Trajectory of Illness for Patients with Congestive Heart Failure

Abstract

Background and Objective:

This study describes progression to death for patients with congestive heart failure (CHF).

Methods:

We used SAS procedure Proc Traj to fit a semiparametric model to longitudinal data on prognosis of patients with CHF in the 12 months prior to death. Data were collected on 744 patients with CHF in 2010 at Bay Pines VA Healthcare System; 386 subjects had sufficient data points (minimum of five encounters) to trace their risk in 12 months prior to death. The prognosis of the patient was calculated using the comorbidities of the patient.

Results:

Unexpected death occurred in 20.5% of patients; all remaining patients had a gradual progression toward death. For 13.3% of patients, progression toward death started 12 months prior to death. For 29.9% of patients, increased risk started at 6 months prior to death. For 36.3% of patients, it started 3 months prior to death. One month prior to death, 79.5% of the patients had a more than 97% chance of mortality. It may be possible to use progression toward death over 3 consecutive months as a predictor of need for hospice consultation.

Conclusions:

Five typical illness trajectories have been described for patients with progressive heart failure. The needs of patients and their caregivers are likely to vary according to the trajectory patients are following. Contrary to reports in the literature about unexpected death in patients with CHF, the majority of decedents in our study had a predictable and gradual progression toward death. Recognizing these trajectories may help clinicians implement an appropriate plan to meet the needs of patients and their caregivers.

Introduction

With more than 5.8 million persons affected, heart failure in Americans is on the rise.¹ Survival rates of patients with myocardial infarction (MI) continue to improve, but with that advance comes a rise in disabling heart failure; within 6 years of a MI 22% of men and 46% of women are disabled with heart failure. The illness and death trajectory of heart failure is unpredictable.² Heart failure is typically characterized by periods of stability interrupted by episodes of acute exacerbation.^3,4 Patients with cancer present with steadily functional deterioration and usually a clear terminal phase prior to death. In contrast, patients with congestive heart failure (CHF) present with episodes of acute deterioration followed by some recovery, but no clear functional deterioration prior to death.⁵ O'Leary and colleagues reported that at 6 months before death patients with heart failure had significantly better social functioning than patients with cancer.⁶ Levenson and coworkers reported that patients with CHF experienced a relatively good quality of life:

There is no significant decrement in quality of life as death approaches. Reflecting the unpredictable course of CHF during the last month of life, many patients have good median model-based 6-month prognoses and enjoy good to excellent quality of life.⁷

At the same time, other patients with heart failure experience pain and functional disabilities like those of patients with cancer.⁸ Collectively, studies show that patients may experience the end stage of heart failure in radically different ways. This article describes the trajectory of patients with end-stage heart failure in the last 12 months of life.

One obvious explanation of why patients with end-stage heart failure have different experiences could be their comorbidities. Attempts to develop accurate predictors of risk of mortality for patients with CHF are numerous.^9–15 Existing prognostic indices examine measures of heart function (e.g., peak oxygen consumption, B-type natriuretic peptide), creatinine levels, cardiac medications (e.g., beta blockers), and implanted devices. These indices do not include comorbidities that contribute to mortality. Patients with CHF have many concurrent diseases, including cancer, diabetes, renal, or other problems. A recent study found that 45% of patients with CHF died from noncardiac causes.¹⁶ Prognostic indicators that overlook mortality from comorbidities of CHF may be suboptimal. The claim that the course of CHF is unpredictable may be due to inaccurate measures of prognosis. The current study examines the trajectory of patients with CHF using the patients' various comorbidities.

Clinical data on the patient's heart condition were deliberately excluded (e.g., heart rate, blood pressure, serum creatinine, ejection fraction, etc.) to allow a focus on CHF comorbidities only. Clinical data are easily available in heart failure encounters but rarely available in encounters for other conditions. By focusing on comorbidities we ensured that prognostic indicators could be evaluated in every encounter.

