Methods Study sample. Data sources

Transcription

1 Predictors of costs of caring for elderly patients discharged with heart failure Deborah J. Wexler, MD, MSc, a Jersey Chen, MD, MPH, a Grace L. Smith, MPH, b Martha J. Radford, MD, b,c,d Shlomit Yaari, BSc, e W. David Bradford, PhD, f and Harlan M. Krumholz, MD b,c,d,g New Haven and Middletown, Conn, Ramat-Gan, Israel, and Charleston, SC Background Investments in programs to improve outcomes and reduce readmissions for patients who survive hospitalization with heart failure will be economically most favorable for those who have the highest risk. Little information is available, however, to stratify the risk of these patients incurring costs after discharge. In this study, we sought to determine correlates of costs in a representative sample of patients with heart failure in the 6 months after discharge. Methods We reviewed medical records of 2181 patients aged 65 years who were discharged alive from 18 Connecticut hospitals in 1994 and 1995 with a principal discharge diagnosis of heart failure. Outcomes 6 months after discharge, including all-cause readmission and cost, heart failure related readmission and cost, and death, were obtained from the Medicare administrative database. A 2-stage sample selection model was used to identify the independent correlates of cost. Risk scores were calculated to identify subsets of patients at risk for generating high costs. Results On average, patients discharged with heart failure incurred costs of $2388 resulting from heart failure related admissions and $7101 resulting from admissions from any cause during the 6 months after discharge. An average admission for heart failure cost $7174, whereas an admission resulting from any cause cost $8589. The multivariate models explained 7% of the variation in cost, although clinical characteristics such as recent heart failure admissions, kidney failure, and hypertension were significant independent correlates of increased cost. Older age and a history of stroke were independently associated with decreased cost. Patients without any of the risk factors associated with increased costs still incurred $1500 to $5000, on average, in the 6 months after discharge. Conclusions Patients with heart failure generate substantial hospital costs in the 6 months after discharge. Given the emerging evidence for effective programs to reduce readmission, investments in interventions that produce even modest reductions in risk would be economically favorable. (Am Heart J 2001;142:350-7) From the a Department of Medicine, the c Section of Cardiovascular Medicine, and the g Section of Health Policy and Administration, Department of Epidemiology and Public Health, Yale University School of Medicine, b Yale New Haven Hospital Center for Outcomes Research and Evaluation, New Haven, d Qualidigm, Middletown, e Bar-Ilan University, Ramat-Gan, and the f Department of Health Administration and Policy, Medical University of South Carolina, Charleston. The analyses upon which this publication is based were performed under Contract Number P549, entitled Utilization and Quality Control Peer Review Organization for the State of Connecticut, sponsored by the Health Care Financing Administration, Department of Health and Human Services. The contents of this publication do not necessarily reflect the views or policies of the Department of Health and Human Services, nor does mention of trade names, commercial products, or organizations imply endorsement by the US Government. The authors assume full responsibility for the accuracy and completeness of the ideas presented. This article is a direct result of the Health Care Quality Improvement Program initiated by the Health Care Financing Administration, which has encouraged identification of quality improvement projects derived from analysis of patterns of care, and therefore required no special funding on the part of this Contractor. Ideas and contributions to the author concerning experience in engaging with issues presented are welcomed. Dr Wexler is currently affiliated with Massachusetts General Hospital, Boston, Mass. Dr Chen is currently affiliated with the University of Pennsylvania. Submitted September 3, 2000; accepted April 12, Reprints not available from the authors. Copyright 2001 by Mosby, Inc /2001/$ /1/ doi: /mhj Much attention has focused on strategies to prevent exacerbations and keep patients out of the hospital, both to improve quality of life and to contain costs for patients with heart failure. Studies suggest that disease management programs that adopt a multidisciplinary approach to heart failure can improve quality of life, reduce hospitalizations, and decrease costs in the elderly 1 and transplant candidates. 2 These interventions have reduced all-cause readmission, not merely readmission for heart failure. However, these programs are generally not being reimbursed. The investment in these programs will have the most clinical benefit and the greatest cost-effectiveness for patients with the highest risk of hospitalization and health expenditures. On the other hand, the program might not be economically attractive for patients at low risk. Accordingly, we sought to determine whether patients with heart failure could be stratified by their risk of incurring costs after discharge. Specifically, we aimed to identify patients with heart failure, on the basis of demographic and clinical characteristics at hospital discharge, who would be expected to gener-

