Abstract
Background
Prognostic models, such as the Seattle Heart Failure Model (SHFM), have been developed to predict patient survival. The extent to which they predict medical resource use and costs has not been explored. In this study, we evaluated relationships between baseline SHFM scores and 1-year resource use and costs using data from a clinical trial.
Methods and Results
We applied generalized linear models to examine the relative impact of a 1-unit increase in SHFM scores on counts of medical resource use and direct medical costs at 1 year of follow-up. Of 2331 randomized patients, 2288 (98%) had a rounded integer SHFM score between –1 and 2, consistent with predicted 1-year survival of 98% and 74%, respectively. At baseline, median age was 59 years, 28% of patients were women, and nearly two-thirds of the cohort had New York Heart Association class II heart failure and one-third had class III heart failure. Higher SHFM scores were associated with more hospitalizations (rate ratio per 1-unit increase, 1.86; P < .001), more inpatient days (2.30; P < .001), and higher inpatient costs (2.28; P < .001), outpatient costs (1.54; P < .001), and total medical costs (2.13; P < .001).
Conclusion
Although developed to predict all-cause mortality, SHFM scores also predict medical resource use and costs.
Trial Registration
clinicaltrials.gov Identifier: NCT00047437.
Keywords: Health Care Costs, Health Resources, Health Status, Heart Failure, Questionnaires, Risk Assessment
Introduction
The direct and indirect costs of heart failure in the United States were estimated at more than $39.2 billion in 2010.1 Heart failure primarily affects older persons and is the leading cause of hospitalization in persons 65 years and older. Approximately 1 in 5 Medicare beneficiaries hospitalized for heart failure requires additional inpatient care within 30 days after discharge.2 An analysis of Medicare data revealed that typical inpatient care costs for a patient with severe heart failure were approximately $24,000 per year, compared to about $3000 for a typical Medicare beneficiary.3
To facilitate assessment of the costs and efficiency of disease management programs for heart failure, our group is developing a Web-based probabilistic disease simulation model that researchers and health care managers can use to evaluate the cost-effectiveness of specific heart failure disease management programs. The Seattle Heart Failure Model (SHFM), developed and externally validated in several large cohorts, is integrated into this model to predict survival based on clinical characteristics, evidence-based treatments, and laboratory measures.4–6 We sought to evaluate whether the SHFM could also be used to predict medical resource use and thereby estimate direct medical costs.
Data from a recently completed large-scale clinical trial of exercise training in outpatients with heart failure provided an opportunity to use the prospectively collected clinical data to examine these relationships.7 We hypothesized that patients with heart failure and higher SHFM scores incurred more hospitalizations and inpatient days and higher medical costs than patients with lower SHFM scores.
Methods
Study Population
The Heart Failure: A Controlled Trial Investigating Outcomes of Exercise Training (HF-ACTION) trial randomly assigned patients to an exercise training program plus usual care or to usual care alone.7 Between April 2003 and February 2007, 2331 patients were enrolled from 82 study sites in the United States (n = 2068), Canada (n = 188), and France (n = 75) for a median follow-up period of 2.5 years. The trial collected a wide range of data on patients’ clinical and treatment characteristics and laboratory measures. It demonstrated that exercise training was safe and produced nonsignificant reductions in the primary end point, a composite of all-cause mortality or all-cause hospitalization, which was observed in 65% of patients randomly assigned to receive exercise training and usual care vs 68% of patients assigned to receive usual care alone (hazard ratio [HR], 0.93; 95% CI, 0.84 to 1.02; P = .13).8 The economic evaluation demonstrated that measures of resource use and costs were similar between the groups, and the total direct medical costs per participant were estimated at $50,857 (SD, $81,488) in the exercise group and $56,177 (SD, $92,749) in the usual care group (95% CI for the difference, −$12,755 to $1547; P = .10).9
Seattle Heart Failure Model Scores
The SHFM incorporates 20 variables representing patients’ demographic, clinical, and treatment and laboratory characteristics. Of these, 3 variables (ie, lymphocytes, uric acid, and allopurinol use) were not collected in HF-ACTION. To compute the SHFM scores, we generated patient-level predicted values for lymphocytes and uric acid using regression models developed using the original SHFM cohort,4 and we assumed no allopurinol use. For patients with missing laboratory values for cholesterol (35%), hemoglobin (24%), or sodium (11%), we imputed the data using mean values from HF-ACTION participants with nonmissing data. SHFM scores for HF-ACTION participants were generated using the original published equation.4 We limited the analysis to patients with a rounded SHFM score between –1 and 2.
Medical Resource Use and Costs
The HF-ACTION case report form was designed to collect extensive medical resource use data every 3 months for the first 2 years of follow-up and yearly thereafter for all participants. Medical resources included all-cause hospitalizations, including length of stay and inpatient procedures performed, urgent and emergent care visits, and nonurgent outpatient visits and procedures.
