The role of a palliative care intervention in moderating the relationship between depression and survival among individuals with advanced cancer

Abstract

Objective

Randomized controlled trials (RCTs) of early palliative care interventions in advanced cancer have positively impacted patient survival, yet the mechanisms remain unknown. This secondary analysis of two RCTs assessed whether an early palliative care intervention moderates the relationship between depressive symptoms and survival.

Methods

The relationships among mood, survival, and early palliative care intervention were studied among 529 advanced cancer patients who participated in two RCTs. The first (N=322) compared intervention vs. usual care. The second (N=207) compared early vs. delayed intervention (12 weeks after enrollment). The interventions included an in-person consultation, weekly nurse coach-facilitated phone sessions, and monthly follow-up. Mood was measured using the Center for Epidemiologic Studies Depression Scale (CES-D). Cox proportional hazard analyses were used to examine the effects of baseline CES-D scores, the intervention, and their interaction on mortality risk while controlling for demographic variables, cancer site, and illness severity.

Results

The combined sample was 56% male (mean = 64.7 years). Higher baseline CES-D scores were significantly associated with greater mortality risk (HR = 1.042, 95% CI 1.017 to 1.067, p = .001). However, participants with higher CES-D scores who received the intervention had a lower mortality risk (HR = .963, CI 0.933–0.993, p = .018) even when controlling for demographics, cancer site, and illness-related variables.

Conclusions

This study is the first to demonstrate that patients with advanced cancer who also have depressive symptoms benefit the most from early palliative care. Future research should be devoted to exploring the mechanisms responsible for these relationships.

The relationship between depression and survival has been studied extensively. A meta-analysis of 293 studies totaling over 1.8 million participants from both patient and community samples concluded that depression was associated with shorter survival (Cuijpers et al., 2014). Research focusing specifically on individuals with cancer has shown that depression measured after cancer diagnosis predicts shorter survival (Lloyd-Williams, Shiels, Taylor, & Dennis, 2009; Pirl et al., 2008, 2012) and that improvement in depression predicts longer survival (Giese-Davis et al., 2011). Though some studies of cancer patients have failed to find a depression-survival link (e.g., Bredal, Sandvik, Karesen, & Ekeberg, 2011), two recent meta-analyses concluded that depression is reliably associated with shorter survival among individuals with cancer (Pinquart & Duberstein, 2010; Satin, Linden, & Phillips, 2009).

Because of its association with survival in patients with advanced cancer and its negative impact on quality of life, depression is a focus of much research in palliative care. Studies have shown palliative care services to be associated both with fewer depressive symptoms (Bakitas et al., 2009; Giese-Davis et al., 2011; Temel et al., 2010) and longer survival (Bakitas et al., 2015; Temel et al., 2010). Given the evidence outlined above that depression is related to higher mortality, one might assume that improvement in depression mediates the influence of palliative care interventions on survival. Although the evidence described above (that palliative care reduces depression and that lower depression is associated with longer survival) is suggestive of mediation, dedicated analyses conducted within a single study are necessary to confirm such a relationship (Baron & Kenny, 1986). However, two groups that tested for mediation in this way did not find evidence of such an association (Giese-Davis et al., 2011; Pirl et al., 2012).

An alternative way of conceptualizing the relationship between depression, palliative care, and survival is that a palliative care intervention may moderate the relationship between depression and survival. For example, the intervention may be particularly beneficial among depressed individuals by focusing attention on factors that promote survival and are often ignored when depressed. According to such a moderation account, depressed individuals would show lower survival rates, the intervention would improve survival, and the impact of the intervention would be greater among depressed individuals. To our knowledge no one has yet tested for such a moderating effect by including baseline depression, palliative care intervention, and their interaction as predictors of survival in a single model, and the present work aims to address this gap.

Current Analysis

The current analysis tested whether the association between depression and mortality risk would be moderated by a palliative care intervention designed to enhance quality of life. Using data from two RCTs of a palliative care intervention, these analyses focused on the relationship between baseline depression scores, intervention status (if and when patients received the intervention), and survival among patients with advanced cancer.

It was hypothesized that receiving a palliative care intervention designed to improve quality of life would moderate the relationship between baseline depression and survival, and that this effect would remain after controlling for demographic variables, cancer site, and illness severity. Specifically, it was predicted that receiving the intervention would disrupt the relationship between depression and survival by increasing survival probability primarily among individuals with higher baseline depression scores.