2 American Heart Journal Volume 142, Number 2 Wexler et al 351 ate substantial costs over the next 6 months. If such patients could be identified, they would be ideal candidates for intensive outpatient monitoring and intervention, perhaps by enrollment in a disease management program. Even a small relative reduction in their costs could produce substantial savings, justifying investments in these programs. As a secondary objective, we sought to characterize the costs of 6- month hospital care in a representative sample of patients discharged with heart failure. Although many studies have estimated the aggregate costs of patients with heart failure, to our knowledge, this is the first that has followed a cohort of patients admitted for heart failure to determine correlates of in-hospital costs over the subsequent 6 months. Methods Study sample The study sample was composed of patients aged 65 years and older with a principal discharge diagnosis of heart failure (International Classification of Diseases, Ninth Revision, Clinical Modification [ICD-9-CM] codes , , , , , , , , and ) at 18 participating acute care Connecticut hospitals. All were fee-forservice insured patients identified from Medicare s National Claims History File. Selection of the patient sample occurred in 2 stages. In 1994 and 1995, 200 consecutive discharges on the basis of ICD-9-CM codes for heart failure were selected from each hospital for review. If a patient was admitted more than once during the sample period, the first hospitalization was selected as the index admission, with subsequent admissions counted as readmissions. These medical records were then reviewed to confirm diagnosis and collect clinical information. Patients were considered to have heart failure if they had symptoms of heart failure or consistent radiographic findings. 3 Patients were excluded if they had severe aortic stenosis, severe mitral stenosis, or heart failure resulting from an acute medical illness (acute myocardial infarction, hyperthyroidism, sepsis, etc), if their primary residence was outside of Connecticut, or if they died during the index hospitalization (7% of the original sample). Data sources Patient characteristics and clinical information were abstracted directly into standardized forms and entered into a computerized database by trained nurses and medical record technicians. During data collection, strategies to decrease abstraction errors and increase reliability included training sessions and detailed data definitions for each field. Also, a random subsample of cases was reabstracted for each abstractor followed by weekly team meetings to ensure reliability. Outcome variables, including 6-month allcause readmission, 6-month readmission resulting from heart failure, and charges associated with all-cause and heart failure related readmissions, were obtained from the Health Care Financing Administration Medicare Provider Analysis and Review file for Connecticut, which contains discharge abstracts for all Medicare-insured inpatients in Connecticut Table I. Variables included in bivariate analysis Demographic Age* Sex* Race* Medical history Atrial fibrillation CABG* CHF* COPD* Hypertension* Myocardial infarction* PTCA* Renal insufficiency* Stable angina* Stroke* Diabetes* Current smoking* Medications at discharge ACE inhibitor Aspirin β-blocker Calcium channel-blocker Warfarin Digoxin Diuretic Nitrate NSAID Vasodilator Course in hospital Cardiac arrest Hypotension* Infection* Myocardial infarction Pneumonia Kidney failure* Shock Stroke Length of stay* Clinical variables at discharge Creatinine* BUN* Ejection fraction* Terminal illness* Functional/mental status Dementia* Discharge to home* Confusion at discharge* Impaired ability to perform ADL Impaired mobility* Other Insurance status* Admissions for CHF before index hospitalization* Admissions for all causes before index hospitalization* Death 180 days after discharge CABG, Coronary artery bypass grafting; CHF, congestive heart failure; COPD, chronic obstructive pulmonary disease; PTCA, percutaneous transluminal coronary angioplasty; ACE, angiotensin-converting enzyme; NSAID, nonsteroidal anti-inflammatory drug; BUN, blood urea nitrogen; ADL, activities of daily living. *Variables that were included in multivariate analysis. hospitals. Costs were calculated by use of each hospital s cost-to-charge ratio listed in the Medicare Hospital Cost Report. 4

