Clinical Investigations

Congestive Heart Failure

Repeated hospitalizations predict mortality in the community population with heart failure
Soko Setoguchi, MD, DrPH,a Lynne Warner Stevenson, MD,b and Sebastian Schneeweiss, MD, ScDa Boston, MA

Background Identification of patients at high risk of death is critical for appropriate management of patients and health care resources. The impact of repeated heart failure (HF) hospitalization on mortality has not been studied for a large community population with HF. We aimed to characterize survival of patients in relation to the number of HF hospitalizations. Method Using the health care utilization databases, we identified a cohort of patients with a first hospitalization for HF among all residents of British Columbia between 2000 and 2004. Survival time was measured after patients first and each subsequent HF hospitalization. Kaplan-Meier cumulative mortality curves were constructed after each subsequent HF hospitalization. Hazard ratios for the number of HF hospitalizations were estimated using a multivariate Cox regression adjusting for major comorbidities. Results Of 14 374 patients hospitalized for HF, 7401 died during the 24 766 person-years of follow-up. Mortality significantly increased after each HF hospitalization. After adjusting for age, sex, and major comorbidities, the number of HF hospitalizations was a strong predictor of all-cause death. Median survival after the first, second, third, and fourth hospitalization was 2.4, 1.4, 1.0, and 0.6 years. Advanced age, renal disease, and history of cardiac arrest attenuated the impact of the number of HF hospitalizations. Conclusions The number of HF hospitalizations is a strong predictor of mortality in community HF patients. This simple predictor of mortality in HF patients should help triage management and resources for HF and trigger patient planning for prognosis. (Am Heart J 2007;154:26026.)
The prevalence and economic burden of heart failure (HF) have been increasing during the past several decades,1,2 with an estimated 5 million people currently diagnosed.3 The number of hospitalizations with HF as the first listed diagnosis has increased steadily over the last 3 decades, with 1 093 000 hospitalizations for HF in 2003.4 Heart failure is also the leading cause of hospital readmission within 60 days of discharge.5 Heart failure is the end stage of cardiac disease, with 1-year mortality rate of 30% to 50% described previously after the first hospitalization has occurred.6,7 Although new management options, including costly medications, implantable cardioverter-defibrillators, left ventricular assist devices, and transplantation, have improved survival8 and quality of life for selected patients with HF, health care systems face difficult decisions regarding the allocation of these and other limited resources within the broader community. Health care staffs need guidance to time discussions with individual patients and families about prognosis and plans for the end of life. Most studies examining predictors of survival in HF describe single-center experiences or trial populations with many exclusions that limit enrollment of older patients with HF.7,9-11 A recent study suggested that one previous HF hospitalization is a predictor of death.9 However, in the community population, the first HF hospitalization is often the time of first HF diagnosis, after which evaluation and treatment are initiated. The mortality of patients with HF in relation to the number of additional HF hospitalizations has not been studied in a community population. The aim of the current study is to characterize the mortality of HF patients with repeated hospitalizations.

From the aDivision of Pharmacoepidemiology and Pharmacoeconomics, Department of Medicine, Brigham and Womens Hospital and Harvard Medical School, Boston, MA, and bAdvanced Heart Disease Section, Division of Cardiovascular Medicine, Department of Medicine, Brigham and Womens Hospital and Harvard Medical School, Boston, MA. This project was supported by grants from the National Institute on Aging (R01AG021950) and from the Agency for Healthcare Research and Quality (2-RO1HS10881), Department of Health and Human Services, Rockville, MD. Submitted October 9, 2006; accepted January 16, 2007. Reprint requests: Soko Setoguchi, MD, DrPH, Division of Pharmacoepidemiology, 1620 Tremont St, Suite 3030, Boston, MA 02130. E-mail: soko@post.harvard.edu 0002-8703/$ - see front matter D 2007, Mosby, Inc. All rights reserved. doi:10.1016/j.ahj.2007.01.041

Methods
Data sources
We used health care utilization databases that contain information on discharge abstract, outpatient diagnoses, and

American Heart Journal Volume 154, Number 2

Setoguchi et al 261

procedure codes for all residents of British Columbia, Canada, from 1996 to 2004, where the Canadian national insurance system provides comprehensive coverage for health care for all including the elderly and disabled. These data sources provide basic demographic information, as well as coded diagnostic, procedural, and pharmacy dispensing information with high accuracy.12 The discharge abstract provided up to 25 fields for diagnoses and 10 fields for procedures that were relevant during the hospitalization. The institutional review boards of the Brigham and Womens Hospital and University of Victoria approved this study, and data use agreements were established. All potentially traceable personal identifiers were removed from the data before analyses to protect patients privacy.