Method

Study Design

This work uses a combined data set from the two ENABLE (Educate, Nurture, Advise, Before Life Ends) RCTs. ENABLE, which introduces palliative care principles early in the cancer trajectory, was designed to improve quality of life among patients newly diagnosed with an advanced cancer. The primary results of both RCTs have been published previously (Bakitas et al., 2009, 2015), however the analyses reported here test novel hypotheses not addressed in previous publications. In the first study (Study 1; enrollment between November 2003 and May 2007; Bakitas et al., 2009) participants were randomly assigned upon enrollment to either (a) an intervention condition or (b) a usual cancer care control condition. In the second study (Study 2; enrollment between October 2009 and March 2013; Bakitas et al., 2015) a wait-list control design was employed wherein all participants received the intervention but were randomly assigned to receive it either (a) early (upon enrollment) or (b) after a delay of 12 weeks. Target sample sizes of 400 (Study 1) and 360 (Study 2) were chosen to provide 80% power to identify key differences in outcomes of interest in the original studies (quality of life, symptom intensity, mood; See Bakitas et al., 2009). Due to slower accrual than anticipated the final enrollment totals were 322 (Study 1) and 207 (Study 2). The protocols and data and safety monitoring board plans for both studies were approved by the institutional review board of Dartmouth College and the Research and Development Committee of the Veterans Administration Medical Center, White River Junction, Vermont. The trials were registered in clinicaltrials.gov (Identifier Study 1: NCT00253383; Study 2: NCT01245621).

In both studies, the primary dependent variables were patient-reported depression, quality of life, symptom intensity, and resource use. In Study 1, participant survival was examined in a post-hoc analysis. In Study 2, survival was examined as an a priori primary outcome. In the current analyses, we focus on the effects of the intervention and depression on survival. Specifically, we examine the association between participants’ baseline depression scores and subsequent survival time, as well as the moderating effect of receiving the palliative care intervention on this association.

Participants

For both studies, participants were recruited from the Norris Cotton Cancer Center at Dartmouth-Hitchcock Medical Center within 60 days of diagnosis with advanced cancer. Eligibility criteria included diagnosis of a new advanced solid tumor or hematological malignancy, recurrence, or new disease progression following stable disease, and an oncologist-predicted prognosis of approximately 6–24 months. In addition, all participants had to be English-speaking and over age 18. Individuals were excluded if they had impaired cognition, or a severe psychiatric (e.g. schizophrenia, bipolar disorder) or active substance use disorder. After enrollment patients were randomly assigned to receive the intervention or usual care (Study 1) or early or delayed intervention (Study 2). Full details of the two studies are described elsewhere (Bakitas et al., 2009, 2015).

The combined sample for this analysis included 529 patients, of whom 161 received usual care (Study 1 only), 265 received the early intervention (161 from Study 1; 104 from Study 2), and 103 received the delayed intervention (Study 2 only).

Intervention

The ENABLE intervention has been described in detail elsewhere (Bakitas et al., 2009; Bakitas et al., 2015) and consists of a psychoeducational approach to encourage patient activation and enhanced quality of life. In brief, after an initial in-person palliative care consult, advanced practice nurses specializing in palliative care facilitated 4 (Study 1) or 6 (Study 2) semi-structured psychoeducational telephone coaching sessions with patients using an author-developed informational guidebook called Charting Your Course, followed by monthly check-in calls until the patient died or the study ended. Topics discussed during sessions included problem-solving, coping, self-care, symptom management, building a support system, communication skills, decision-making, advance care planning, and life review. It is relevant to note that treatment of depression was not included as a specific goal of the intervention.

Data Collection and Instruments

The primary variables of interest in these analyses were baseline depression and survival. The Center for Epidemiological Studies Depression Scale (CES-D; (Radloff, 1977) is a 20-item measure that yields scores from 0–60 (α = .86), where higher values represent greater severity of depressive symptoms and a score ≥16 generally indicates a clinically significant level of depressed mood (Okun, Stein, Bauman, & Silver, 1996). Survival time was calculated from the date of enrollment to the date of death; patients who were still alive at study end (Study 1: May 1, 2008; Study 2: September 5, 2013) were censored on that date.