Although data on medical resource use were collected beyond 1 year, we chose to examine relationships between SHFM scores and resource use in the first year of follow-up only to reduce variance attributable to changes in patients’ prognoses (ie, higher SHFM scores) over time. The costing methods were described in detail previously.9 As part of the economic evaluation, hospital billing data were collected for more than 80% of hospitalizations and emergency department visits that occurred during the follow-up period. We converted department-level hospital charges to costs using department-specific cost-to-charge ratios derived from each hospital’s annual Medicare cost report. For hospitalizations without available bills, we imputed costs by multiplying the length of stay for each hospitalization by estimates of median daily costs that corresponded to 1 of 47 potential reasons for hospitalization. To avoid overestimating costs for procedure-based stays typically characterized by high costs and short stays (eg, percutaneous coronary revascularization and device implant), we assigned the median total costs associated with hospitalizations for these procedures. Then, for patients with stays that extended beyond the median length of stay for each procedure type, we applied the median daily cost for heart failure, $1202, to all remaining hospital days. We used the 2008 Medicare Physician Fee Schedule to assign costs for physician services, including inpatient services, outpatient visits, and inpatient and outpatient procedures. For hospitalizations that continued beyond the 1-year time horizon for the analysis, we calculated the daily cost of inpatient care by dividing the total inpatient cost by the length of stay for the admission and summed daily costs incurred up to 1 year beyond randomization. For emergency department visits without available bills, we applied the median cost calculated from the available bills, $479, as a proxy. We valued all costs in 2008 US dollars.
Statistical Analysis
For reporting purposes, we rounded SHFM scores to the nearest integer and grouped patients accordingly. We present baseline demographic and clinical variables and 1-year medical resource use and costs for each group. We report frequencies and percentages for categorical variables and means and SDs for continuous variables. We used means and SDs to describe counts of medical resource use and costs during the first year of follow-up. Given the right-skewed distributions of resource use and costs, we also report medians and interquartile ranges.
In the statistical analysis, we analyzed SHFM score as a continuous variable. We used analysis of variance to evaluate relationships between categorical baseline variables and continuous SHFM scores and Pearson correlation tests to evaluate relationships between continuous baseline variables and SHFM scores. We applied generalized linear models to evaluate relationships between SHFM scores and medical resource use and costs during the first year of follow-up. Specifically, we applied logit links and binomial error distributions to model the proportions of patients who used each type of medical resource. We applied log links and negative binomial error distributions to examine the relative associations between SHFM scores and 1-year counts of medical resource use. We applied log links and gamma error distributions to model 1-year medical costs. In each of the models, we included an offset variable to adjust differential follow-up periods in the first study year. We also performed a sensitivity analysis of costs in which we excluded patients who died during the first year of follow-up. We used SAS version 9.2 (SAS Institute, Cary, North Carolina) for all analyses.
Results
Among the 2331 patients randomly assigned in HF-ACTION, SHFM scores were available for 2293 (98%), of whom 5 (0.2%) had a rounded SHFM score greater than 2 and were excluded from subsequent analyses. Table 1 shows the baseline characteristics of the 2288 patients in the final study cohort. The mean baseline SHFM score was 0.24 (SD, 0.64; median, 0.18; interquartile range, –0.20 to 0.63), corresponding to a predicted 1-year survival of 95%. When we rounded the SHFM scores to the nearest integer, 11.5% of patients had a score of –1, 57.3% had a score of 0, 27.1% had a score of 1, and 3.9% had a score of 2, corresponding to predicted 1-year mortality of 1.5%, 4.0%, 10.5% and 26.0%, respectively. At 1 year, 92.4% of patients were alive and participating, 4.7% had died, and 2.8% had discontinued participation.
Table 1.