3 352 Wexler et al American Heart Journal August 2001 Independent variables The independent variables used in the initial bivariate analysis are listed in Table I. Candidate variables included demographic characteristics, medical history, hospital course at index hospitalization, in-hospital procedures, discharge medications, discharge clinical variables, functional and mental status at discharge, insurance status, and prior admissions. Data analysis All analyses were performed by use of standard statistical software. 5 Bivariate analyses were performed for both allcause and heart failure related readmission and costs to determine mean and median costs related to each independent variable. The distribution of costs was skewed, with extreme outliers. Because any population of patients with heart failure could be expected to contain such outliers, and because the actual cost of those outliers is relevant, we avoided truncating the sample, reassigning outliers to a lower percentile, or logtransforming the cost variables. The available data measure inpatient costs, so patients may generate cost only if they are readmitted to the hospital. Moreover, readmission is not a random event; sicker patients are readmitted. To estimate a model that is valid for the entire population of patients discharged with heart failure, regardless of whether patients were readmitted, we used a 2- stage limited dependent variable model proposed by Heckman. 6 The first stage is a probit model of readmission with the whole sample that calculates an instrumental variable, termed the inverse Mills ratio, which is included in the second stage, a cost regression that is run on the subset of patients who were readmitted. Conceptually, the Heckman procedure allows estimation of cost among those who generated cost, while accounting for the nonrandom likelihood of readmission using an instrumental variable. Analyses were performed for both all-cause readmissions and heart failure related readmissions with Heckman s 2-stage model to predict cost. Subsequently, analyses were conducted to examine the influence of events that occurred in the 6 months after discharge, but that could not be known at discharge. Variables for death occurring by 180 days after discharge, cumulative inpatient length of stay by 180 days after discharge, and procedures performed in the 180 days after discharge were added to the model to determine their impact on cumulative cost at 180 days after discharge and to check the validity of the model. Finally, on the basis of the results of the Heckman models, a risk score to predict patients likely to generate high costs was developed for each model. Coefficients in the risk score equations were derived from the parameter estimates for each of the significant variables in the linear regression on cost. These parameter estimates were standardized by dividing them by the smallest parameter estimate. The resulting quotient for each variable was used to calculate the risk score for the generation of heart failure related and all-cause costs, with and without age used in the equation. Each scale was divided into high-, medium-, and low-risk groups by the top quartile, middle half, and bottom quartile of calculated risk score. The costs of each group were calculated. In addition, costs for the highest and lowest risk groups identified by the presence or absence of risk factors were calculated. Results Characteristics of the study sample The characteristics of the study sample are summarized in Table II,A. The sample represents an elderly cohort nearly 70% of the patients were older than 75 years. During the 6 months after discharge, 512 (23%) patients were readmitted with a primary diagnosis of heart failure, and 1076 (49%) were readmitted for all causes, including heart failure. Twenty-nine percent were readmitted only once during this period, whereas 20% had 2 or more readmissions. Twenty percent of the patients (435) died by 180 days after discharge, 5% (109) of whom died within 30 days after discharge. Unadjusted outcomes are summarized in Table II,B. Medical resources and costs The mean cost per hospitalization for heart failure was $7174, and the mean cost per hospitalization for any cause was $8589. Of patients readmitted, those admitted for heart failure cost an average of $10,173. The median cost was $7179, with an interquartile range of $4181 to $11,611. Patients readmitted for all causes had an average cost of $14,327 and a median cost and interquartile range of $9629 and $4737 to $18,471. The mean cost of readmissions for heart failure alone was $2388 per discharged patient, averaged over all patients. The mean cost of all readmissions during this