Given the correlational nature of the primary independent variables (individual differences in baseline depression and an intervention that spanned two RCTs), results were assessed in the context of a variety of statistical covariates. Specifically, demographic information collected at baseline included age, gender, ethnicity/race, marital status, employment status, level of education, and rural/urban residence. The site of cancer diagnosis was identified from the electronic health record. Variables representing illness severity characteristics included resource use 3 months prior to study enrollment (days spent in the hospital and intensive care unit [ICU] and emergency room [ER] visits), use of chemotherapy and radiation, and the presence of advance directives and do not resuscitate orders. All analyses were conducted using SPSS software, version 22.

Results

Descriptive Statistics and Comparisons Between Study Samples

Descriptive statistics calculated for baseline depression, demographics, cancer site, and illness severity characteristics are reported in Table 1. For all continuous variables (CES-D score, days in hospital, days in ICU, ER visits, and age), the mean and standard deviation were calculated. The remaining variables were categorical: advanced directives (ADs; yes/no), do not resuscitate order (DNR; yes/no), undergoing chemotherapy at enrollment (yes/no), undergoing radiation at enrollment (yes/no), gender, residence (rural vs. urban), education (college graduate vs. not), currently married (yes/no), employment (employed full or part time vs. not currently employed), and race (white vs. other). Cancer site was dummy coded and each diagnosis site (lung, gastrointestinal, genitourinary, breast, hematological, and other) treated as a separate variable. For all dichotomous variables a count and percentage of the total were calculated.

Table 1.

Baseline characteristics of combined sample and comparisons between Study 1 and Study 2

	Combined Sample		Comparison between study groups
			Study 1 (N = 322)		Study 2 (N = 207)
	N	Mean (SD) or No. (%)	N	Mean (SD) or No. (%)	N	Mean (SD) or No. (%)	p *
Demographic variables
Age	529	64.7 (10.7)	322	65.0 (11.2)	207	64.3 (9.9)	.453
Gender (male)	529	296 (56%)	322	187 (58.1%)	207	109 (52.7%)	.221
Rural residence	529	306 (57.8%)	322	181 (56.2%)	207	125 (60.4%)	.343
College graduate	482	166 (34.4%)	275	81 (29.5%)	207	85 (41.1%)	.008
Married	526	356 (67.7%)	319	221 (69.3%)	207	135 (65.2%)	.331
Employed	525	112 (21.3%)	318	63 (19.8%)	207	49 (23.7%)	.291
White	485	475 (97.9%)	279	275 (98.6%)	206	200 (97.1%)	.257
Cancer site
Lung	529	205 (38.8%)	322	117 (36.3%)	207	88 (42.5%)	.155
Gastrointestinal	529	183 (34.6%)	322	133 (41.3%)	207	50 (24.2%)	<.001
Genitourinary	529	55 (10.4%)	322	39 (12.1%)	207	16 (7.7%)	.107
Breast	529	56 (10.6%)	322	33 (10.2%)	207	23 (11.1%)	.753
Hematological	529	10 (1.9%)	322	0 (0%)	207	10 (4.8%)	<.001
Other cancer	529	20 (3.8%)	322	0 (0%)	207	20 (9.7%)	<.001
Illness-related variables
Days in hospital	528	2.79 (5.2)	322	2.95 (5.2)	206	2.54 (5.2)	.381
Days in ICU	404	.18 (.92)	322	.03 (.25)	82	.74 (1.9)	.001
ER visits	529	.43 (.84)	322	.34 (.78)	207	.57 (.90)	.003
Advanced directives	527	238 (45.2%)	322	149 (46.3%)	205	89 (43.4%)	.520
Do not resuscitate	515	43 (8.3%)	322	23 (7.1%)	193	20 (10.4%)	.201
Chemotherapy	529	427 (80.7%)	322	271 (84.2%)	207	156 (75.4%)	.012
Radiation	529	81 (15.3%)	322	41 (12.7%)	207	40 (19.3%)	.040
Depression
Baseline CES-D	471	13.5 (9.3)	268	12.9 (8.7)	203	14.2 (10.1)	.132
CES-D ≥ 16 **	471	167 (35.5%)	268	89 (33.2%)	203	78 (38.4%)	.241

Abbreviations: SD = Standard Deviation; % = Percent of total patients; ICU = Intensive Care Unit; ER = Emergency Room; CES-D = Center for Epidemiologic Studies Depression Scale.

^*Independent samples t-tests were used for continuous variables (Age, Days in hospital, Days in ICU, ER visits, and Baseline CES-D); Chi-Squared analyses were used for categorical variables.

^**Number of participants with CES-D ≥16 at baseline, indicative of clinical levels of depression. This dichotomous variable is reported for descriptive statistics and group comparisons only; the continuous Baseline CES-D score was used for all other analyses.

The two study samples were compared on baseline depression and all statistical covariates described above using independent-samples t-tests for continuous variables and chi-square tests for categorical variables. Of the 21 variables examined, 8 differed between studies: days in ICU, ER visits, chemotherapy, radiation, gastrointestinal cancer, hematological cancer, other cancer, and college graduate. See Table 1 for results of all comparisons.

Association of Baseline Variables with Depression

Pearson’s bivariate correlations between baseline CES-D scores and all other continuous covariates were calculated; point-biserial correlations were used for dichotomous covariates. Lung cancer was associated with relatively higher depression (r = .094, p = .041, N = 471) and older age was associated with lower depression (r = −.108, p = .019, N = 471). No other associations achieved significance (See Table S1 in Supplementary Material available online).

Survival Analyses

Cox proportional-hazards regression analyses (Cox, 1972) were used to model the effects of intervention status and baseline depression on survival, with and without adjustment for baseline covariates. The design involved survival over time among a control group that never received the intervention (usual care), an experimental group that received the intervention upon enrollment (early intervention), and an experimental group that received the intervention 12 weeks after enrollment (delayed intervention). Because the intervention was the same for the latter two groups and simply involved implementation at different points in time, the data were handled using a survival analysis with intervention as a time-varying covariate (see Cox, 1972). Patients with missing covariate data were excluded as needed from each Cox model.

To test whether the intervention moderated the effect of baseline depression on survival, we conducted a Cox analysis that included intervention status, baseline CES-D, and their interaction as simultaneous predictors. The Cox analysis confirmed an effect of depression (Wald = 12.377, HR = 1.038, CI: 1.017–1.060, p < .001) such that higher depression at baseline was associated with shorter survival. In addition, there was a significant interaction between depression and the intervention (Wald = 4.451, HR = .973, CI: 0.949–0.998, p = .035), such that depression was more strongly associated with shorter survival among participants receiving usual cancer care than among those receiving the intervention.

The Cox analysis was repeated including only the covariates that correlated significantly with depression (lung cancer and age) as a test of whether these variables, not depression, could be responsible for the relationships of interest. In this analysis, depression was once again associated with shorter survival (Wald = 12.645, HR = 1.039, CI: 1.017–1.061, p < .001) and the depression by intervention interaction remained significant (Wald = 4.431, HR = .973, CI: 0.949–0.998, p = .035).

Finally, to provide an even more conservative test, the analysis was repeated including all baseline covariates (the variable other cancer was omitted to avoid model co-linearity). In this full model, depression remained associated with shorter survival (Wald = 10.869, HR = 1.042, CI: 1.017–1.067, p = .001) and the depression by intervention interaction again remained significant (Wald = 5.636, HR = .963, CI: 0.933–0.993, p = .018) despite inclusion of the 19 covariates. Of these, six were associated with longer survival (married, college graduate, gastrointestinal cancer, lung cancer, hematological cancer, and DNR) and one (days in hospital) was associated with shorter survival (See Table 2).

Table 2.

Results of Cox Proportional Hazards Regression Analysis including all statistical covariates

				95% CI for HR
Variables in the Equation	Wald	HR	p	Lower	Upper
Intervention	0.23	1.128	.635	0.685	1.858
Depression (CES-D at baseline)	10.87	1.042	.001	1.017	1.067
Intervention × Depression	5.64	0.963	.018	0.933	0.993
Demographic variables
Age	0.03	1.001	.870	0.987	1.015
Gender (male)	1.91	0.796	.167	0.577	1.100
Rural	0.19	0.940	.666	0.710	1.245
College graduate	4.28	0.724	.039	0.534	0.983
Married	4.27	0.719	.039	0.525	0.983
Employed	0.01	1.016	.934	0.705	1.464
White	2.16	0.342	.142	0.082	1.432
Cancer site
Lung	3.84	0.306	.050	0.094	1.000
Gastrointestinal	4.09	0.295	.043	0.090	0.964
Genitourinary	1.32	0.487	.251	0.142	1.665
Breast	1.17	0.489	.279	0.134	1.784
Hematological	4.65	0.210	.031	0.051	0.868
Illness-related variables
Days in hospital	13.43	1.050	<.001	1.023	1.078
Days in ICU	0.22	0.963	.638	0.823	1.127
ER visits	0.62	0.930	.431	0.777	1.114
Advanced directives	0.07	1.041	.789	0.777	1.395
Do not resuscitate	21.14	0.312	<.001	0.190	0.512
Chemotherapy	1.48	0.789	.223	0.539	1.155
Radiation	2.55	0.705	.110	0.460	1.083

CI = Confidence Interval; HR = Hazard Ratio (risk of death); Wald = Wald statistic; Intervention = having the palliative care intervention (vs. not having it) entered as a time-varying covariate; CES-D = Center for Epidemiologic Studies Depression Scale; ICU = Intensive Care Unit; ER = Emergency Room.