Table I. Characteristics of 14 374 study patients at their first HF hospitalization

Study population (N = 14,374) No. of patients Follow-up, y Age, y Age category, y b65 65-74 75-84 85+ Male Prior hospitalization for any reason (non-HF) Prior hospitalization for cardiovascular (non-HF) reasons MI Cerebrovascular disease Atrial fibrillation or flutter VT/VF/cardiac arrest Ischemic heart disease Hypertension Diabetes Cancer CKD (no dialysis) Renal disease on dialysis Chronic pulmonary disease Rheumatoid arthritis Dementia ICD implantation Pacemaker implantation 14,374 1.7 (1.4) 77 (12) 13 20 38 29 54 80 50 18 11 36 2 47 71 36 9 21 2 24 2 6 0.5 7

Study cohort
We assembled a cohort of patients who were admitted to a hospital for the first time for HF between January 1, 2000, and December 31, 2004. The hospitalization due to HF was defined as an admission with HF as the primary diagnosis in the discharge abstract (International Classification of Diseases, Ninth Revision [ICD-9], code of 428.XX), which was previously shown to have a positive predictive value (PPV) of 94% for HF by Framingham criteria.13 Patients with a previous hospitalization for HF within the past 4 years were excluded.

Covariates
Patients characteristics and other comorbid conditions were assessed during the 4 years before the HF hospitalization. The comorbid conditions included history of myocardial infarction (MI), other ischemic heart diseases, cerebrovascular attacks, atrial fibrillation, cardiac arrest/ventricular arrhythmia, hypertension, diabetes, chronic kidney diseases (CKDs), chronic pulmonary diseases, cancer, rheumatoid arthritis, and dementia. These conditions were defined by the presence of corresponding ICD-9 diagnosis codes in hospital discharge diagnosis and/or outpatient files. Many of the definitions used were previously validated and had good to excellent PPVs.10,14-17 For example, the definition of MI was previously validated by medical chart review to have a PPV of 94%.15 We used a definition of CKD previously shown to have a PPV of 94% compared with estimated glomerular filtration rate of b60 mL/min/1.73 m2.16 Cardiac arrest/Ventricular arrhythmia was defined by ICD-9 codes in hospital discharge summary, previously shown to have a PPV of 82%.17

Values represent percentage for categorical variables and mean (SD) for continuous variables. Covariates were assessed during the 4-year period before the first HF hospitalization. VT , Ventricular tachycardia; VF , ventricular fibrillation.

cardiovascular diagnosis or for any cause, in addition to the initial hospitalization for HF.

Multivariate regression analyses

Using the same cohort of patients, we estimated the effect of the number of HF hospitalizations after controlling for known comorbid conditions affecting survival in HF patients using multivariate Cox regression analyses. In contrast to the KaplanMeier curves, for these analyses, patients who had multiple HF hospitalizations contributed their person-time for each HF hospitalization until the time of the subsequent hospitalization. Patients characteristics and other comorbid conditions were assessed and updated at each HF hospitalization. Multivariate Cox proportional hazard models were constructed to assess the effect of the number of previous HF hospitalizations after controlling for age, sex, and the comorbid conditions.