period was $7101, also averaged over the whole sample (Table II,B). Correlates of cost Bivariate analyses were performed for both heart failure related and all-cause readmission and costs (data not shown). Although many variables were significantly associated with differences in mean cost, for several variables the absolute difference in cost associated with a given characteristic was large and considered to be clinically relevant. For heart failure related cost, history of renal insufficiency (as well as elevated blood urea nitrogen and creatinine) and diabetes were associated with at least a $2000 difference in median cost ($1990 for diabetes, and more than $2000 for the others). For all-cause cost, only elevated creatinine and >2 prior admissions were associated with a median cost difference of at least $2000. History of chronic obstructive pulmonary disease and length of stay at index hospitalization >4 days were associated with a $1700 higher median cost in the subsequent 6 months. Although death in the 180 days after discharge was not an individual patient characteristic that could be known at discharge, it was a frequent event, occurring in 1 of 5 patients. Of those who died, slightly more than one third died without being readmitted to the hospital. In both heart failure related and all-cause samples, death was associated with at least $1500

4 American Heart Journal Volume 142, Number 2 Wexler et al 353 Table II, A. Baseline characteristics of the study sample (n = 2181) Independent variables Mean N Percent N missing Demographic variables Age, years 78.7 ± Male sex White race Medical history Ejection fraction Not measured <20% %-39% % Atrial fibrillation CHF COPD Hypertension Renal insufficiency Diabetes Current smoking Medications at discharge ACE inhibitor Digoxin Diuretic Calcium channel-blocker Insurance status Medicare Medicare plus other insurance Medicare plus Medicaid Other insurance CHF, Congestive heart failure; COPD, chronic obstructive pulmonary disease; ACE, angiotensin-converting enzyme. higher mean and median costs (P =.0001 for both samples) irrespective of whether the patient was readmitted. The probit stage of the Heckman 2-stage model demonstrated that patients with a history of heart failure, renal insufficiency, and diabetes; those who were discharged to home; and those who had more than 2 admissions in the 6 months before the index hospitalization (P <.001 in both samples) and ejection fraction <20% (P <.05 in both samples) were more likely to be readmitted. The second stage of the model identified independent correlates of cost (Table III). Both the heart failure related and all-cause models accounted for very little of the variation in costs during the 6 months after discharge. The model to predict costs related to heart failure explained only 5% of the variation, whereas the model predicting costs related to all-cause readmissions performed slightly better, explaining 7% of the cost variation. The models did achieve statistical significance, and the variables that emerged as significant are important in spite of their low overall explanatory value. Heart failure related cost was correlated with age and history of stroke (associated with lower cost), prior admission for heart failure, and history of kidney failure. Cost related to all-cause readmission was correlated with age (negative association), prior admission for heart failure, the instrumental variable representing the correlation between being readmitted and costs conditional on readmission (negative association), history of stroke (negative association), and history of hypertension. The correlates of cost were similar in both models, with age and past stroke predicting lower cost, and prior heart failure related admission predicting increased cost. The difference was that a history of kidney failure was likely to be associated with higher costs related to heart failure readmission, and history of hypertension was likely to be associated with higher costs related to readmission for any cause. Finally, the instrumental variable, which was adjusted for the influence of unobserved factors, was significant and negative in the model for all-cause costs, implying that use of the 2-stage model was warranted to control for such bias. When a 4-level categorical variable for ejection fraction (not measured, <20%, 20% to 39%, 40%) was forced into both multivariate models, the dummy vari-

