With more than 5 million Americans living with heart failure (HF), and another 670,000 new cases being diagnosed each year, HF is the most costly diagnosis in the Medicare population and is the most common cause for hospitalization in patients older than 65 years (1). Clinical depression also is common in HF, with prevalence estimates ranging from 24% to 42% (2). Studies from multiple investigative teams, including our own, have found that, independent of HF disease severity, depressive symptoms are associated with adverse clinical outcomes for HF patients (3- 8). Moreover, depressive symptoms that increase over time also have been associated with worse outcomes in patients with myocardial infarction (MI) (9- 11). Although changes in depressive symptoms have been assessed previously in HF patients (12- 13), their relationship to HF clinical outcomes has not been evaluated. The present study extends our prior report (3) by evaluating the unique contribution of changes in depressive symptoms at 1 year of follow-up in the context of established risk factors, including baseline depressive symptoms and established HF disease severity biomarkers, on subsequent clinical outcomes defined by hospitalization or death across an overall median follow-up period of 5 years.

This was a prospective, observational study of a nonhospitalized cohort of HF patients, recruited from outpatient HF clinics in central North Carolina. We performed medical and psychosocial assessments at baseline and at the 1-year follow-up to evaluate their association with events, including mortality and hospitalizations over a median follow-up period of 5 years from baseline assessments (median of 4 years from the 1-year follow-up assessments with range of 3 to 6 years and no losses to follow-up).

The study sample comprised 147 patients recruited from the Heart Failure Programs at Duke University Medical Center and the University of North Carolina Health Care at Chapel Hill from January 2000 through December 2002. This sample represents a subset of the 204 participants whom we initially recruited (3); 27 patients died within the first year and 30 were alive but were unavailable to complete the depression assessment at the 1-year follow-up. Inclusion criteria were: left ventricular ejection fraction (LVEF) of 40% or less, documented within the last year by angiography, radionuclide ventriculography, or echocardiography; and New York Heart Association functional classes I to IV for at least 3 months in duration. Exclusion criteria were: pacemaker dependence; uncontrolled hypertension; MI, percutaneous coronary intervention, or coronary artery bypass graft within the last 3 months; and a patient's first visit to the heart failure clinic. Pacemaker dependence, fewer than 3 months after MI, percutaneous coronary intervention, and coronary artery bypass graft were excluded because the study also was designed to evaluate autonomic nervous system regulation. The study was approved by the Institutional Review Board at Duke University Medical Center, where all assessments were performed. Written informed consent was obtained from all participants before their participation.

Clinical information and medical history were obtained from medical records. Medications were documented as medications being taken at the time of baseline assessments, including both cardiac and psychotropic medications.

After 20 min of seated relaxation, blood was collected from an antecubital vein in a 5-ml phlebotomy tube containing ethylenediaminetetraacetic acid. Samples were placed on ice and were cold-centrifuged at 1,000 × g for 10 min. The resulting plasma was pipetted into plastic vials and frozen and maintained at −80°C until assay. On the day of NT-proBNP analysis, samples were thawed in a room temperature water bath and then mixed thoroughly. NT-proBNP measurements were performed within 4 hours of thawing using an electrochemiluminescence immunoassay in accordance with the manufacturer's instructions (Elecsys proBNP, Roche Diagnostics Corporation, Indianapolis, Indiana). The analytical measurement range for the assay is 5 to 35,000 pg/ml, and typical day-to-day imprecision has a coefficient of variation of less than 5%.

LVEF was determined by 2-dimensional echocardiography. Apical 4- and 2-chamber images of the heart were acquired by a single sonographer using an Acuson (Mountain View, California) ultrasound machine and were stored as digital loops. The endocardial borders of the left ventricle in the 2 views were traced by a single experienced echocardiography specialist (A.L.H.) using customized off-line software (Access Point 2000, Freeland Systems, LLC, Westfield, Indiana), and ventricular volumes and LVEF were computed using the biapical Simpson's rule.

The Beck Depression Inventory (BDI) is a 21-item self-report measure of depressive symptomatology (14). A score of 10 or more is considered indicative of clinically significant levels of depressive symptoms (15). Although the BDI is not used to diagnose clinical depression, it is a well-validated instrument for assessing the severity of depressive symptoms. Moreover, elevated symptoms of depression measured by the BDI are associated with an increased risk of adverse events in cardiac patients, including patients with HF (3,7,16- 17). These studies also document that the cut point of a BDI score of 10 or more is applicable in cardiac patient populations, where it has been associated with poorer prognosis (3,16- 17).