Methods

Type of study

This is a descriptive classification study. Readers may be familiar with traditional studies, wherein patient attributes classify them into different groups. In contrast, this study examines patient characteristics over time. Thus two patients who have the same prognosis at one point in time but different rates of change are classified into two different groups. This allows the identification of distinct subgroups within the population that have a similar rate of change in prognosis.

Source and type of data

Retrospective diagnoses data were used to measure prognosis over the last year of life in patients with CHF. This approach focuses on comorbidities of patients with CHF and classifies patients into different trajectories based on their comorbidities. Data were examined from the electronic health records of patients with CHF seen in the last 5 years prior to 2011 in the Bay Pines VA Healthcare System. A brief description of the patients is provided in Table 1. The median CHF decedents in our sample were white, non-Hispanic, married, male, and 80 to 90 years old.

Table 1.

Demographics of 744 CHF Decedents

Category	Percent
Not reported or unknown race	28.63%
American Indian or Native Alaskan race	0.13%
Black or African American	1.61%
Native Hawaiian or other Pacific Islander	1.61%
White	68.01%
Not reported or unknown ethnicity	27.55%
Hispanic origin	0.81%
Non-Hispanic origin	71.64%
Male	97.85%
Divorced	22.18%
Married	55.51%
Never married	3.63%
Separated	1.75%
Widow/Widower	16.67%
30 to 40 years old	0.13%
40 to 50 years old	0.67%
50 to 60 years old	5.65%
60 to 70 years old	15.05%
70 to 80 years old	22.04%
80 to 90 years old	48.79%
90 to 100 years old	7.66%

Eligibility criteria

To be eligible for inclusion in the study the patient must have met the following criteria:

CHF: The patient must have had a diagnosis of CHF, defined as International Classification of Diseases, 9th Revision (ICD-9) codes 402.01, 402.11, 402.91, 425.1, 425.4, 425.5, 425.7, 425.8, 425.9, 428.0, 428.1, 428.20, 428.21, 428.22, 428.23, 428.27, 428.30, 428.31, 428.32, 428.33, 428.40, 428.41, 428.42, 428.43, or 428.9. In addition to these diagnoses, patients presented with various other diagnoses, which were considered the CHF comorbidities.

Reported date of death: All patients must have a reported date of death. Reliance on deaths recorded in electronic health records is problematic because some study subjects may have died at home and their death may not be recorded. Several studies have shown that, because financial benefits are linked to reported date of death for veterans, the date of death is reported accurately¹⁷ We identified 744 patients with CHF with reported date of death during the study time frame.

Minimum data required: The patient should have at least 5 months in which they had an encounter with the VA health care system in the 12 months preceding death. In short, we examined only those patients who were closely monitored within the VA. Patients with fewer visits were assumed to have received part of their care outside of the VA and therefore they were excluded from the analysis. Among the patients with reported dates of death, 378 met the criteria of at least 5 months of visits in the 12 months preceding death.

Occasionally, there were inconsistent dates of death. Patients with a recorded date of death 30 days prior to their last visit were excluded from the analysis. A 30-day interval was allowed because at times test results become available after date of death or visits are coded to have been completed after date of death.

Prognosis of CHF comorbidities

Patients' prognosis within a particular month was calculated from their diagnoses during that month. To accomplish this, we assigned severity scores to a pair of diagnoses/procedure codes. This is similar in concept to the approach first examined by Alemi and Walter's Severity Index.¹⁸ A brief description of the procedure is provided here:

1. Prognosis of a pair of diagnoses: For each pair of diagnoses (across all patients), we calculated the average number of days to mortality. The average was standardized to range from 0 to 1 by dividing by the maximum days to mortality for any pair of diagnosis. We examined pairs of diagnoses during an encounter to insure that we accounted for interactions between diagnoses. For example, the combined CHF and MI had a different prognosis than either one separately or the average of the two. We excluded pairs of diagnoses that do not repeat at least five times in the dataset, on grounds that these estimates were unreliable.

2. Prognosis for diagnoses during a month: A patient's prognosis within a month was based on the average prognosis for all pairs of diagnoses during that month.

Definition of unexpected death

We defined gradual progression toward death as patients who start at lower risk groups and add 0% to 5% to the chance of mortality month after month. We defined the unexpected death group as patients who had died despite the fact that one month prior to death they had a less than 80% chance of mortality. A 20% jump in mortality risk was considered an unexpected change.

Method of statistical analysis

The data were analyzed using trajectory modeling.^19,20 Decedents were classified into the five groups. The focus of the Proc Traj procedure is on grouping membership and identifying distinct subgroups within the population. Proc Traj does not provide any individual-level information on the pattern of change over time; subjects are grouped and it is assumed that every subject in the group follows the same trajectory. We used the trajectory censored normal modeling. We used the Bayesian information criterion (BIC) to test from two to six trajectories and to determine whether each trajectory was the best fit. All analyses were performed with the use of SAS software (version 9.2; SAS Institute Inc., Cary, NC). A two-sided p value of less than 0.05 was considered to indicate statistical significance. The software reports the proportions of decedents classified according to each CHF trajectory. Additionally, the analysis reported 95% confidence intervals using 1000 bootstrap samples.

We also examined if certain diagnoses could anticipate patient's classification into a specific trajectory group. To do so we examined the likelihood ratio associated with the most severe diagnoses. For diagnosis “d” we calculated the likelihood ratio of belonging to a specific trajectory group as follows: \documentclass{aastex}\usepackage{amsbsy}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{bm}\usepackage{mathrsfs}\usepackage{pifont}\usepackage{stmaryrd}\usepackage{textcomp}\usepackage{portland, xspace}\usepackage{amsmath, amsxtra}\pagestyle{empty}\DeclareMathSizes{10}{9}{7}{6}\begin{document} \begin{align*}LR{_d}=\frac {\rm Prevelance \,of\,{\it d} \,among\,members\,of\,trjectory\,group}{\rm Prevalence\,of\,{\it d}\,among \,members\,of\,other\,trajectory\,groups}\end{align*} \end{document}

The more the likelihood ratio is different from 1, the more informative it is. Large likelihood ratios indicate diagnoses that increase probability of membership into the trajectory group. Likelihood ratio less than 1 can be used to rule out membership in the trajectory group. Bayesian probability models use the likelihood ratio associated with the patient's diagnoses to predict the posterior odds of membership in any specific trajectory group.

We reported the cross-validated accuracy of predicting membership in trajectory groups by examining the accuracy of Bayesian predictions in 1/5 of data set aside for validation of the Bayesian probability model. The accuracy of patients' diagnoses in classifying patients into unexpected death versus gradual progression was tested using Receiver Operating Characteristic (ROC) curve. The C-statistic, which calculates the area under the ROC curve, was used as a measure of accuracy of the predictions.

IRB review

This project was reviewed and approved by Bay Pines VA Healthcare System.

Results

Fig. 1 shows trajectories of CHF patients in the last year of life. The probability of mortality is represented by the Y-axis, as established by the severity of the comorbidities associated with the patient's CHF during specific months. The X-axis shows months before mortality; it ranges from 12 months to 1 month. At the zero month (not shown), all decedents died and therefore have the probability of 1.0 of dying. The probability of mortality in the months prior to death ranged from low of 0.73 to high of 0.95. Progression to death was shown by increase in monthly probability of mortality. The Trajectory Model classified the patterns of progression towards death into five distinct groups of patients. These groups ranged from medium to high risk of mortality. It should have been expected that no groups would have low risk of mortality, as all patients in this study die in 12 months. The five progressions towards death identified through Trajectory Modeling are:

FIG. 1.

Trajectory of congestive heart failure.

Constant Moderate Risk and Unexpected Death (abbreviated to Unexpected Death): This group indicates patients that are at constant moderate risk of mortality, whose condition is not getting worse over time, but unexpectedly die. This group fits best the notion that CHF patients die in unpredictable fashion. Of the patients examined, 78 patients, 20.5% fall in this group. Despite stable risk of mortality throughout the 12 months these patients' risk of mortality jumps 22 points, from 0.78 to 1.0 after the last month. The jump in mortality risk suggests an unanticipated risk of mortality.

Rapid Progression: This group starts at the lowest risk of death, probability of 0.73, among the five decedent groups. The group shows no increased risk of mortality for six months; however, six months prior to death they begin to show elevated risk. In the last six-months, patients in this group increase their risk of mortality to the highest level, probability of 0.95, among all five groups. In the last six months, this group progresses towards death at the fastest rate than any other gorup. For these patients death is not unexpected as the rapid decline in their health increases the risk of death. Of the decedents, 13.0% fell into this group.

Late Progression: This group starts at moderate risk of mortality, with a probability of 0.83. Decedents in this group maintain their prognosis for 9 of the 12 months but in the last 3 months increase their risk by 17 points to 0.90. The last 3 months of data for these patients reveal a progressively sicker patient and increasing likelihood of death. Of the decedents, 36.3% fell into this group.

Early Progression: This group starts at relatively low probability of mortality among the decedents, 0.76, and within the first 9 months increases the probability of mortality by 15 points to 0.91. For the next 3 months, the patients maintain their high risk of mortality and then die. Death of these patients could be anticipated given their high risk of mortality. Of the decedents, 13.3% fell into this group.

Constant High Risk: This group starts at highest risk of mortality among the decedents, with a probability of 0.89, and gradually risk increases to 0.96. Each encounter is for diagnoses that are severe and have high probability of mortality. Of the decedents, 16.9% fell into this group.

Table 1 shows the top 10 diagnoses that rule out or rule in membership in the Unexpected Death group. Pairs of diagnoses were used to establish membership in Unexpected Death group. The top diagnoses that rule in Unexpected Death were: “long-term (current) use of anticoagulants,” “acute, but ill-defined, cerebrovascular disease,” “encounter for therapeutic drug monitoring,” “mixed hyperlipidemia,” “atrial fibrillation,” and “of unspecified type of vessel, native or graft.” The diagnoses that rule out membership in the Unexpected Death group included: “speech therapy,” “dysphagia, oropharyngeal phase,” “dehydration,” “unspecified hypothyroidism,” “long-term (current) use of anticoagulants,” “other,” and “bone and bone marrow malignant neoplasm of prostate.”

Of interest are the causes of mortality for the Unexpected Death group. These causes can be inferred from the nature of diagnoses that increase the probability of membership in this group. Patients may die from CHF itself or from accompanying comorbidities. Table 2 illustrates that most (86%) of the diagnoses predictive of membership in the Unexpected Death group involved cardio- or cerebrovascular problems, whereas all of the diagnoses in the rule-out group involved noncardiac or noncardiovascular problems. Clearly, the Unexpected Death group had more cardiac or cardiovascular diseases.

Table 2.

Predictors of Membership in Unexpected Death Group

Pair of diagnoses that rule in Unexpected Death trajectory			Pair of diagnoses that rule out Unexpected Death trajectory
ICD9 code	Description	Likelihood ratio	ICD9 code	Description	Likelihood ratio
V58.61 & 436	Long-term (current) use of anticoagulants and acute, but ill-defined, cerebrovascular disease	33.2	198.5 & 185	Bone and bone marrow and malignant neoplasm of prostate	1/16.7
V58.83 & 250.00	Encounter for therapeutic drug monitoring and diabetes mellitus	29.2	V58.61 & 415.19	Long-term (current) use of anticoagulants and other pulmonary embolism and infarction	1/16.7
V58.83 & 272.2	Encounter for therapeutic drug monitoring and mixed hyperlipidemia	29.2	250.00 & 244.9	Diabetes mellitus and unspecified hypothyroidism	1/14.3
436 & 427.31	Acute, but ill-defined, cerebrovascular disease and atrial fibrillation	20.7	276.51	Dehydration	1/12.5
V58.61 & 414.00	Long-term (current) use of anticoagulants and coronary atherosclerosis of unspecified type of vessel, native or graft	20.7	V57.3 & 787.22	Speech therapy and dysphagia, oropharyngeal phase	1/12.5

We used the likelihood ratios associated with most severe diagnoses in each of the 12 months preceding death to predict membership in the Unexpected Death group. If severe diagnoses are causes of death, then predicting membership in the Unexpected Death group is akin to arguing that patients in this trajectory have distinct causes of death. Likelihood ratios were estimated from the experience of 299 training cases and cross validated on 87 validation cases. An ROC curve was organized to show the accuracy of prediction. The curve shows both the sensitivity and specificity of the predictions. (See Fig. 2.) The area under the curve measures the extent of accuracy. This area was calculated to be 0.75.

FIG. 2.

Receiver operating characteristic curve.

Discussion

We have shown that CHF patients follow different trajectories to death; this may not be surprising to most clinicians. This report confirms clinical observations that CHF patients have different trajectories. It describes five distinct trajectories. These trajectories contradict the observation that progression to mortality for patients with CHF is unexpected or unpredictable. For the majority (79.5%) of patients progression was gradual. Only a small minority had unpredictable mortality, where the patients' risk of mortality jumped by 20%. For all other patients, progression to mortality was gradual, either early in the first 12 months, or late in the last 6 or 3 months. In fact, one month prior to death, all except those in the Unexpected Death group had a greater than 90% chances of mortality.

Many patients attempt to gain control over their illness by acquiring knowledge about how it is likely to progress.²¹ Patients and their caregivers may want to know which of the five groups may describe their progression. Patients' comorbidities can be used to classify the patient into one of the five groups. When patients know the possible trajectory of their illness, they can plan better. A dialogue about the illness trajectory among patient, family, and professionals can allow for the provision of realistic hope and supportive care, focusing on quality of life, and symptom control. Death is inevitable but whether disability and progression toward death occurs gradually or unexpectedly has profound implications.^22,23 Many patients describe the “good” death as one without significant functional disability.^24,25 Patients often express an interest in knowing how they will deteriorate and whether they can have a controlled, dignified end.²⁶ For instance, patients within the Rapid Progression group should expect more admissions to the hospital and higher caregiver burden, and perhaps higher functional disabilities in the 6 months prior to death. Patients and families in the Unexpected Death category may want to plan for a relatively consistent need for caregiving over time. The trajectories described may help clinicians and patients set their expectations.

This study did not create a prognostic index but could be used to generate two hypotheses about how these indices could be improved. First, our data showed that it is important to fully take into account the patients' various comorbidities, as many die from non-CHF comorbidities. Some noncardiac comorbidities may be considered the systemic manifestations of heart failure. These comorbidities mark the progression of the disease from one organ to another. Expanding the variables used in prognostic indicators to include noncardiac comorbidities may be reasonable.

Second, our data suggest that mortality may be better predicted from change in prognosis than the value of the index at any particular point of time. For example, Figure 1 showed many patients at elevated risk who do not die shortly afterwards. For example, patients in the Constant High Risk group had a 90% of chance of death, yet they survived month after month. At 12 months prior to death, relying on the elevated risk scores would not have been accurate. At this point in time, admitting these patients to a hospice program would require repeated renewels of eligibility under the Center for Medicare and Medicaid Services. This also holds true for the 11th month prior to death. In fact, month after month, the patients in the Constant Elevated Risk group survived despite having more than 90% chance of mortality. In contrast, if we examine the change in prognosis, we see a consistent increase in risk of mortality starting 4 months prior to death. In fact, the trajectories for all groups, except the Unexpected Death group, showed an increased risk of mortality within 6 months of death. Therefore, the increased risk, and not the starting risk value, maybe a better predictor of death. None of the current CHF prognostic indices focus on rate of change. Our data suggest a way of improving prognostic indices by switching from risk scores to changes in risk scores.

Our data also suggest how far ahead of time one could anticipate the impending death. Every patient dies at some point in time. The difficulty with estimating patients' prognoses is in predicting the length of survival time. Predicting mortality only becomes useful, and difficult, when one gives it a time frame. The question of what is a suitable time frame for predicting mortality may be policy related or empirical, as when data are analyzed to find the largest interval in which death can be anticipated. We did not look at patients who lived. Therefore, we cannot be sure what is an optimal time frame for predictions. Nevertheless, the data from decedents are suggestive. Suppose we predict that in the Rapid Progression group, anyone whose prognosis grows worse by 5% is likely to die within a month. Our prediction would be wrong in many instances. We see five periods in which the patient increased his/her risk of mortality by 5% and in only one of these five times the patient died. Clearly, using a 5% increase is a poor predictor of mortality for this group of patients. Suppose that we predict that patients will die if they have two consecutive 5% increases in risk of mortality. There were two instances that met this criterion, the first one was the change from 4 month prior to death to 2 months prior to death, and the second was the change from 3 months prior to death to one month prior to death. Two consecutive 5% increases in prognosis may be a marker for death in the Rapid Progression group. Our data provide clues to how soon we can predict a patient's mortality.

This study is limited in several ways. First it is descriptive and not normative; it is possible that treatment or quality of care may affect membership in different trajectories of illness. Progression of the disease depends on quality of care. We have not controlled for these differences. Second, the study was limited to patients with CHF within one institution and therefore may reflect the idiosyncratic institutional policies that are not relevant to other settings. Third, this is a small sample of patients; a study with a larger sample of patients may change our conclusions. Fourth, we rely on comorbidities to calculate risk of mortality; these are based on patients' diagnoses as coded in the ICD-9. These codes may not reflect the patients' true condition. Fifth, many clinicians are interested in better methods of predicting patients prognosis; we have not shown that we can predict mortality of a cohort of patients with CHF. Such a cohort would include patients who live and therefore were not part of this study. Our data showed that there are many patients at high risk who survive for long times and many at moderate risk who die quickly. These data highlight the difficulty of predicting which patient with CHF will live and which will die. Future studies could use the information learned in this study to design tools that can predict mortality among a cohort of patients.

Footnotes

Acknowledgments

We acknowledge help received from John Winiecki in the initial phases of the analysis of these data. This work has benefited from data samples prepared under direction of Dr. William Liao. The paper was edited by Joseph Davis. This project was supported by the resources of the District of Columbia and the Bay Pines Veterans Affairs Healthcare Systems.

Author Discosure Statement

The contents of this paper do not represent the views of the Department of Veterans Affairs or the United States government. Authors have no conflict of interest. The procedures described here for measurement of severity of comorbidities are in the public domain.

References

Bui

, Horwich

, Fonarow

. Epidemiology and risk profile of heart failure. Nat Rev Cordiol, 2011; 8:30–41.

Hanratty

, Hibbert

, Mair

, May

, Ward

, Capewell

, Litva

, Corcoran

. Doctors' perceptions of palliative care for heart failure: Focus group study. BMJ, 2002; 325:581–585.

Field

, Cassel

. Approaching Death: Improving Care at the End of Life. Washington, D.C.: National Academy of Medicine, 1997.

Gott

, Barnes

, Parker

, Payne

, Seamark

, Gariballa

, Small

. Dying trajectories in heart failure. Palliat Med, 2007; 21:95–99.

McCarthy

, Phillips

, Zhong

, Drews

, Lynn

. Dying with cancer: Patients' function, symptoms, and care preferences as death approaches. J Am Geriatr Soc, 2000; 48,5 Suppl:S110–S121.

O'Leary

, Murphy

, O'Loughlin

, Tiernan

, McDonald

. A comparative study of the palliative care needs of heart failure and cancer patients. Eur J Heart Fail, 2009; 11:406–412.

Levenson

, McCarthy

, Lynn

, Davis

, Phillips

. The last six months of life for patients with congestive heart failure. J Am Geriatr Soc, 2000; 48,5 Suppl:S101–S109.

Bekelman

, Rumsfeld

, Havranek

, Yamashita

, Hutt

, Gottlieb

, Dy

, Kutner

. Symptom burden, depression, and spiritual well-being: A comparison of heart failure and advanced cancer patients. J Gen Intern Med, 2009; 24,5:592–598.

Aaronson

, Schwartz

, Chen

, Wong

, Goin

, Mancini

. Development and prospective validation of a clinical index to predict survival in ambulatory patients referred for cardiac transplant evaluation. Circulation, 1997; 95:2660–2667.

10.

Rose

, Moskowitz

, Packer

, Sollano

, Williams

, Tierney

, Heitjan

, Meier

, Ascheim

, Levitan

, Weinberg

, Stevenson

, Shapiro

, Lazar

, Watson

, Goldstein

, Gelijns

. The REMATCH trial: rationale, design, and end points: Randomized Evaluation of Mechanical Assistance for the Treatment of Congestive Heart Failure. Ann Thorac Surg, 1999; 67:723–730.

11.

Lee

, Austin

, Rouleau

, Liu

, Naimark

, Tu

. Predicting mortality among patients hospitalized for heart failure: derivation and validation of a clinical model. JAMA, 2003; 290:2581–2587.

12.

Fonarow

, Adams

Jr , Abraham

, Yancy

, Boscardin

. Risk stratification for in-hospital mortality in acutely decompensated heart failure: Classification and regression tree analysis. JAMA, 2005; 293:572–580.

13.

Brophy

, Dagenais

, McSherry

, Williford

, Yusuf

. A multivariate model for predicting mortality in patients with heart failure and systolic dysfunction. Am J Med, 2004; 116:300–304.

14.

Lund

, Aaronson

, Mancini

. Predicting survival in ambulatory patients with severe heart failure on beta-blocker therapy. Am J Cardiol, 2003; 92:1350–1354.

15.

Koelling

, Joseph

, Aaronson

. Heart failure survival score continues to predict clinical outcomes in patients with heart failure receiving beta-blockers. J Heart Lung Transplant, 2004; 23:1414–1422.

16.

Muzzarelli

, Leibundgut

, Maeder

, Rickli

, Handschin

, Gutmann

, Jeker

, Buser

, Pfisterer

, Brunner-La Rocca

. TIME-CHF investigators: Predictors of early readmission or death in elderly patients with heart failure. Am Heart J, 2010; 160:308–314.

17.

Sohn

, Arnold

, Maynard

, Hynes

. Accuracy and completeness of mortality data in the Department of Veterans Affairs. Popul Health Metr, 2006; 4:2.

18.

Alemi

, Walters

. A mathematical theory for identifying and measuring severity of episodes of care. Qual Manag Health Care, 2006; 15:72–82.

19.

Jones

, Nagin

, Roeder

. A SAS procedure based on mixture models for estimating developmental trajectories. Sociol Methods Res, 2001; 29:374–393.

20.

Jones

, Nagin

. Advances in group-based trajectory modeling and an SAS procedure for estimating them. Sociol Mehods Res, 2007; 35:542–571.

21.

Asbring

, Narvanen

. Patient power and control. Qual Health Res, 2004; 14:226–240.

22.

Murray

, Kendall

, Boyd

, Sheikh

. Illness trajectories and palliative care. BMJ, 2005; 330:1007–1011.

23.

Buetow

, Coster

. Do general practice patients with heart failure understand its nature and seriousness, and want improved information? Patient Educ Couns, 2001; 45:181–185.

24.

Nahlén

, Saboonchi

. Coping, sense of coherence and the dimensions of affect in patients with chronic heart failure. Eur J Cardiovasc Nurs, 2010; 9:118–125.

25.

Buetow

, Goodyear-Smith

, Coster

. Coping strategies in the self-management of chronic heart failure. Fam Pract, 2001; 18:117–122.

26.

Asbring

, Närvänen

. Patient power and control: A study of women with uncertain illness trajectories. Qual Health Res, 2004; 14:226–240.