Baseline Characteristics of the Study Population by Seattle Heart Failure Model Score
| Characteristic | All Patients (N = 2288) | Rounded Seattle Heart Failure Model Score | P Valuea | |||
|---|---|---|---|---|---|---|
| –1 (n = 264) | 0 (n = 1313) | 1 (n = 621) | 2 (n = 90) | |||
| Predicted mortality rate at 1 year, % | 1.5 | 4.0 | 10.5 | 26.0 | ||
| Age, mean (SD) | 59 (13) | 53 (12) | 58 (12) | 63 (12) | 68 (12) | < .001 |
| Sex, No. (%) | < .001 | |||||
| Female | 648 (28) | 87 (33) | 410 (31) | 143 (23) | 8 (8.9) | |
| Male | 1640 (72) | 177 (67) | 903 (69) | 478 (77) | 82 (91) | |
| Race, No. (%) | .02 | |||||
| Black or African American | 739 (32) | 92 (35) | 424 (32) | 196 (32) | 27 (30) | |
| White | 1394 (61) | 156 (59) | 789 (60) | 390 (63) | 59 (66) | |
| Other or missing | 155 (7) | 16 (6.1) | 100 (7.6) | 35 (5.6) | 4 (4.4) | |
| NYHA classification, No. (%) | < .001 | |||||
| II | 1457 (64) | 249 (94) | 978 (74) | 219 (35) | 11 (12) | |
| III | 811 (35) | 15 (5.7) | 335 (26) | 394 (63) | 67 (74) | |
| IV | 20 (0.9) | — | — | 8 (1.3) | 12 (13) | |
| Ischemic etiology, No. (%) | 1176 (51) | 66 (25) | 634 (48) | 406 (65) | 70 (78) | < .001 |
| Ejection fraction, mean (SD), % | 25.2 (7.5) | 26.7 (6.7) | 25.8 (7.6) | 23.5 (7.1) | 23.4 (7.5) | < .001 |
| Systolic blood pressure, mean (SD), mm Hg | 114 (18) | 129 (19) | 115 (17) | 107 (16) | 104 (17) | < .001 |
| Diastolic blood pressure, mean (SD), mm Hg | 70 (11) | 79 (12) | 71 (11) | 67 (10) | 64 (9) | < .001 |
| Laboratory tests | ||||||
| Total cholesterol, mean (SD), mg/dL | 168 (45) | 185 (42) | 172 (44) | 155 (43) | 135 (43) | < .001 |
| Hemoglobin, mean (SD), g/dL | 13 (1.6) | 14 (1.3) | 14 (1.5) | 13 (1.7) | 12 (1.7) | < .001 |
| Medications | ||||||
| β-Blocker, No. (%) | 2163 (95) | 264 (100) | 1269 (97) | 550 (89) | 80 (89) | < .001 |
| Loop diuretic, No. (%) | 1776 (78) | 160 (61) | 971 (74) | 560 (90) | 85 (94) | < .001 |
| Diuretic dose, mean (SD), mg/kg/d | 0.7 (0.7) | 0.3 (0.4) | 0.5 (0.5) | 1.0 (0.8) | 1.8 (1.6) | < .001 |
| ACE inhibitor, No. (%) | 1701 (74) | 222 (84) | 996 (76) | 438 (71) | 45 (50) | < .001 |
| Lipid-lowering agent, No. (%) | 1074 (47) | 165 (63) | 664 (51) | 213 (34) | 32 (36) | < .001 |
| Spironolactone/eplerenone, No. (%) | 1030 (45) | 167 (63) | 594 (45) | 238 (38) | 31 (34) | < .001 |
| Angiotensin receptor blockers, No. (%) | 537 (24) | 61 (23) | 308 (24) | 144 (23) | 24 (27) | .77 |
| Implantable cardioverter-defibrillator | 917 (40) | 129 (49) | 499 (38) | 251 (40) | 38 (42) | .15 |
| Medical history, No. (%) | ||||||
| Hypertension | 1362 (60) | 172 (65) | 770 (59) | 358 (58) | 62 (69) | .42 |
| Myocardial infarction | 962 (42) | 57 (22) | 519 (40) | 334 (54) | 52 (58) | < .001 |
| Diabetes mellitus | 731 (32) | 65 (25) | 403 (31) | 221 (36) | 42 (47) | < .001 |
| Angina | 577 (25) | 40 (15) | 338 (26) | 170 (27) | 29 (32) | < .001 |
| Chronic obstructive pulmonary disease | 243 (11) | 11 (4.2) | 127 (9.7) | 89 (15) | 16 (18) | < .001 |
| Stroke | 235 (10) | 28 (11) | 114 (8.7) | 77 (12) | 16 (18) | .001 |
| Peripheral vascular disease | 151 (6.6) | 9 (3.4) | 77 (5.9) | 55 (8.9) | 10 (11) | < .001 |
P value from analysis of variance for categorical variables and Pearson correlation coefficient for continuous variables.
Patients with higher SHFM scores were more likely to be men (P < .001) and white (P = .02). Ninety-one percent of patients with a rounded SHFM score of 2 were men, compared with about 67% of patients with a score of –1 (P < .001). Higher SHFM scores were significantly associated with older age, worse New York Heart Association (NYHA) functional classification, lower ejection fraction, lower systolic and diastolic blood pressure, and lower hemoglobin. Patients with higher SHFM scores were significantly more likely to be hospitalized (P < .001) and to have had an emergent or urgent outpatient visit (P < .001) (Table 2). The mean number of hospitalizations during the first year after randomization ranged from 0.4 among patients with predicted 1-year mortality of 1.5% (rounded SHFM score, –1) to 1.5 among patients with predicted 1-year mortality of 26% (rounded SHFM score, 2). The mean number of inpatient days ranged from 2.1 to 11.4 days, respectively. Higher SHFM scores were significantly associated with more hospital admissions and more inpatient days; a 1-unit increase in SHFM score was associated with an 86% increase in hospital admissions and a 130% increase in inpatient days (P < .001 for both comparisons). Similar relationships existed for emergent and urgent outpatient visits and nonurgent outpatient visits. The mean number of emergent and urgent visits approximately doubled (1.5) for patients with a rounded SHFM score of 2, compared with patients with a rounded SHFM score of –1 (0.7; P < .001).
