Introduction

Heart failure (HF) is defined as a clinical syndrome in which patients have typical symptoms such as breathlessness, ankle swelling, and fatigue and signs such as elevated jugular venous pressure, pulmonary crackles, and displaced apex beat, resulting from an abnormality of cardiac structure or function [1]. Approximately 1–2 % of the adult population in developed countries has HF, with the prevalence rising to ≥10 % among persons 70 years of age or older [2]. HF is one of the most common causes of hospital readmission and mortality.

Psychological factors such as depression or anxiety are often reported with high prevalence and strong association with negative outcomes in patients with cardiovascular disease [3]. Many studies have reported high rates of depression among HF patients. A prior systematic review and meta-analysis published by Rutledge in 2006 [4] reported an overall aggregated depression prevalence rate of 21.6 % among HF patients, while individual study prevalence estimates ranged from 9 to 60 %. Moreover, in 2005 Konstam [5] reported that approximately 40 % of HF patients may suffer from major anxiety, and overall anxiety levels are 60 % higher than levels seen in the healthy population.

Depression has been linked to increased risk of negative outcomes, such as rehospitalization and mortality among HF patients. According to a previous meta-analysis, the aggregated risk estimate derived from 8 studies suggested a >2-fold risk of death and secondary events for HF patients with heightened depressive symptoms or a depressive disorder [4]. A similar analysis was also published by Fan [6] in 2014 on 9 prospective studies, who reported a pooled Hazard Ratio of 1.51 for patients with depression compared to patients without depression. In both cases the result was strongly heterogeneous but no further analysis, such as meta-regression, was performed to examine the sources of this heterogeneity. On the other hand, there is, to the best of our knowledge, no meta-analysis published about the prevalence of anxiety among HF patients and the effect of anxiety on mortality outcome. Even though anxiety is usually correlated with depression, it has not extensively been studied among patients with HF.

Our aim is to provide an updated systematic review of prospective or retrospective studies and a meta-analysis of the effect of depression and the effect of anxiety on mortality among HF patients. To reach this objective, we searched extensively for available studies investigating the impact of depression and anxiety on mortality of HF patients. Within these studies, we identified also the reported prevalence of depression or anxiety among HF patients.

Methods

Search strategy and selection criteria

This systematic review and meta-analysis were conducted according to the guidelines introduced in the Preferred Reporting Items for Systematic reviews and Meta-analysis (the PRISMA Statement) [7]. The 27 checklist items of the PRISMA methodology followed are given in Appendix 1. Three electronic databases (MEDLINE, BIOSIS and EMBASE) were searched for studies that investigated the relationship between depression or anxiety and mortality among HF patients. No publication time restriction was applied. All papers written in English and published before the 25th of June 2015 were included. Selected journals as well as the references of full text papers were also hand-searched, when necessary, in order to identify studies that meet the inclusion criteria.

The database search string was created according to the PICO model (P population/patient, I intervention/indicator, C comparator/control, O outcome). For the “P” in PICO the “HEART FAILURE” keyword was included. For the “I”, the following keywords: “DEPRESS? OR STRESS OR ANXIETY OR PSYCHOLOG?”. For the “C”, no particular terms were used in our case. For “O”, we used the following keywords: “MORTALITY OR DEATH”. The complete query as used for the databases search is given in Appendix 2.

Study selection

In our analysis, several inclusion and exclusion criteria were defined. All studies that met those criteria were included. The inclusion criteria were articles presenting studies focusing on the association between depression or anxiety and mortality in a HF adult population. All mortality outcomes such as all-cause or cardiac-related mortality were included and studies focusing on inpatient, outpatient or both care settings were taken into account. On the other hand, publications analyzing data that had already been used before for the same purpose, studies introducing no quantitative assessment of the impact of depression or anxiety on the outcome or analyzing the use of antidepressants as primary focus were excluded from our analysis.

Review process and data collection

All titles and abstracts of studies identified by the electronic and hand search were screened by the reviewer (IS) to identify those meeting the inclusion/exclusion criteria. Then, all the selected full texts were screened independently by two reviewers (IS, GJdV) to identify which articles should be included in the systematic review. Any disagreement between the reviewers was resolved by a third reviewer (SP). For each of the selected articles, the reviewers extracted data about author, year of publication, follow-up period, outcome variable, location, study design, study population (size/type), prevalence of depression or anxiety, assessment method of the psychological parameter, other parameters, statistical method and results.

