitalizations, differences by gender, and the importance of differences in costs by LOS. Of note, one consequence of this approach was that our estimated costs were considerably higher than the average reimbursement for diagnosis-related group 127, heart failure. Some possible explanations for this difference are that reimbursement might not fully cover costs and that blacks are more likely to receive care in urban teaching hospitals with greater associated costs. Furthermore, A-HeFT subjects had relatively severe heart failure and thus incurred hospitalizations of longer duration and higher intensity than average. Nevertheless, sensitivity analysis demonstrated that the cost-effectiveness profile remained similar even if hospitalizations were assumed to be considerably less expensive. A number of healthcare costs, including skilled nursing care use, lost productivity, and informal caregiver support, were not included in our analysis. Given the general trend to decreased healthcare resource use and improved clinical outcomes, it seems likely that these other costs would be greater in the placebo arm; thus, their inclusion would have amplified the favorable profile of ISDN/HYD therapy. Economic analyses frequently express cost-effectiveness ratios as costs per utility such as quality-adjusted life-years. Quality of life was assessed in A-HeFT with the Minnesota Living With Heart Failure survey, for which no validated approach to generate utilities exists. Nevertheless, subjects in the ISDN/HYD arm reported better quality of life, so adjustment for quality of life also would have magnified the favorable economic profile of ISDN/HYD therapy.
In summary, treatment with fixed-dose combination of ISDN and HYD, previously shown to have improved clinical outcomes in blacks with advanced heart failure, also reduced heart failure–related and all healthcare resource use and costs in A-HeFT, primarily because of a large reduction in heart failure–related hospitalizations. Excluding the cost for study drug, heart failure–related costs were decreased by 34% and total healthcare costs were decreased by 22%. Even extending to a projected lifetime horizon under the conservative assumption that there would be no benefit beyond the trial, this therapy provided a reference case cost-effectiveness that is still very favorable. Long-term use of ISDN/HYD therapy should be associated with a favorable cost-effectiveness profile in this population.
This work was supported by NitroMed, Inc, Lexington, Mass. The sponsor assisted with the collection of clinical data but played no role in the study design or in the analysis and interpretation of data. The authors performed all analyses, and the sponsor had no editorial control of the manuscript. We thank Dr Shelby Reed, Center for Clinical and Genetic Economics, Duke University Medical Center, for critical review of the manuscript.
Dr Angus served as a consultant for ZD Associates. W.T. Linde-Zwirble has a significant consulting relationship with NitroMed, Inc. Drs Tam, Sabolinski, and Worcel are employees of and hold stock in NitroMed, Inc. Dr Ghali has received research grants from and has served as a consultant to and on the speaker’s bureau of Nitromed, Inc. Dr Winkelmayer has served as a consultant to ZD Associates. Dr Villagra reports no conflicts.
Guest editor for this article was Robert O. Bonow, MD.
American Heart Association. Heart Disease and Stroke Statistics: 2005 Update. Available at: http://www.americanheart.org/downloadable/heart/11050481905292005Statcharts.ppt. Last update: January 2005. Accessed May 18, 2005.
Kalinowski L, Dobrucki IT, Malinski T. Race-specific differences in endothelial function: predisposition of African Americans to vascular diseases. Circulation. 2004; 109: 2511–2517.
2003 Medicare Hospital Discharge Database. Washington, DC: DHHS; 2004.
US Department of Labor, Bureau of Labor Statistics. Consumer Price Indexes. Available at: http://www.bls.gov/cpi/home.htm. Last update: May 2005. Accessed: May 20, 2005.
Ellis SG, Miller DP, Brown KJ, Omoigui N, Howell GL, Kutner M, Topol EJ. In-hospital cost of percutaneous coronary revascularization: critical determinants and implications. Circulation. 1995; 92: 741–747.
Medicare Utilization for Part B (Supplementary Medical Insurance SMI). Baltimore, Md: Centers for Medicare and Medicaid Services; 2004.
2004 Drug Topics Red Book. Montvale, NJ: Thompson Medical Economics; 2004.
Bardy GH, Lee KL, Mark DB, Poole JE, Packer DL, Boineau R, Domanski M, Troutman C, Anderson J, Johnson G, McNulty SE, Clapp-Channing N, Davidson-Ray LD, Fraulo ES, Fishbein DP, Luceri RM, Ip JH. Amiodarone or an implantable cardioverter-defibrillator for congestive heart failure. N Engl J Med. 2005; 352: 225–237.
Cheng RCH. Bootstrap methods in computer simulation experiments. In: Lilegdon WR, Goldsman D, Goldsman C, Alexopoulos C, Kang K, eds. Proceedings of the 1995 Winter Conference on Simulation (WSC ’95, Arlington, VA, Dec. 3–6). New York, NY: ACM Press; 1995.
Fieller EC. Some problems in interval estimation. J Royal Stat Soc Series B (Methodological). 1954; 16: 175–183.