Mortality after repeated HF hospitalization

All patients were followed until they died, until they lost eligibility for the health care system, or until the end of the study period. To characterize survival after each hospitalization for HF, we first estimated mortality after the first HF hospitalization, calculating patients survival time from the first hospitalization until they died or were censored. Subsequently, mortality after the second hospitalization was estimated by using survival time after the second hospitalization till the end of follow-up, and similarly for the third and fourth hospitalization. When patients died during HF hospitalization, such inpatient deaths were counted toward the current HF hospitalization. We constructed Kaplan-Meier cumulative mortality curves to compare mortality among the patients after the first, second, third, and fourth hospitalization. Kaplan-Meier curves were also constructed to determine whether there were the same or different effects for subsequent hospitalizations for any

Subgroup analyses
To assess whether the effect of the number of HF hospitalizations is enhanced or diminished by other risk factors, we constructed Kaplan-Meier curves stratified by characteristics such as age categories, sex, presence of CKD, previous MI, and history of cardiac arrest/ventricular arrhythmia. We also tested for effect modification (interaction) of these factors by the

262 Setoguchi et al

American Heart Journal August 2007

Figure 1

Kaplan-Meier cumulative mortality curve for all-cause mortality after each subsequent hospitalization for HF.

number of hospitalizations in multivariate analyses. All statistical analyses were performed using SAS (SAS Institute, Cary, NC).

Figure 2

Results
Study population and its characteristics We identified a cohort of 14,374 patients who were hospitalized for HF for the first time during January 1, 2000, to December 31, 2004. Characteristics of the study population are summarized in Table I. The mean age of the cohort was N75 years, reflecting the high incidence of HF in the elderly population. Patients had a high prevalence of comorbid conditions, including other cardiovascular diseases, diabetes, chronic pulmonary diseases, and CKD. Of 14,374 patients, 3,358 patients had second, 1,123 had third, and 417 had fourth hospitalizations for HF during the study period. The mean number of days between hospitalizations was 212 for the first and second, 158 for the second and third, and 117 for the third and fourth hospitalizations. Mortality of patients with HF hospitalizations The 30-day all-cause mortality after the first HF hospitalization was 12%, and the 1-year mortality was 34%. The median survival (50% mortality) was 2.4 years. We observed a significant increase in all-cause mortality with advanced age, as shown previously.1 The mortality significantly increased after each additional HF hospi-

Median survival (50% mortality) and 95% confidence limits in patients with HF after each HF hospitalization.

talization (Figure 1). Median survival times after the first, second, third, and fourth hospitalization were 2.4 (95% CI 2.3-2.5), 1.4 (95% CI 1.2-1.5), 1.0 (95% CI 0.91.1), and 0.6 (95% CI 0.5-0.9) years (Figure 2). Most patients were alive 2 years after the first HF hospitalization, but approximately half were dead by 1 year after 3 hospitalizations.

American Heart Journal Volume 154, Number 2

Setoguchi et al 263

Table II. Effects of number of HF hospitalizations on mortality

2 Hospitalizations4 Adjustment Unadjusted Age and sex adjusted Fully adjustedy HR 1.32 1.31 1.22 95% CI (1.25-1.41) (1.23-1.39) (1.14-1.30) 3 Hospitalizations4 HR 1.51 1.51 1.33 95% CI (1.37-1.67) (1.37-1.67) (1.20-1.47) 4 Hospitalizations4 HR 1.92 1.93 1.64 95% CI (1.65-2.24) (1.66-2.24) (1.40-1.91) > _ 5 Hospitalizations4 HR 2.25 2.25 1.84 95% CI (1.87-2.71) (1.87-2.71) (1.53-2.23)

HR , Hazard ratio. 4Hazard of deaths compared with the risk after 1 hospitalization. yModel included age; sex; and history of atrial fibrillation, ventricular arrhythmia/cardiac arrest, chronic kidney diseases, dialysis, diabetes, cancer, chronic pulmonary diseases, rheumatoid arthritis, MI, cerebrovascular attack, hypertension, ischemic heart diseases, and dementia.

Figure 3

Kaplan-Meier cumulative mortality curves for all-cause mortality after each subsequent hospitalization for HF stratified by age category.