On the 1-year anniversary (±2 weeks) of participants' baseline assessments, 147 surviving participants repeated the BDI to reassess their depression symptoms. Medical records also were examined to document current health status, hospitalizations during the past year, and antidepressant medication use.

Participants' medical records were reviewed on a yearly basis, over a median of 5 consecutive years from baseline (with a range of 4 to 7 years and no losses to follow-up) on the anniversary of their baseline assessments by trained research assistants blinded to patients' depression status. Patients also were contacted annually by mail and asked to indicate whether they had been hospitalized during the past year and provided consent for retrieval of their hospitalization records. The primary end point was defined as the time to cardiovascular hospitalization or death (whichever occurred first) within the follow-up period. Patient mortality was verified through hospital and emergency medical services records. Cardiovascular hospitalizations included hospitalizations for MI, stroke, treatment for worsening heart failure, cardiac surgery including coronary artery bypass grafts, and heart transplantation. To develop a fuller understanding of the relationship of depression and clinical outcomes, we also considered all-cause hospitalization or mortality as a secondary composite end point.

Cox proportional hazards regression models were used to examine the relationship between baseline characteristics and events (death and hospitalizations) during the median 5-year follow-up (with range of 4 to 7 years and no losses to follow-up). Age, HF cause, NT-proBNP level, LVEF, baseline BDI score, 1-year change in BDI score (BDI score at 1 year minus baseline BDI score), and antidepressant medication use had planned inclusion in the primary model. Hospitalizations occurring during the first year were used as an indicator of medical status for that year and also were included in the planned primary proportional hazards regression model to address the possibility that clinical events during the first year of follow-up could drive changes in depression symptoms during the interval between baseline and the reassessment 1 year later. To minimize the possible influence of clinical events during the first year on subsequent 1-year BDI scores, outcomes were considered as clinical events occurring at least 1 year after initial baseline assessments. Thus, outcomes were defined as death or hospitalizations occurring only after the first year of follow-up data had been collected (including 1-year BDI score) for each participant. Secondary sensitivity analyses for the evaluation of robustness of findings from the planned model allowed other potential explanatory factors (including New York Heart Association functional class; diabetes, hypertension; hypercholesterolemia; myocardial infarction; tobacco use; glomerular filtration rate [modified Cockcroft-Gault equation]; implantable defibrillator; and treatment with a beta-blocker, diuretic, angiotensin converting enzyme inhibitor, nitrate, warfarin, or statin) to be available for entry into the planned model by stepwise selection (significance level entry/exit = 0.1). For Cox regression analyses, baseline NT-proBNP and BDI scores were trimmed at the ninety-fifth percentiles to prevent excessive influence of outliers; NT-proBNP was expressed as NT-proBNP/1,000 and age was expressed as age/10. Descriptive statistics were means with standard deviations for continuous variables and counts with percentages for categorical variables. Also, Student t tests and chi-square tests were used for the comparison of characteristics of study participants who died before or were otherwise unavailable for reassessment of depressive symptoms, with those comprising the study sample (n = 147) evaluated using the Cox regression analyses.

Demographic and clinical characteristics of the study sample are shown in (Table 1). The mean age at baseline was 57 years, with a range of 27 to 88 years. Most patients were male (70%), and approximately half were minorities. Most patients had New York Heart Association functional class II (59%) or class III (37%) levels, and most patients were taking a beta-blocker (88%) or an angiotensin converting enzyme inhibitor (86%). Twenty percent of patients were taking antidepressant medication (selective serotonin reuptake inhibitors, tricyclic or tetracyclic agents, and monoamine-oxidase inhibitors) at baseline, and the average BDI score at baseline was 10 (SD: 6, range: 0 to 37). The study sample consisted of 147 patients, which represents 83% of the 177 patients who survived 1 year after baseline assessments from our original study (3). Compared with the survivors, patients who died during the first year of follow-up (n = 27) had, at baseline, higher resting heart rates (p = 0.021), lower LVEFs (p = 0.028), higher NT-proBNP levels (p = 0.0001), and were less likely to be taking nitrates (p = 0.041). Cox proportional-hazards regression analyses revealed that only NT-proBNP and LVEF were related to death (all p < 0.01) within the first year after baseline assessments. There were few differences in any of the demographic or clinical characteristics of the 30 patients who survived at least 1 year, but who were unavailable for follow-up compared with the remaining 147 patients, with the exceptions of their having higher baseline systolic blood pressure (p = 0.011), higher NT-proBNP levels (p = 0.021), lower hemoglobin levels (p = 0.015), and being more likely to be diabetic (p = 0.024). It is of note that there were no differences in baseline BDI score between the present study sample (n = 147) and those 30 patients who were unavailable for 1-year follow-up (10.2 ± 6.9 vs. 10.9 ± 6.4, respectively, p = 0.579).