5 354 Wexler et al American Heart Journal August 2001 Table II, B. Outcomes (n = 2181) Independent variables Mean No.Percent No.missing Admission and length of stay Readmissions for CHF by 180 days after discharge No readmissions readmission readmissions or more readmissions Readmissions for all causes by 180 days after discharge No readmissions readmission readmissions or more readmissions Distribution of length of stay of patients readmitted for any cause 1-3 days days days days Discharged to home Cost Mean CHF-related cost $ Mean all-cause cost $ Mean CHF cost per CHF admission $ Mean all-cause per admission $ Total 180-day CHF cost $5,208,534 Total 180-day cost $15,486,696 Death Death in hospital before discharge Death 180 days after discharge Death between 30 and 180 days CHF, Congestive heart failure. Table III. Results of the Heckman procedure: stepwise regression on cost Parameter Independent variable estimate SE F Prob > F Cost related to all readmissions, R 2 for model = Intercept 37, Inverse Mills ratio Age Prior CHF admission* History of stroke History of hypertension Costs related to readmissions with heart failure as a primary diagnosis code, R 2 for model = Intercept 20, Inverse Mill ratio Age History of stroke Prior CHF admission* History of kidney failure CHF, Congestive heart failure. *This is a dummy variable that is coded as 1 if the patient has been admitted for heart failure at least once in the 6 months before the index hospitalization. ables 40% versus not measured, 20% to 39% versus not measured, and <20% versus not measured were nonsignificant predictors of cost and did not change the models. The analysis that included events in the 6 months after discharge (cumulative inpatient length of stay, coronary artery bypass grafting, catheterization, percutaneous transluminal coronary angioplasty) in the 2-

6 American Heart Journal Volume 142, Number 2 Wexler et al 355 Table IV. Six-month costs by risk score model Model Number at risk Cost ($) F value P value CHF cost* High Medium Low CHF cost including age High Medium Low All cost High 15211,956 Medium Low All cost including age High Medium Low Score groups were stratified by top 25%, middle 50%, lowest 25%. One-way analysis of variance was used to determine differences between groups. CHF, Congestive heart failure. Formulas for calculating scores for each model: * 1.5 history of stroke + prior 6 months heart failure related admission history of kidney failure. 1 age 19 history of stroke + 12 prior 6 months heart failure related admission + 16 history of kidney failure. 1.2 prior 6 months heart failure related admission 1 history of stroke history of hypertension. 1 age 10 history of stroke + 9 prior 6 months heart failure related admission + 7 history of hypertension. Table V. Costs of highest and lowest risk groups Heart failure model All-cause model Lowest risk Highest risk Lowest risk Highest risk History of CHF History of hypertension N/A N/A 0 1 History of kidney failure 0 1 N/A N/A History of stroke Age > Number Mean cost $1516 $4120 $5282 $17,294 CHF, Congestive heart failure. stage model explained 79% of the variation in costs. The number of inpatient days in the 6 months after discharge alone contributed 71% of the explanatory power. A secondary analysis on the subset of patients with low ejection fraction (<40%) did not show angiotensin-converting enzyme inhibitors at discharge to be a significant predictor of lower costs. Although the discharge variables that were significant in the cost models explained only a small fraction of the variation in cost, they nonetheless proved useful in stratifying subsets of patients at risk of generating particularly high costs in the 6 months after discharge. Formulas derived from the cost models and the subsequent categorization of patients into risk strata are listed in Table IV. The models that included age in the score equation predicted that the group with all of the risk factors would generate $2604 more than the group with none of the risk factors for heart failure related readmissions. The model of costs related to all-cause admission showed that the highest risk group would have an excess of $12,012. The lowest risk group, which had none of the risk factors, generated substantial mean costs of $1516 for heart failure related admissions and $5282 for all-cause admissions (Table V). In addition, these lowest risk groups were very small. The group of patients with no risk factors for increased costs constituted only 9% of the group of patients readmitted for heart failure and 3% of the group readmitted for all causes. The basic risk score models, both including and excluding age (because all patients were