Multivariate analyses for the effect of repeated hospitalization In the multivariate analyses, the number of previous hospitalizations for HF was a strong predictor of mortality after controlling for other known risk factors for HF survival (Table II). The adjusted hazard ratio for all-cause deaths was 1.2 (95% CI 1.1-1.3) for 2 HF hospitalizations, 1.3 (95% CI 1.2-1.5) for 3 hospitalizations, 1.6 (95% CI 1.4-1.9) for 4 hospitalizations, and

1.8 (95% CI 1.5-2.2) for 5 or more hospitalizations. The effect estimates of major comorbidities in the model were 1.6 (95% CI 1.5-1.6) for CKD without dialysis, 2.6 (95% CI 2.2-2.9) for CKD with dialysis, 1.5 (95% CI 1.41.6) for cancer, 1.8 (95% CI 1.6-2.0) for previous cardiac arrest, 1.2 (95% CI 1.1-1.3) for chronic pulmonary diseases, 1.2 (95% CI 1.1-1.3) for cerebrovascular diseases, 1.3 (95% CI 1.1-1.5) for rheumatoid arthritis, and 1.6 (95% CI 1.5-1.7) for dementia.

264 Setoguchi et al

American Heart Journal August 2007

Figure 4

Figure 5

Median survival and 95% CIs after HF hospitalizations by age category.

Median survival and 95% CIs after HF hospitalization with and without CKDs (the CKD definition has a PPV of 94% compared with estimated glomerular filtration rate b60 mL/[min 1.73 m2]).

Subgroups and modifying factors To estimate the effect of repeated hospitalization among age groups, we constructed Kaplan-Meier cumulative mortality curves after each hospitalization for each age category (b65, 65-74, and 85+ years) (Figure 3). The effects of repeated hospitalizations decreased progressively for advanced age. Because the mortality of older patients was already high after the first HF hospitalization, additional hospitalization added less additional prognostic information than that for younger patients (Figure 4). After the fourth hospitalization, patients in all age groups had very high mortality. The similar diminution of the effect of HF hospitalization on mortality was observed in the presence of CKD with or without dialysis (Figure 5) and in the patients who had a history of cardiac arrest (Figure 6). The tests for interaction terms with the number of hospitalizations were significant for age ( P b .0001), CKD ( P = .03), and cardiac arrest ( P = .0062). We did not observe significant modification of the effect of the number of HF hospitalization by previous MI. Sensitivity analyses with alternative definition of HF hospitalization We focused on the definition of HF hospitalization as an admission with HF being the primary diagnosis because this definition has been shown to have a PPV of 94% when the Framingham criteria are used.13 Nonetheless, we explored whether the results were sensitive to expanding the definition of HF hospitalization to HF as either primary or secondary diagnosis. We identified more patients by using the alternative definition of HF admissions (22,789 patients). However, the Kaplan-

Meier curves stratified by repeated HF hospitalizations were similar to the primary analyses. To test whether the increase in the mortality after subsequent hospitalization is specific to HF hospitalizations or observed for other non-HF hospitalizations, we further analyzed groups of HF patients who had subsequent hospitalizations for any reason or for non-HF cardiovascular diagnoses after the first HF hospitalization. The mortality curves of these groups of patients with non-HF hospitalization were constructed and compared with the mortality after the first and second HF hospitalizations. We found that the subsequent hospitalizations for any reason or any cardiovascular diseases other than HF did not significantly affect mortality. However, when we specifically assessed mortality after subsequent hospitalizations for MI, we observed that mortality after a subsequent MI was significantly increased; and the risk was especially high within the first few months after the MI admission. The 3-month mortality was 17% after the first HF hospitalization, 21% after the second HF hospitalization, and 32% after subsequent MI hospitalizations, whereas 1-year mortality was 34% after the first HF hospitalization, 44% after the second HF hospitalization, and 53% after subsequent MI hospitalizations.