The median follow-up period from baseline was 5 years, with a range of 4 to 7 years; no patients were lost to follow-up. Of the 147 patients who survived at least 1 year and completed a second BDI assessment, 127 (86%) either died or were hospitalized at least once, including 112 (76%) such events the result of cardiovascular events during the subsequent follow-up period (with a median of 4 years). During the follow-up period, 53 (36%) of the 147 participants died, of which 40 deaths were the result of cardiac causes; 15 of these 53 deaths occurred before any hospitalizations during the follow-up period. The planned multivariate Cox proportional-hazards regression analysis revealed that ischemic cause (hazard ratio [HR]: 1.83, 95% confidence interval [CI]: 1.22 to 2.75, p = 0.004), NT-proBNP (1,000 pg/ml HR: 1.17, 95% CI: 1.01 to 1.36, p = 0.03), LVEF (HR: 0.97, 95% CI: 0.95 to 0.99, p = 0.011), and hospitalization within the first year after baseline assessments (HR: 2.4, 95% CI: 1.57 to 3.66, p < 0.001) were associated with cardiovascular hospitalization or mortality (Table 2). In this same model, the extent of baseline depression symptoms also was strongly related to risk for cardiovascular hospitalization or mortality (1-point BDI HR: 1.1, 95% CI: 1.06 to 1.14, p < 0.001), and so was change in BDI score at 1 year (1-point BDI change HR: 1.07, 95% CI: 1.02 to 1.12, p = 0.007). In secondary analyses, of the variables eligible for contributing further to the planned model through stepwise selection, current smoking status (HR: 1.92, 95% CI: 1.14 to 3.23, p = 0.014) and presence of an implantable defibrillator (HR: 0.51, 95% CI: 0.28 to 0.92, p = 0.024) were added to the model. However, the addition of these factors did not influentially alter the results for the explanatory values of the variables in the original model, including baseline depression symptoms and 1-year change in depression symptoms, confirming that the original planned model was robust for its results.

The change in BDI score from baseline to 1-year follow-up ranged from a 14-point reduction to an increase of 19 points. To provide a better understanding of how changes in depressive symptoms were related to change in risk of adverse clinical outcomes, we created 3 descriptive categories. Ideally, each of the categories would have approximately one third of the participants, and 25% at a minimum; moreover, one would like the middle category for a variable that is reflective of change to represent little or no change. Sixty-five participants (44%) showed a 2-point change or less in either direction in BDI score from baseline to the 1-year follow-up, whereas 29% showed either an increase in BDI of 3 points or more (n = 43), and 27% showed a decrease of at least 3 points (n = 39). In exploratory analyses, these BDI-change categories were entered into our planned regression model in place of the raw BDI-change score to assess the nature of changing BDI effects on outcomes. Compared with patients showing minimal (≤2-point) change in BDI, those exhibiting a 3-point increase or more, indicating worsening depression symptoms, were at more than twice the risk of adverse outcomes (HR: 2.12, 95% CI: 1.31 to 3.43, p = 0.002); however, patients whose BDI scores were 3 points or more lower at the 1-year follow-up were at a relatively similar risk compared with those exhibiting a minimal BDI change (HR: 0.87, 95% CI: 0.5 to 1.48, p = 0.59).

In our planned model, age (10-year HR: 1.25, 95% CI: 1.04 to 1.51, p = 0.019) and hospitalization within the first year after baseline assessments (HR: 2.36, 95% CI: 1.59 to 3.51, p < 0.001) were associated with all-cause hospitalization or mortality over the subsequent follow-up period (Table 2). The extent of baseline depression symptoms also was related to increased risk for all-cause hospitalizations or mortality (1-point BDI HR: 1.09, 95% CI: 1.05 to 1.13, p < 0.001), and so was change in BDI score at 1 year (1-point BDI change HR: 1.06, 95% CI: 1.01 to 1.11, p = 0.023). Secondary analyses showed that no additional variables were added to the planned model through stepwise selection, confirming that the planned model was robust.