All variables were entered into a Cox proportional hazards regression model simultaneously.

In order to provide a better understanding of these results, patients were divided into groups as a function of intervention condition and their categorization as relatively high (CES-D > 12) or low (CES-D ≤ 12) in baseline depression as determined by a median split. These six patient groups are represented as separate Kaplan-Meier survival curves in Figure 1. Notably, those in the usual care group with relatively higher baseline depression (indicated by a red solid line in Figure 1) experienced the shortest survival. That said, depression is treated as a continuous variable in all statistical hypothesis tests, and patients were categorized into these high and low depression groups merely for illustrative purposes.

Figure 1.

Kaplan-Meier survival curves for patients with relatively low vs. high depression (determined by a median split on the entire patient sample) separated by intervention group: Usual Care (Study 1), Delayed Intervention (Study 2), and Early Intervention (Studies 1 and 2). Filled circles on each curve indicate censored data. Not corrected for covariates.

Subsidiary Analyses

A series of analyses was conducted to explore whether change in depression over time mediated the positive impact of the intervention on survival (see Supplemental Material available online). As in past research (Pirl et al., 2012) these results failed to demonstrate clear evidence for mediation.

Discussion

We hypothesized that a palliative care intervention designed to improve quality of life and encourage patient activation would moderate the effect of depression on survival in a sample of individuals with advanced cancer. Our results supported this hypothesis. The intervention had a greater impact on survival for individuals with higher baseline depression, and this moderation effect remained significant independent of demographics, cancer site, and illness severity. Figure 1 illustrates the nature of the moderation effect: those with high depression (solid lines) experienced the greatest improvements in survival as a result of the intervention, while the survival outcomes of those low in depression (dashed lines) were relatively unaffected. Not only do these findings add to the literature reporting a link between depression and shorter survival (Cuijpers et al., 2014; Pinquart & Duberstein, 2010; Satin et al., 2009), they demonstrate for the first time that a palliative care intervention designed to improve quality of life among patients with advanced cancer can moderate this relationship by improving survival among individuals with more depressive symptoms. From a clinical perspective, this suggests programs should prioritize such patients when offering early palliative care services, because these individuals are most likely to benefit from these types of interventions.

Mechanisms

Although the structure of the observed moderation effect is relatively straightforward, the underlying mechanisms responsible for it are less clear. We offer four possibilities. First, it is possible that the intervention moderates the association of depression and survival because of its direct effect on the depressive state and its biological sequelae. Thus, there is evidence that psycho-social stressors and depressed mood can alter biological pathways involved in immune function, inflammation, circadian cycles, and stress hormone regulation (Antoni et al., 2006). According to this account, palliative care interventions decrease mortality because they reduce depression and the harmful biological changes that accompany it. Clear support for this mechanism would require showing that the magnitude of individuals’ reduction in depression predicts changes in relevant biomarkers and ultimately longer survival. Although direct biomarker assessment is beyond the scope of this article, we were able to conduct exploratory analyses to test for evidence that change in depression mediated survival outcomes in these data (see Supplemental Material available online). As in past research (Pirl et al., 2012) the results failed to demonstrate clear evidence for mediation. Although these findings do not speak to the role of biological mechanisms in the depression-survival link, they do not support changes in the depressive state as a mechanism by which the intervention improves survival in the present sample.

A second possibility is that depression is associated with shorter survival as a proxy through its association with reduced health-relevant behaviors. Thus, it is possible that certain behaviors (e.g., adhering to treatments, communicating with care providers, etc.) are associated with improved survival and are less likely to be enacted among those who are depressed than among those who are not (Holland & Alici, 2010). The intervention often helped participants discuss and enhance their own day-to-day behaviors, much like a personalized medicine approach. By promoting these behaviors, interventions may improve survival to the extent that the behaviors are otherwise lacking. This line of reasoning could explain why our intervention improved survival primarily among individuals with greater depressive symptoms: These individuals may only have engaged in health-promoting behaviors when explicitly encouraged to do so via the intervention, whereas non-depressed individuals may have engaged in healthy behaviors regardless of the intervention. Furthermore, it suggests improvements in healthy behaviors may occur even when depressive symptoms persist, such that an intervention that encourages healthy behaviors directly could reduce mortality risk regardless of whether or not it also reduces depression. This account could explain our results, as well as similar findings by Pirl and colleagues (2012) that improvement in depression was not associated with improved survival even though their intervention was independently associated with both.

A third possibility is that depression is associated with shorter survival as a consequence of its relation to other unmeasured variables. Although individuals with certain severe psychiatric and substance use disorders were excluded from participating, patients were not screened for all mental health disorders, personality variables, or underlying biological mechanisms that may be associated with both depression and survival. Because baseline depression levels cannot be randomly assigned, we are unable to rule out the possible involvement of a confounding variable. Relatedly, insofar as participants from two independent studies were combined, it is possible that variations in patient characteristics across studies affected our findings. Although it is impossible to eliminate such alternative accounts, we did seek to decrease their plausibility by demonstrating that (a) few differences existed in patient characteristics across studies, (b) few patient characteristics were found to be related to depression, and (c) our observed effects were essentially unchanged when we tested models that took into account cancer site and a variety of demographic and illness severity characteristics.

Finally, rather than framing the observed effects in terms of the moderating influence of the intervention on the association of depression and mortality, it is equally valid to frame them in terms of the moderating influence of depression on the association of the intervention and mortality. Indeed, insofar as the research design meets the criteria of the MacArthur approach to moderation (i.e., depression is independent of the intervention and precedes it in time), adherents of this approach would advocate adopting the depression-moderates-intervention logic in the service of generating hypotheses for future research (e.g., Kraemer et al., 2001; 2002; 2008). For example, it might be that different responses to specific elements of the intervention as a function of individual differences in depression are themselves responsible for the observed differences in mortality.

Limitations

Specific aspects of the current study limit the generalizability of the findings in at least three key respects. First, because we lacked depression measures prior to study enrollment, the nature and timing of patients’ symptomatology remain unknown. This is important because research has identified distinct trajectories of depression leading up to and following negative health events that differentially impact mortality risk (Kyranou et al., 2014; Oksholm et al., 2015). For example, one study of heart attack victims showed that newly emerging depression, but not chronic depression, was associated with mortality (Galatzer-Levy & Bonanno, 2014). On the other hand, the meta-analysis by Pinquart and Duberstein (2010) found that studies in which depression was assessed prior to the cancer diagnosis as well as those that assessed chronic depression based on a clinical diagnosis both found depression to be associated with increased cancer mortality. Future studies might specifically differentiate between newly-developed versus chronic depression in a single design to clarify the relationships among depression, mortality, and palliative care.

Second, our study samples were relatively homogenous, comprised of predominantly white participants from the northeast United States. There is reason to suspect that different populations might exhibit different patterns of results, limiting the generalizability of the present findings. For example, the link between depression and health has been shown to be weaker among East Asian cultures, perhaps due to differences in individual responsibility for negative affect (Curhan et al., 2014). On the other hand, the link between depression and poor health has been found worldwide (Pressman, Gallagher, & Lopez, 2013). Regardless, it would be helpful to explore these relationships among more diverse samples.

Third, our study involved only cancer patients with a prognosis of 2 years or less, and it is therefore possible that the intervention’s moderating role is limited to only severely ill patients nearing the end of life. Additional research is needed before extending these results to populations recovering from past disease or with a chronic, long-term illness. Conversely, it is possible that the observed effect might be stronger among patients at an even earlier stage of disease progression. We are currently testing this intervention in patients with advanced heart failure (Dionne-Odom et al., 2014), but ultimately the generality of the effects will need to be established in a wider variety of patient populations.

Conclusion

This research demonstrates the power of early palliative care to moderate the relationship between depression and survival among individuals newly diagnosed with advanced cancer. Although depression was associated with higher mortality risk in the sample overall, this relationship was diminished among individuals who received the ENABLE palliative care intervention. In contrast, depressed individuals who received only usual cancer care demonstrated significantly higher mortality risk than other patients. Despite decades of investigation, researchers have yet to agree on the mechanisms explaining the relationship between depression and mortality. Studies that focus on identifying moderators of this relationship can move the field toward a more complete understanding of why those who are depressed are at a greater risk to die and how interventions can improve not only the quality but also the length of life.