Mendeley 1.13.8 software was used for organizing and managing of the articles.

Data analysis

All studies were categorized according to the psychological factor investigated (depression or anxiety). Information was extracted according to whether the analysis was adjusted for confounders such as age, gender, and clinical severity. For both groups the association between depression or anxiety and mortality was reported by collecting information of the hazard ratios/odds ratios, 95 %CI and/or p values.

Random-effects meta-analysis was applied to combine the results. We decided to pool not only the adjusted effect but also the unadjusted effects in order to avoid the bias of the different adjustments. For the few cases where Odds Ratios were reported, they were converted [8] into Hazard Ratios in order to be comparable with the other Hazard Ratios. In studies where results were presented for several periods of follow-up we selected the longest follow-up period to avoid bias of including multiple results on the same patient data.

Studies collected in our analysis were different with respect to patient population, locations and depression or anxiety assessment methods. The random-effects method allows for heterogeneity by assuming that the effects being estimated in the different studies are not identical, but follow a normal distribution. Heterogeneity across the studies was quantified by the I 2 statistic [9]. The I 2 statistic summarizes the fraction of the variation across studies due to heterogeneity relative to chance. Random-effect meta-regression was used in an attempt to explain between-study heterogeneity and identify possible sources of bias. Meta-regression is a method to quantify the association between the estimated effect of depression and different study characteristics.

Meta analyses were presented in the form of forest plots created with the meta for package for R statistics version 3.0.3 (The R Foundation for Statistical Computing).

Results

Search result

A total of 906 potentially relevant articles was identified from the electronic search and five from the hand search. After removing the duplicates and reviewing the titles and abstracts, we ended up with 62 articles for a full text review. From these, 35 more articles were excluded, leaving 27 articles for the systematic review (Fig. 1).

Fig. 1
figure 1

PRISMA flow diagram

Full size image

Characteristics of the selected studies

Depression and mortality

Among the identified studies, 26 reported on the effect of depression. The prevalence of depression varied from 10 to 79 % in the identified literature studies (Table 1). The unadjusted effect of depression is presented in Table 1a, while the effect of depression after adjusting for several confounders in Table 1b. The most common confounders, used in more than 10 studies, were age, gender, NYHA class and (left ventricular) ejection fraction.

Table 1 (a) Unadjusted/(b) adjusted effect of depression on mortality among HF patients
Full size table

There were various techniques used among the studies to assess depression levels. We included all studies assessing for clinically significant depression. The most common scale used was the Beck Depression Inventory (BDI) [10], followed by the Patient Health Questionnaire (PHQ) [11].

The pooled hazard ratio for the unadjusted effect of depression on mortality was strongly significant across 15 studies (HR = 1.57; 95 %CI 1.30–1.89; p < 0.001). The pooled estimation was strongly heterogeneous as reflected by the I 2 statistic (I 2 = 94 %, heterogeneity p < 0.001). The pooled adjusted Hazard Ratio was also significant (HR = 1.40; 95 %CI 1.22–1.60; p < 0.001) and again heterogeneous (heterogeneity p < 0.001; I 2 = 97 %; Fig. 2).

Fig. 2
figure 2

Meta-analysis—forest plot calculating the effect of depression a unadjusted effect, b adjusted effect

Full size image

A random-effect meta-regression was performed to understand the sources of the higher than 90 % observed heterogeneity between the studies. The potential study-level covariates analyzed were the study characteristics introduced in Table 1. There was no association found between heterogeneity and the depression assessment method, the adjusted or univariate analysis, the location where the study was conducted, the inpatient or outpatient predictive period, the year of the study, the type of the study and the follow-up period. On the other hand, significant heterogeneity was associated with the total population size (smaller effect in larger studies p < 0.01) and the prevalence of the depression in the study (smaller effect for prevalence >29 %; p < 0.01; Table 2).

Table 2 Random-effect meta-regression
Full size table

Anxiety and mortality

Only six studies analyzing the effect of anxiety on mortality among HF patients were identified with a prevalence of anxiety varying from 9 to 53 % (Table 3). Table 3a shows the unadjusted effects reported in the studies, and Table 3b the reported effects on mortality after adjusting for a group of confounders. Age, NYHA class and (left ventricular) ejection fraction were the most common confounders in the identified studies.