7 356 Wexler et al American Heart Journal August 2001 elderly) showed a significant trend toward increasing costs with increasing risk factors, but no truly low-risk group could be identified. Discussion The primary purpose of this analysis of 2181 patients aged 65 years with a discharge diagnosis of heart failure was to identify patients at hospital discharge who were likely to generate substantial hospital costs in the subsequent 6 months. The best predictive model on the basis of factors known at discharge accounted for very little of the variation in cost only 5% of the heart failure related cost and 7% of all cost was explained by these variables. Thus we found that commonly available variables at discharge do not account for most of the variation in future costs. Risk stratification scores derived from the models revealed that no low-risk group of patients could be identified, although patients with certain clinical characteristics were at risk for excess costs. Given the substantial costs incurred by patients in this group and the difficulty of stratifying them, all elderly patients discharged after an episode of heart failure should be considered to have a high risk. This result is consistent with previously published results. In a study that characterized the correlates of inhospital costs for acute myocardial infarction, only 7% of the variation was explained by demographic, clinical, and discharge characteristics, and similarly, a study of hospitalization rates and chronic disease found that sociodemographic, location, and clinical variables explained only 7% of hospitalizations. 7 In our sample, inclusion of cumulative inpatient days and procedures performed 180 days after discharge as explanatory variables yielded a model that explained 79% of the variation in cost, similar to findings from Krumholz et al 4 that showed hospital treatments and procedures explained 53% of the variation in cost for acute myocardial infarction patients. Although including the treatments and procedures that patients with heart failure receive in the 6 months after discharge explains more of the cost, it does not help us to identify those patients before they require such treatment, the goal of this study. A secondary purpose of this study was to characterize this cohort of patients and the in-hospital costs associated with caring for them. In many respects, this sample represents a typical cohort of patients with heart failure. The all-cause readmission rate of 49% over 6 months after discharge is consistent with earlier estimates, as is the in-hospital mortality rate of 7% and average length of stay of index hospitalization of 8.25 days (median 6.5). 8 The burden of disease in this study was substantial. The average cost of a readmission resulting from heart failure was $7174; for any cause, the average cost was $8589. Another recent study of Medicare-insured patients with chronic heart failure estimated the cost of an average admission to be between $7668 and $ One reason for the discrepancy between heart failure related and all-cause readmissions may be coding practices. One study found that heart failure related admissions resulting from sodium retention cost less than those from other causes, with fewer intensive care unit admissions and procedures but a higher readmission rate. 8 It may be that in our study, primary exacerbations of heart failure were given a principal discharge diagnosis code of heart failure, whereas angina, acute myocardial infarction, and infection had different primary diagnosis codes and warranted more-expensive treatment. Patients with heart failure who are admitted to the hospital for any cause may not have a primary diagnosis of heart failure recorded as a reason for readmission, but it is likely in many cases that the readmission is related to having heart failure as a comorbid condition. A 10-year National Hospital Discharge Survey found that only one third of hospitalized patients with heart failure had it listed as the principal diagnosis. 10 In our models, clinical characteristics such as recent heart failure related admissions, kidney failure, and hypertension were associated with increased costs, and older age and history of stroke were associated with decreased costs. Other studies have shown that some variables associated with greater burden of illness, such as age and selected comorbidities, are paradoxically associated with lower resource consumption. Patients with these characteristics potentially receive less aggressive treatment, resulting in lower cost. Kidney failure was a significant predictor of heart failure related readmission costs but not of all-cause readmission costs. Our results may reflect the variation in documentation of this imprecise variable rather than a true lack of association of previous kidney failure with all-cause readmission costs. Using readily available clinical characteristics, we identified a particularly high-risk subset that is likely to generate excess costs of $4120 for heart failure related admissions and $17,294 for all admissions, on average. In our sample 25% of all hospitalizations, including the index admissions, were readmissions with a primary diagnosis code of heart failure. If this figure can be extrapolated to the 1 million admissions per year nationwide, then 25% of those are readmissions that cost approximately $2.5 billion. Preventing only 10% of those admissions would yield a cost savings of $250 million. This study is purposely designed to focus on inpatient costs. Although inpatient costs represent most expenditures, other costs should not be ignored. In another study, hospitalization for patients with heart failure was estimated at 56% of total costs, whereas outpatient medical care (13%), drugs (16%), and various tests (12%) represented smaller but still significant fractions. 11 Nonetheless, the importance of hospital costs