In exploratory regression analyses, we substituted our categorical variables indicating worsening, improving, or minimal change in BDI score for the continuous measure of 1-year change in BDI score. For all-cause hospitalizations, there was a trend for increased risk associated with worsening depression symptoms compared with minimal or no change in depression symptoms (HR: 1.48, 95% CI: 0.96 to 2.29, p = 0.077), whereas improving depression symptoms were not associated with altered risk compared with minimal or no change (HR: 0.78, 95% CI: 0.47 to 1.28, p = 0.32).

The present findings confirm and extend our previous observations by showing that symptoms of depression are associated with adverse clinical outcomes, including cardiovascular (or all-cause) hospitalization and death, over a median of 5 years of follow-up from baseline (with range of 4 to 7 years and no losses to follow-up), even after controlling for the severity of HF disease and other established risk factors, and that worsening of depression additionally was associated with such adverse events compared with patients whose depressive symptoms remained relatively stable over the initial 1-year follow-up period. To our knowledge, this is the first study to evaluate prospectively the clinical impact of changing depression symptoms in HF patients, and our observations demonstrate that such changes do indeed affect clinical outcomes. Specifically, a 1-point change in BDI score was associated with a 7% alteration in risk (per unit of time) for cardiovascular hospitalizations and mortality. Importantly, the effects of change in depression symptoms were independent of baseline depression symptoms and HF disease severity biomarkers, as well as hospitalizations occurring during the year over which depression symptoms were monitored. Exploratory analyses designed to understand better the latter finding suggest that worsening symptoms of depression in HF patients may be of particular concern; an increase in depressive symptoms of 3 points or more in BDI score over a 1-year follow-up period was associated with a more than 2-fold increased risk of cardiovascular hospitalization or death compared with minimal change (<2 BDI points). Our findings are consistent with observations in MI patients, in whom worsening symptoms of depression have been associated with greater risk of subsequent mortality (9- 11).

Our study found no association between antidepressant use and clinical outcomes. However, because the study was not designed to assess the impact of antidepressant treatment on outcomes, the significance of this observation is unclear. There are currently no published data from randomized clinical trials of HF patients indicating whether interventions designed to modify depression symptoms may alter long-term clinical outcomes. Although the SADHART (Sertraline Against Depression and Heart Disease in Chronic Heart Failure) trial (18) was designed to be a safety and efficacy trial and was not powered to evaluate adverse clinical outcomes, the incidence of cardiovascular events in a subgroup of more severely depressed patients with acute coronary syndromes randomized to sertraline treatment was 14.5%, compared with 22.4% among those receiving a placebo. In the Enhancing Recovery in Coronary Heart Disease trial, the rates of nonfatal MI and all-cause mortality were similar between post-MI patients randomized to cognitive behavior therapy and education controls (19), although a post hoc analysis revealed that patients who received antidepressant medication had improved survival compared with similar patients not receiving antidepressants (20). Moreover, patients randomized to the cognitive behavior therapy condition who exhibited an increase in depressive symptoms had higher event rates compared with those patients who showed an improvement in depressive symptoms (9). In a recently completed randomized clinical trial in HF patients with depression (21), no antidepressant benefit of sertraline compared with placebo was observed, although remitting depression symptoms were found to be associated with improved health status and quality of life (22).

It is important to note that the observational data reported in the present study do not lend themselves to causal inferences, nor to insights into the potential effects of interventions designed to modify symptoms of depression. Other limitations of the present study include absence of changes in depression during the follow-up period. However, changes in depression symptoms during the follow-up period would be difficult to interpret because such changes would be concomitant with changing HF disease severity. The reported analyses evaluate the role of change in depressive symptoms before the onset of the follow-up period and therefore are explanatory. The sample size of the present study also was relatively small, but a significant relationship between 1-year change in depression symptoms and subsequent clinical outcomes was observed despite this limitation. It also should be noted that the HF disease severity biomarkers (NT-proBNP and LVEF) were assessed only at baseline, not at the time of the reassessment of depressive symptoms 1 year later; however, our inclusion of hospitalizations during that 1-year period was designed to minimize this potential limitation.

In summary, elevated depression symptoms and worsening depression symptoms were found to be explanatory risk factors for adverse clinical outcomes in HF patients, independent of HF disease severity. Our findings support the recent American Heart Association's position encouraging depression screening (23) and further suggest that it may be prudent for clinicians to reassess symptoms of depression routinely in HF patients to determine better appropriate medical management of these patients who are at increased risk for adverse clinical outcomes and impaired quality of life.