Table 3 (a) Unadjusted/(b) adjusted effect of anxiety on mortality among HF patients
Full size table

There was no evidence found for anxiety as an independent predictor of mortality. The pooled hazard ratio for the unadjusted effect of anxiety on mortality, which was based on 2 studies, was 1.02 (95 % CI 1.00–1.04; p = 0.24, heterogeneity p = 0.38; I 2 = 0 %). The pooled hazard ratio for the adjusted effect of anxiety on mortality could be based on 5 studies and was identical (HR = 1.02; 95 % CI 1.00–1.04; p = 0.09) and reasonably homogenous (heterogeneity p = 0.97; I 2 = 0 %, Fig. 3).

Fig. 3
figure 3

Meta-analysis—forest plot calculating the effect of anxiety a unadjusted effect, b adjusted effect

Full size image

Discussion

This systematic review was conducted according to the PRISMA guidelines to assess the evidence on the effect of depression (26 studies) and anxiety (6 studies) on all-cause mortality outcome among heart failure (HF) patients. <Key results: 1.6 for depression but very heterogeneous across studies; no effect for anxiety>.

In contrast to other reviews, our study was not limited on follow-up duration or only in prospective studies reporting adjusted effects of the two parameters. We reviewed all studies published quantifying the effect of depression or anxiety.

The prevalence of depression varied among the 26 different studies with an average of approximately 29 % ranging from 10 to 79 %. The meta-analysis showed that the unadjusted risk of death among HF patients facing depression was 1.57 times higher than the risk among HF patients without depression and the pooled estimate of the adjusted Hazard Ratio was 1.40. In both univariate and adjusted analysis, strong heterogeneity among the studies was found. Our findings are more conservative than previous reviews published [4, 6]. Rutledge et al. reported a 2.10 higher adjusted risk of mortality and secondary events based on 8 studies, and Fun et al. reported a pooled adjusted Hazard Ratio of 1.51 based on 9 studies, both with substantial heterogeneity. From our attempt to explain heterogeneity, we found that the effect of depression is weaker in larger studies; this suggests publication bias: small studies were published if they found relatively large effect estimates, while small studies with modest effect estimates were not. The weaker effect in studies with higher prevalence of depression may relate to the use of different cut-offs on an underlying, latent, scale for depression. If a more liberal cut-off was used, those labeled as depressed actually were milder than with a more strict definition of depression.

Our results for anxiety do not have the same weight as the results with respect to depression since anxiety was less studied in the literature. Anxiety had a similar prevalence to depression among the six identified studies (average 29 %, range 9–45 %), but patients with anxiety had no increased risk of death compared to those without anxiety. However, since anxiety is usually correlated with other factors such as depression, further research of anxiety as a covariate to other factors is recommended.

One limitation of our study is related to the variation in follow-up times. Follow-up times varied from 30 days to a number of years; furthermore, there were studies covering different follow-up periods but in these cases we always selected the longest follow-up. Further analysis such as subgroup analysis would be recommended to investigate the effect variation in different follow-up periods; however, limited information in some of the literature publications is restrictive toward this direction.

Moreover, we focused only on mortality. Nevertheless, there is evidence that depression and anxiety are also associated with other adverse events such as readmission. Further investigation is needed also toward this direction.

One limitation of the meta-regression is that even though we tried to cover a broad selection of study-level covariates, there are more that might also be related to the heterogeneity. Further research on different factors’ interactions would be recommended.

The “gold standard” test of causality of a putative risk factor is a randomized clinical trial. Such a trial minimizes concerns about confounders [12–14]. To the best of our knowledge, there is no randomized clinical trial conducted for depression among a HF population. Based on our findings, we strongly recommend such a trial in order to evaluate the causality of depression.

Finally, according to our findings from the meta-regression, depression should not be underestimated in clinical practice within HF population groups where prevalence is low. Furthermore, based on our overall findings on the effect of depression, we recommend further research on the recognition and management of depression in clinical practice which might improve patient outcomes. Further analysis such as subgroup analysis and interventional studies is required for stronger evidence toward this direction.