Discussion
Among 14,374 patients with repeated HF hospitalizations, mortality increased sharply with recurrent HF hospitalizations. The number of repeated HF hospitalizations remained a strong independent predictor of

American Heart Journal Volume 154, Number 2

Setoguchi et al 265

Figure 6

Median survival after HF hospitalization with and without history of cardiac arrest or ventricular tachycardia/ventricular fibrillation (labeled as arrest).

mortality even after adjusting for other major risk factors for mortality among HF patients. However, the effects of the number of repeated HF hospitalizations were diminished with advanced age, presence of CKD, and history of cardiac arrest. History and timing of prior HF hospitalization have previously been examined as a covariate predicting mortality in patients with HF.9 To our knowledge, this is the first study that quantified the magnitude of risk associated with repeated hospitalization in patients with HF in a large community population outside a clinical trial. Mortality among the patients who were hospitalized for HF for the first time was 12.4% in 30 days and 33.7% in 5 years in our study, comparable with those of previous studies.7,10 The magnitudes of effects of other predictors of mortality in HF patients estimated in our multivariate model (eg, CKD, cerebrovascular disease, rheumatoid arthritis, dementia, cancer) were similar to those noted in previous studies.7,13 The relationship between hospitalization and mortality most likely represents the progressive symptomatic decompensation that occurs as HF approaches the end stages. In addition, hospitalization itself may not only be the consequence of worse prognosis but may contribute to worse outcome. Large doses of intravenous diuretics, and in particular inotropic therapy, even when used for short periods in hospital, have been associated with myocardial troponin leak, worsening renal function, and adverse outcomes 6 to 12 months later.18,19 Appreciation of the prognosis should influence decisions for patients who have been hospitalized for HF. For example, physicians may be more likely to consider expensive treatment options with delayed benefits, such

as placement of an implantable cardioverter-defibrillator, in a patient with good renal function and no comorbidities during the first admission with an HF diagnosis because his or her 50% survival is N2.5 years. On the other hand, a patient hospitalized for HF for the third time has b50% chance of surviving the next year. Those few patients who might be appropriate for cardiac transplantation or permanent left ventricular assist device placement should be referred to centers offering those procedures. For most of the patients without such definitive options, a defibrillator may not increase lifeyears; and risks of elective noncardiac surgery such as joint replacement may subtract from limited available time. After multiple HF hospitalizations, physicians should initiate frank discussion on prognosis and focus on management that improves the quality of life. Several limitations of the study should be noted. First, our data did not allow distinction between systolic dysfunction and HF with preserved ejection fraction. However, approximately one half of HF admissions are due to systolic dysfunction,20 for which prognosis is slightly worse than that for preserved ejection fraction HF. Thus, survival after repeated hospitalizations for HF due to systolic dysfunction would be overestimated by these analyses. Second, because no clinical information on measures of HF severity is available, we do not know whether the number of HF hospitalizations is independently predictive of mortality after adjusting for those clinical measures of HF severity such as functional status or quality of life. However, these measures could not be systematically obtained in the large number of community patients reflected in the study. Most HF hospitalizations result from symptoms occurring at rest or with minimal exertion, typical of at least transient class IV status. Our study suggests that the number of HF hospitalizations may serve as an integrated measure of severity over time. Throughout the progression of HF, many decisions need to be individualized in light of all available information. The number of hospitalizations does not replace consideration of multiple factors such as age, cardiac and renal function, functional status, and comorbidities. However, the current study suggests that the number of HF hospitalizations may be useful to triage patients for the therapies that benefit most at different stages of the disease. Perhaps most importantly, repeated rehospitalizations for HF should trigger individualized discussion with the patient and family about the goals of care for the limited time remaining.

References
1. McCullough PA, Philbin EF, Spertus JA, et al. Confirmation of a heart failure epidemic: findings from the Resource Utilization Among Congestive Heart Failure (REACH) study. J Am Coll Cardiol 2002;39:60 - 9.

266 Setoguchi et al

American Heart Journal August 2007

2. Levy D, Kenchaiah S, Larson MG, et al. Long-term trends in the incidence of and survival with heart failure. N Engl J Med 2002;347:1397 - 402. 3. Miller LW, Missov ED. Epidemiology of heart failure. Cardiol Clin 2001;19:547 - 55. 4. Hospital discharges for cardiovascular diseases. 2004 (Accessed January 19, 2006, at http://www.americanheart.org/ downloadable/heart/1136821961654HospDischarges06.pdf.). 5. Haldeman GA, Croft JB, Giles WH, et al. Hospitalization of patients with heart failure: National Hospital Discharge Survey, 1985 to 1995. Am Heart J 1999;137:352 - 60. 6. Thom T, Haase N, Rosamond W, et al. Heart disease and stroke statistics2006 update. A report from the American Heart Association Statistics Committee and Stroke Statistics Subcommittee. Circulation 2006. 7. Jong P, Vowinckel E, Liu PP, et al. Prognosis and determinants of survival in patients newly hospitalized for heart failure: a population-based study. Arch Intern Med 2002;162:1689 - 94. 8. Roger VL, Weston SA, Redfield MM, et al. Trends in heart failure incidence and survival in a community-based population. JAMA 2004;292:344 - 50. 9. Pocock SJ, Wang D, Pfeffer MA, et al. Predictors of mortality and morbidity in patients with chronic heart failure. Eur Heart J 2006;27:65 - 75. 10. Lee DS, Austin PC, Rouleau JL, et al. Predicting mortality among patients hospitalized for heart failure: derivation and validation of a clinical model. JAMA 2003;290:2581 - 7. 11. Ho KK, Moody GB, Peng CK, et al. Predicting survival in heart failure case and control subjects by use of fully automated methods for deriving nonlinear and conventional indices of heart rate dynamics. Circulation 1997;96:842 - 8.

12. Schneeweiss S, Avorn J. A review of uses of health care utilization databases for epidemiologic research on therapeutics. J Clin Epidemiol 2005;58:323 - 37. 13. Lee DS, Donovan L, Austin PC, et al. Comparison of coding of heart failure and comorbidities in administrative and clinical data for use in outcomes research. Med Care 2005;43:182 - 8. 14. Birman-Deych E, Waterman AD, Yan Y, et al. Accuracy of ICD-9CM codes for identifying cardiovascular and stroke risk factors. Med Care 2005;43:480 - 5. 15. Kiyota Y, Schneeweiss S, Glynn RJ, et al. Accuracy of Medicare claims-based diagnosis of acute myocardial infarction: estimating positive predictive value on the basis of review of hospital records. Am Heart J 2004;148:99 - 104. 16. Winkelmayer WC, Schneeweiss S, Mogun H, et al. Identification of individuals with CKD from Medicare claims data: a validation study. Am J Kidney Dis 2005;46:225 - 32. 17. De Bruin ML, van Hemel NM, Leufkens HG, et al. Hospital discharge diagnoses of ventricular arrhythmias and cardiac arrest were useful for epidemiologic research. J Clin Epidemiol 2005;58:1325 - 9. 18. Gheorghiade M, Zannad F, Sopko G, et al. Acute heart failure syndromes: current state and framework for future research. Circulation 2005;112:3958 - 68. 19. Silver MA, Horton DP, Ghali JK, et al. Effect of nesiritide versus dobutamine on short-term outcomes in the treatment of patients with acutely decompensated heart failure. J Am Coll Cardiol 2002;39:798 - 803. 20. Vasan RS, Larson MG, Benjamin EJ, et al. Congestive heart failure in subjects with normal versus reduced left ventricular ejection fraction: prevalence and mortality in a population-based cohort. J Am Coll Cardiol 1999;33:1948 - 55.

Access to American Heart Journal Online is reserved for print subscribers!

Full-text access to American Heart Journal Online is available for all print subscribers. To activate your individual online subscription, please visit American Heart Journal Online, point your browser to http://www.ahjonline.com, follow the prompts to activate your online access, and follow the instructions. To activate your account, you will need your subscribe account number, which you can find on your mailing label (note: the number of digits in your subscriber account number varies from 6 to 10). See the example below in which the subscriber account number has been circled:

Sample mailing label

This is your subscriber account number

**************************3-DIGIT 001 SJ P1 FEB 0 0 J004 C: 1 1234567-89 U 05/00 Q: 1 J. H. DOE, MD 531 MAIN ST CENTER CITY, NY 10001-001

Personal subscriptions to American Heart Journal Online are for individual use only and may not be transferred. Use of American Heart Journal Online is subject to agreement to the terms and conditions as indicated online.