Table 2.
Medical Resource Use and Costs at 1 Year by Seattle Heart Failure Model Score
| Variable | All Patients (N = 2288) | Rounded Seattle Heart Failure Model Score | P Value | |||
|---|---|---|---|---|---|---|
| –1 (n = 264) | 0 (n = 1313) | 1 (n = 621) | 2 (n = 90) | |||
| Medical resources | ||||||
| Proportion of patients with ED or urgent care visits, No. (%) | 960 (42) | 96 (36) | 530 (40) | 280 (45) | 54 (60) | < .001 |
| Proportion of patients hospitalized, No. (%) | 911 (40) | 66 (25) | 463 (35) | 329 (53) | 53 (59) | < .001 |
| Non-urgent outpatient visits | < .001 | |||||
| Mean (SD) | 12.3 (13.2) | 9.6 (12.2) | 11.8 (13.2) | 14.2 (13.0) | 15.4 (14.3) | |
| Median (interquartile range) | 9.0 (4.0–16.0) | 6.0 (3.0–11.0) | 8.0 (4.0–15.0) | 11.0 (6.0–20.0) | 12.0 (6.0–18.0) | |
| ED or urgent outpatient visits | < .001 | |||||
| Mean (SD) | 0.8 (1.6) | 0.7 (1.4) | 0.8 (1.4) | 1.0 (2.0) | 1.5 (2.4) | |
| Median (interquartile range) | 0.0 (0.0–1.0) | 0.0 (0.0–1.0) | 0.0 (0.0–1.0) | 0.0 (0.0–1.0) | 1.0 (0.0–2.0) | |
| Hospital admission | < .001 | |||||
| Mean (SD) | 0.8 (1.3) | 0.4 (1.0) | 0.6 (1.1) | 1.1 (1.5) | 1.5 (1.8) | |
| Median (interquartile range) | 0.0 (0.0–1.0) | 0.0 (0.0–0.5) | 0.0 (0.0–1.0) | 1.0 (0.0–2.0) | 1.0 (0.0–3.0) | |
| Inpatient days | < .001 | |||||
| Mean (SD) | 4.8 (12.2) | 2.1 (6.0) | 3.6 (11.2) | 7.4 (14.7) | 11.4 (15.8) | |
| Median (interquartile range) | 0.0 (0.0–5.0) | 0.0 (0.0–1.0) | 0.0 (0.0–3.0) | 2.0 (0.0–8.0) | 3.0 (0.0–18.0) | |
| Medical costs | ||||||
| Inpatient care costs, $ | < .001 | |||||
| Mean (SD) | 12,808 (35,788) | 5694 (16,573) | 9912 (30,493) | 20,256 (47,532) | 24,535 (43,068) | |
| Median (interquartile range) | 0 (0–10,356) | 0 (0–622) | 0 (0–6527) | 3330 (0–20,732) | 6341 (0–34,667) | |
| Outpatient care costs, $ | < .001 | |||||
| Mean (SD) | 2522 (4267) | 1876 (2995) | 2314 (3855) | 3023 (4801) | 3999 (7473) | |
| Median (interquartile range) | 1361 (575–2854) | 1007 (415–2277) | 1267 (525–2739) | 1772 (754–3148) | 1891 (1013–3564) | |
| Total medical costs, $ | < .001 | |||||
| Mean (SD) | 15,330 (36,964) | 7570 (17,663) | 12,226 (31,356) | 23,278 (48,923) | 28,535 (45,638) | |
| Median (interquartile range) | 2552 (718–15,055) | 1220 (447–5272) | 2088 (646–9882) | 6617 (1197–24,473) | 10,609 (2055–41,524) | |
Abbreviation: ED, emergency department.
Cumulative medical costs at 1 year averaged $15,330 (SD, $36,964; median, $2552; interquartile range, $718-$15,055). Higher SHFM scores were significantly associated with higher total costs (Figure 1). On average, inpatient care costs accounted for 84% of total costs, estimated at $12,808 (SD, $35,788). Mean inpatient costs ranged from $5694 among those with an SHFM score of –1 to $24,535 among those with a score of 2. Mean outpatient care costs, including emergent, urgent, and nonurgent care visits and outpatient procedures, were $1876 for patients with an SHFM score of –1 and $3999 for patients with a score of 2. As a linear continuous covariate, SHFM score was a significant predictor of inpatient costs (rate ratio per 1-unit increase, 2.28; P < .001), outpatient costs (1.54; P < .001), and total costs (2.13; P < .001). Predicted 1-year total medical costs generated from the model for SHFM scores of −1, 0, 1 and 2 were $6,060, $12,896, $27,442, and $58,398, respectively. For baseline SHFM scores of 1 and 2, predicted 1-year mortality was 10.5% and 26.0%. Thus, with an increase in SHFM scores from 1 to 2, 1-year costs were expected to increase by $1997 for each percentage-point increase in predicted 1-year mortality.
Figure.
Rounded Seattle Heart Failure Model Score and Mean Costs
In the sensitivity analysis that excluded patients who died during the first year, 2180 patients (95%) remained. The proportions of patients who had at least 1 hospitalization (38%), at least 1 emergency visit (41%), mean inpatient costs ($11, 970) and total medical costs ($14,491) over 1 year were consistent with the main analysis (Table 3). Comparisons across the 4 SHFM score groups of counts of medical resource use and medical costs over 1 year after randomization were also consistent. Higher SHFM scores were associated with more hospitalizations, more emergency visits, more outpatient nonurgent visits, and higher inpatient and total costs. Total 1-year medical costs ranged from $7388 for patients with an SHFM score of –1 to $28,524 for patients with an SHFM score of 2 (P < .001).
Table 3.
Medical Resource Use and Costs at 1 Year by Seattle Heart Failure Model Score After Exclusion of Patients Who Died Within 1 Year
| Variable | All Patients (N = 2180) | Rounded Seattle Heart Failure Model Score | P Value | |||
|---|---|---|---|---|---|---|
| –1 (n = 261) | (n = 1277) | 1 (n = 567) | 2 (n = 75) | |||
| Medical resources | ||||||
| Proportion of patients with ED or urgent care visits, No. (%) | 889 (41) | 94 (36) | 505 (40) | 245 (43) | 45 (60) | < .001 |
| Proportion of patients hospitalized, No. (%) | 832 (38) | 64 (25) | 439 (34) | 288 (51) | 41 (59) | < .001 |
| Non-urgent outpatient visits | < .001 | |||||
| Mean (SD) | 12.4 (13.2) | 9.5 (12.2) | 11.8 (13.2) | 14.6 (13.2) | 15.7 (13.6) | |
| Median (interquartile range) | 9.0 (5.0–16.0) | 6.0 (3.0–11.0) | 8.0 (4.0–15.0) | 11.0 (6.0–20.0) | 13.0 (7.0–18.0) | |
| ED or urgent outpatient visits | < .001 | |||||
| Mean (SD) | 0.8 (1.6) | 0.7 (1.4) | 0.8 (1.4) | 1.0 (2.0) | 1.4 (2.2) | |
| Median (interquartile range) | 0.0 (0.0–1.0) | 0.0 (0.0–1.0) | 0.0 (0.0–1.0) | 0.0 (0.0–1.0) | 1.0 (0.0–2.0) | |
| Hospital admission | < .001 | |||||
| Mean (SD) | 0.7 (1.2) | 0.4 (1.0) | 0.6 (1.1) | 1.0 (1.5) | 1.5 (1.9) | |
| Median (interquartile range) | 0.0 (0.0–1.0) | 0.0 (0.0–0.0) | 0.0 (0.0–1.0) | 1.0 (0.0–1.0) | 1.0 (0.0–3.0) | |
| Inpatient days | < .001 | |||||
| Mean (SD) | 4.3 (11.6) | 2.0 (5.9) | 3.5 (11.2) | 6.4 (13.1) | 10.9 (16.1) | |
| Median (interquartile range) | 0.0 (0.0–4.0) | 0.0 (0.0–0.0) | 0.0 (0.0–3.0) | 2.0 (0.0–7.0) | 3.0 (0.0–16.0) | |
| Medical costs | ||||||
| Inpatient care costs, $ | < .001 | |||||
| Mean (SD) | 11,970 (34,855) | 5516 (16,480) | 9634 (30,035) | 18,560 (46,384) | 24,371 (45,586) | |
| Median (interquartile range) | 0 (0–8381) | 0 (0–0) | 0 (0–5635) | 2442 (0–18,138) | 3965 (0–34,667) | |
| Outpatient care costs, $ | < .001 | |||||
| Mean (SD) | 2522 (4311) | 1873 (3,011) | 2304 (3860) | 3095 (4,961) | 4153 (7917) | |
| Median (interquartile range) | 1347 (573–2850) | 993 (415–2266) | 1257 (525–2706) | 1788 (748–3155) | 1938 (1013–3726) | |
| Total medical costs, $ | < .001 | |||||
| Mean (SD) | 14,491 (36,052) | 7388 (17,589) | 11,938 (30,844) | 21,655 (47,920) | 28,524 (48,310) | |
| Median (interquartile range) | 2414 (689–13,045) | 1167 (445–4886) | 1966 (634–9326) | 5494 (1103–22,084) | 9196 (1817–41,524) | |
Abbreviation: ED, emergency department.
Discussion
Seattle Heart Failure Model scores have been widely used in clinical practice to predict survival and to support discussions with patients about the impact of therapeutic interventions and to guide cardiac transplant and ventricular assist device (VAD) placement.10,11 We found that SHFM scores are consistently associated with higher rates of inpatient and outpatient medical resource use and associated costs. Patients with predicted 1-year mortality of 26% incurred costs that were almost 4 times higher than costs incurred by patients with predicted 1-year mortality of 1.5%.
A limited number of studies have reported associations between SHFM scores and hospital days. Ketchum et al10 reported that patients with higher SHFM scores (ie, patients at higher risk) were more likely to have longer hospital stays after VAD placement. Two abstracts reported associations between higher SHFM scores and greater resource use. Bahrainy et al12 found that higher SHFM scores were associated with longer hospital stays in patients with acute decompensated heart failure. Moorman et al13 analyzed data from the Advanced Chronic Heart Failure Clinical Assessment of Immune Modulation Therapy and found linear relationships between SHFM-predicted mortality and annual rates of all-cause and heart failure hospitalizations.
Compared with previous studies that examined relationships between SHFM scores and medical resources, our study has several strengths. The study included a larger sample of patients and examined a wider range of medical resources, including inpatient days, emergency and urgent care visits, and nonurgent outpatient visits. Two of the previous studies had small samples of patients treated at single institutions,10,12 in patients who had undergone VAD placement,10 and both were limited to length of stay associated with an index admission.10,12 The only previous study that included patients from multiple sites and examined associations between SHFM scores and longitudinal measures of resource use was limited to abstract form. Thus, methodological detail is sparse, and the study did not evaluate costs. To our knowledge, ours is the first study to report on relationships between SHFM scores and medical costs.
Our study has limitations. First, patients with more severe heart failure were underrepresented. Less than 1% of patients had signs and symptoms consistent with NYHA class IV, but approximately 1 out of 3 (35%) had NYHA class III signs and symptoms at baseline. Future studies with larger samples of patients with NYHA class IV heart failure would be useful in characterizing relationships between SHFM scores and resource use in these patients. Second, we only examined 1-year medical resources and costs after randomization. However, we chose this time horizon to maintain a closer temporal relationship between SHFM scores measured at baseline and subsequent resource use. We recognized that, as some patients experienced disease progression, their SHFM scores would increase, thereby increasing the variance associated with relationships between SHFM scores and resource use or costs. Third, although HF-ACTION collected data on 17 of the 20 variables needed to compute SHFM scores, the trial did not collect patient-level information on allopurinol use, lymphocytes, and serum uric acid. We imputed values for percent lymphocytes and uric acid levels using patient-level characteristics based on externally generated prediction models developed using the original SHFM cohort and we do not expect that this imputation strategy imparted bias, but the assumption of no allopurinol use for all patients would slightly suppress SHFM scores for a small fraction of individuals, expected to range from 4% to 18% as reported in the external data sets used to validate the SHFM.4 Fourth, generalized linear models do not provide absolute measures of model fit, like ROC curves for logistic models or R2 for OLS models. Thus, a convenient means to demonstrate how accurately the models predict resource use and costs is not available.
Given the rising costs of care associated with heart failure, heart failure has become a primary target of disease management programs.14–16 Earlier studies of disease management programs for heart failure reported short-term cost-savings.17–21 However, as the literature has grown, so has the variation in economic results, with more studies failing to demonstrate cost savings or reductions in hospital admissions.22–25 For programs that may increase costs in the short term, it is useful to evaluate whether the benefits of these programs are commensurate with their costs. However, few studies have examined the long-term costs and benefits of disease management programs.26–28 Also, the variable approaches employed for extrapolation make it difficult to disentangle true differences in the cost-effectiveness between programs or differences in methodological strategies or assumptions. To address challenges of extrapolating short-term findings in studies of disease management programs, we are in the final phase of developing a Web-based cost-effectiveness analysis model that will incorporate the relationships between SHFM scores and resource use reported in this paper. This freely available model will provide a common approach to evaluating long-term costs and health outcomes associated with heart failure disease management programs, thereby facilitating direct comparisons between programs. In addition, others could build SHFM scores into forecasting models to predict medical resource use and costs for individuals or groups of patients with heart failure. Such information will be increasingly valuable as providers take on more financial risk with capitated reimbursement and bundled payment schemes.
Although the SHFM was developed to predict survival for heart failure patients, it also predicts medical resource use and costs. These relationships between SHFM scores and medical resource use and costs will be integral to examining the cost-effectiveness of disease management programs or other interventions designed to improve outcomes of patients with heart failure.
Acknowledgments
Funding/Support: This study was supported by grant 5R01NR011873-02 from the National Institute of Nursing Research. HF-ACTION was funded by grants 5U01HL063747, 5U01HL066461, 5U01HL068973, 5U01HL066501, 5U01HL066482, 5U01HL064250, 5U01HL066494, 5U01HL064257, 5U01HL066497, 5U01HL068980, 5U01HL064265, 5U01HL066491, and 5U01HL064264 from the National Heart, Lung, and Blood Institute; and grants R37AG018915 and P60AG010484 from the National Institute on Aging. The content of this manuscript is solely the responsibility of the authors and does not necessarily represent the official views of the National Institute of Nursing Research, the National Heart, Lung, and Blood Institute, or the National Institutes of Health.
We thank Linda Davidson-Ray, Betsy O’Neal, and Ann Burnette, Duke University, for acquisition of hospital billing data, and Damon M. Seils, MA, Duke University, for assistance with manuscript preparation. They did not receive compensation for their assistance apart from their employment at the institution where the study was conducted.
Footnotes
Disclosures: Dr Levy reported receiving research support from Thoratec, HeartWare, GE Healthcare, Medtronic, and NIHLBI; licensing from Epocrates; and serving as a consultant to GE Healthcare, Novartis, and Amgen. No other disclosures were reported.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
References
- 1.Lloyd-Jones D, Adams RJ, Brown TM, Carnethon M, Dai S, De Simone G, et al. Heart disease and stroke statistics--2010 update: a report from the American Heart Association. Circulation. 2010;121:e46–215. doi: 10.1161/CIRCULATIONAHA.109.192667. [DOI] [PubMed] [Google Scholar]
- 2.Jencks SF, Williams MV, Coleman EA. Rehospitalizations among patients in the Medicare fee-for-service program. N Engl J Med. 2009;360:1418–28. doi: 10.1056/NEJMsa0803563. [DOI] [PubMed] [Google Scholar]
- 3.US News and World Report. [Accessed January 03, 2013];Heart failure accounts for 37% of Medicare spending. Available at: http://health.usnews.com/health-news/family-health/heart/articles/2008/11/12/heart-failure-accounts-for-37-of-medicare-spending.
- 4.Levy WC, Mozaffarian D, Linker DT, Sutradhar SC, Anker SD, Cropp AB, et al. The Seattle Heart Failure Model: prediction of survival in heart failure. Circulation. 2006;113:1424–33. doi: 10.1161/CIRCULATIONAHA.105.584102. [DOI] [PubMed] [Google Scholar]
- 5.Clemens M, Szegedi Z, Kardos L, Nagy-Baló E, Sándorfi G, Edes I, et al. The Seattle Heart Failure Model predicts survival in patients with cardiac resynchronization therapy: a validation study. J Card Fail. 2012;18:682–7. doi: 10.1016/j.cardfail.2012.07.001. [DOI] [PubMed] [Google Scholar]
- 6.May HT, Horne BD, Levy WC, Kfoury AG, Rasmusson KD, Linker DT, et al. Validation of the Seattle Heart Failure Model in a community-based heart failure population and enhancement by adding B-type natriuretic peptide. Am J Cardio. 2007;100:697–700. doi: 10.1016/j.amjcard.2007.03.083. [DOI] [PubMed] [Google Scholar]
- 7.Whellan DJ, O’Connor CM, Lee KL, Keteyian SJ, Cooper LS, Ellis SJ, et al. Heart failure and a controlled trial investigating outcomes of exercise training (HF-ACTION): design and rationale. Am Heart J. 2007;153:201–11. doi: 10.1016/j.ahj.2006.11.007. [DOI] [PubMed] [Google Scholar]
- 8.O’Connor CM, Whellan DJ, Lee KL, Keteyian SJ, Cooper LS, Ellis SJ, et al. Efficacy and safety of exercise training in patients with chronic heart failure: HF-ACTION randomized controlled trial. JAMA. 2009;301:1439–50. doi: 10.1001/jama.2009.454. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Reed SD, Whellan DJ, Li Y, Friedman JY, Ellis SJ, Piña IL, et al. Economic evaluation of the HF-ACTION (Heart Failure: A Controlled Trial Investigating Outcomes of Exercise Training) randomized controlled trial: an exercise training study of patients with chronic heart failure. Circ Cardiovasc Qual Outcomes. 2010;3:374–81. doi: 10.1161/CIRCOUTCOMES.109.907287. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Ketchum ES, Moorman AJ, Fishbein DP, Mokadam NA, Verrier ED, Aldea GS, et al. Predictive value of the Seattle Heart Failure Model in patients undergoing left ventricular assist device placement. J Heart Lung Transplant. 2010;29:1021–5. doi: 10.1016/j.healun.2010.05.002. [DOI] [PubMed] [Google Scholar]
- 11.Levy WC, Mozaffarian D, Linker DT, Farrar DJ, Miller LW REMATCH Investigators. Can the Seattle heart failure model be used to risk-stratify heart failure patients for potential left ventricular assist device therapy? J Heart Lung Transplant. 2009;28:231–6. doi: 10.1016/j.healun.2008.12.015. [DOI] [PubMed] [Google Scholar]
- 12.Cardiovascular Business. [Accessed February, 21, 2013];ACC: Heart failure model could predict outcomes for patients, shorten inpatient stays. Available at: http://www.cardiovascularbusiness.com/index.php?option=com_articles&view=conference&id=publication:41:article:21248.
- 13.Moorman AJ, Ponikowski P, Anker SD, Torre-Amione G, Young JB, Levy WC. Linear relationship between Seattle Heart Failure Model predicted mortality and hospitalizations in patients with chronic heart failure. Circulation. 2011;124:A17913. [Google Scholar]
- 14.Phillips CO, Wright SM, Kern DE, Singa RM, Shepperd S, Rubin HR. Comprehensive discharge planning with postdischarge support for older patients with congestive heart failure: a meta-analysis. JAMA. 2004;291:1358–67. doi: 10.1001/jama.291.11.1358. [DOI] [PubMed] [Google Scholar]
- 15.Göhler A, Januzzi JL, Worrell SS, Osterziel KJ, Gazelle GS, Dietz R, et al. A systematic meta-analysis of the efficacy and heterogeneity of disease management programs in congestive heart failure. J Card Fail. 2006;12:554–67. doi: 10.1016/j.cardfail.2006.03.003. [DOI] [PubMed] [Google Scholar]
- 16.Gwadry-Sridhar FH, Flintoft V, Lee DS, Lee H, Guyatt GH. A systematic review and meta-analysis of studies comparing readmission rates and mortality rates in patients with heart failure. Arch Intern Med. 2004;164:2315–20. doi: 10.1001/archinte.164.21.2315. [DOI] [PubMed] [Google Scholar]
- 17.Rich MW, Beckham V, Wittenberg C, Leven CL, Freedland KE, Carney RM. A multidisciplinary intervention to prevent the readmission of elderly patients with congestive heart failure. N Engl J Med. 1995;333:1190–5. doi: 10.1056/NEJM199511023331806. [DOI] [PubMed] [Google Scholar]
- 18.Stewart S, Pearson S, Horowitz JD. Effects of a home-based intervention among patients with congestive heart failure discharged from acute hospital care. Arch Intern Med. 1998;158:1067–72. doi: 10.1001/archinte.158.10.1067. [DOI] [PubMed] [Google Scholar]
- 19.Krumholz HM, Amatuda J, Smith GL, Mattera JA, Roumanis SA, Radford MJ, et al. Randomized trial of an education and support intervention to prevent readmission of patients with heart failure. J Am Coll Cardiol. 2002;39:83–9. doi: 10.1016/s0735-1097(01)01699-0. [DOI] [PubMed] [Google Scholar]
- 20.Murry MD, Young J, Hoke S, Tu W, Weiner M, Morrow D, et al. Pharmacist intervention to improve medication adherence in heart failure: a randomized trial. Ann Intern Med. 2007;146:714–25. doi: 10.7326/0003-4819-146-10-200705150-00005. [DOI] [PubMed] [Google Scholar]
- 21.Inglis SC, Clark RA, McAlister FA, Ball J, Lewinter C, Cullington D, et al. Structured telephone support or telemonitoring programmes for patients with chronic heart failure. Cochrane Database Syst Rev. 2010 Aug 4;(8):CD007228. doi: 10.1002/14651858.CD007228.pub2. [DOI] [PubMed] [Google Scholar]
- 22.McCall N, Cromwell J. Results of the Medicare Health Support disease-management pilot program. N Engl J Med. 2011;365:1704–12. doi: 10.1056/NEJMsa1011785. [DOI] [PubMed] [Google Scholar]
- 23.Herbet PL, Sisk JE, Wang JJ, Tuzzio L Casabianca JM, Chassin MR, et al. Cost-effectiveness of nurse-led disease management for heart failure in an ethnically diverse urban community. Ann Intern Med. 2008;149:540–8. doi: 10.7326/0003-4819-149-8-200810210-00006. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Smith B, Hughes-Cromwick PF, Forkner E, Galbreath AD. Cost-effectiveness of telephonic disease management in heart failure. Am J Manag Care. 2008;14:106–15. [PubMed] [Google Scholar]
- 25.Chaudhry SI, Mattera JA, Curtis JP, Spertus JA, Herrin J, Lin Z, et al. Telemonitoring in patients with heart failure. N Engl J Med. 2010;363:2301–9. doi: 10.1056/NEJMoa1010029. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Miller G, Randolph S, Forkner E, Smith B, Galbreath AD. Long-term cost-effectiveness of disease management in systolic heart failure. Med Decision Making. 2009;29:325–33. doi: 10.1177/0272989X08327494. [DOI] [PubMed] [Google Scholar]
- 27.Moertl D, Steiner S, Coyle D, Berger R. Cost-utility analysis of nt-probnp-guided multidisciplinary care in chronic heart failure. Int J Technol Assess Health Care. 2013;29:3–11. doi: 10.1017/S0266462312000712. [DOI] [PubMed] [Google Scholar]
- 28.Wijeysundera HC, Machado M, Wang X, Van Der Velde G, Sikich N, Witteman W, et al. Cost-effectiveness of specialized multidisciplinary heart failure clinics in Ontario, Canada. Value Health. 2010;13:915–21. doi: 10.1111/j.1524-4733.2010.00797.x. [DOI] [PubMed] [Google Scholar]