8 American Heart Journal Volume 142, Number 2 Wexler et al 357 justifies a study that focuses specifically on this outcome. Another important issue is that although the hospitals represented a cross-section of community, private, and tertiary care hospitals, the standard of care and associated expenditures in Connecticut may not accurately represent the country at large. The study does, however, present the experience of an entire state, a much more representative sample with detailed chart data than other studies have achieved. Another consideration is that the cost data are 5 years old. There may be a trend toward decreasing readmission costs as management strategies for heart failure and coronary artery disease improve, such as increasing penetration of angiotensin-converting enzyme inhibitors into clinical practice and use of beta blockers as first-line therapy. Studies have also shown that patients with heart failure with lower costs tend to be younger, have fewer comorbidities, and be enrolled in managed care plans, whereas most patients with heart failure are aging, have multiple comorbidities, and remain enrolled in Medicare fee-for-service plans. 12 Although the overall effect of these differences on costs is hard to predict, the correlates of cost would be expected to be similar. Also, although many categories of explanatory variables were included in the models, including variables for dementia, mental status, and mobility, other potential factors were not captured in this analysis, such as social support, emotional stability, and economic resources. Finally, because patients were initially selected by use of ICD-9-CM codes, variability in coding practices may have occurred. However, verifying signs and symptoms of heart failure by use of medical record review increased the specificity of diagnosis. In this study, we analyzed a population of elderly patients with heart failure and demonstrated that readmission is common and that the costs associated with these readmissions are substantial. Given the emerging evidence for effective programs to reduce readmission and considerable costs, investments in this area could be economically favorable even if they produce modest reductions in risk. Although disease management programs may prevent admissions, the current payment structure does not allow them to be funded. These results suggest that compensating health care providers for preventive education and follow-up could be cost-beneficial. Finally, because no low-risk group can be easily identified, our findings suggest that these interventions should be directed to all Medicareinsured patients who survive hospitalization with heart failure. References 1. Rich MW, Beckham V, Wittenberg C, et al. A multidisciplinary intervention to prevent the readmission of elderly patients with congestive heart failure. N Engl J Med 1995;333: Fonarow GC, Stevenson LW, Walden JA, et al. Impact of a comprehensive heart failure management program on hospital readmission and functional status of patients with advanced heart failure. J Am Coll Cardiol 1997;30: Vaccarino V, Chen Y, Wang Y, et al. Sex differences in the clinical care and outcomes of congestive heart failure in the elderly. Am Heart J 1999;138: Krumholz HM, Chen J, Murillo JE, et al. Clinical correlates of in-hospital costs for acute myocardial infarction in patients 65 years of age and older. Am Heart J 1998;135: The SAS Institute. The SAS System ed. Cary, NC; Heckman JJ. Sample selection bias as a specification error. Econometrica 1979;47: Nelson EC, McHorney CA, Manning WG, et al. A longitudinal study of hospitalization rates of patients with chronic disease: results from the Medical Outcomes Study. Health Serv Res 1998;32: Bennett SJ, Saywell RM, Zollinger RW, et al. Cost of hospitalization for heart failure: sodium retention versus other decompensating factors. Heart Lung 1999;28: Sobotka PA, O Connell JB. Malaligned financial incentives of providers in the treatment of chronic diseases: the case of chronic heart failure. CHF 1999;January-February: Haldeman GA, Croft JB, Giles WH, et al. Hospitalization of patients with heart failure: National Hospital Discharge Survey, Am Heart J 1999;137: Mackowiak J. Cost of heart failure to the healthcare system. Am J Manag Care 1998;4:S Philbin EF, DiSalvo TG. Managed care for congestive heart failure: influence of payer status on process of care, resource utilization, and short-term outcomes. Am Heart J 1998;136: