The Epidemiology of Depressive Symptoms in the Last Year of Life

Abstract

BACKGROUND/OBJECTIVE

Depression impacts quality of life at all life stages, but the epidemiology of depression in the last year of life is unknown. This study’s objectives were to document the epidemiology of depressive symptoms in the year prior to death and to assess how the trajectory of depressive symptoms varies by sociodemographic and clinical factors.

DESIGN

Observational, cross-sectional, cohort study using the Health and Retirement Study.

SETTING

Population-based survey.

PARTICIPANTS

A total of 3274 individuals who died within 12 months after assessment.

MEASURES

Primary outcome: eight-item Center for Epidemiologic Studies Depression Scale (CESD-8). Covariates included sociodemographics, self-reported illnesses, and activity of daily living (ADL) limitations.

RESULTS

Average CESD-8 score increased over the last year of life, with 59.3% screening positive for depression in the last month before death. Depression symptoms increased gradually from 12 to 4 months before death (increase of 0.05 points/month; 95% confidence interval [CI] = 0.01–0.08 points/month) and then escalated from 4 to 1 months before death (increase of 0.29 points/month; 95% CI = 0.16–0.39 points/month). Women, younger adults, and nonwhite adults all demonstrated higher rates of depressive symptoms. Individuals with cancer reported escalating rates of depressive symptoms at the end of life, while individuals with lung disease and ADL impairment demonstrated persistently high rates throughout the year before death.

CONCLUSIONS

This study revealed high rates of depressive symptoms in the last year of life as well as differences in the burden of depressive symptoms. A public health approach must be taken to screen for and appropriately treat symptoms of depression across the lifespan.

Psychological symptoms, such as depression, have a negative impact on patients’ quality of life as they near the end of life. Psychological symptoms in patients with serious illness are associated with higher rates of pain, fatigue, dyspnea, and poor quality of life.^1–6 In addition, psychological symptoms are negatively related to treatment adherence, communication with the healthcare team, and patient-provider alliance.^7–9 Patients with depression have worse survival outcomes than nondepressed patients,¹⁰ making depression a critical issue to screen for and manage in the context of serious illness.

Given the impact of psychological symptoms at end of life, the National Consensus Project and National Hospice and Palliative Care Organization Clinical Practice Guidelines recommend that the “interdisciplinary team assesses and addresses psychological and psychiatric aspects of care based upon the best available evidence to maximize patient and family coping and quality of life.”¹¹ Despite these guidelines, the prevalence and epidemiology of depressive symptoms for patients nearing death are largely unknown. This makes it difficult to assess which groups are at highest risk for depression and what depression interventions may be most effective for specific populations based on the timing of depression onset and death.

Prior research on depressive symptoms at the end of life has been limited to focus on the very end of life, only the last days of life, and specific populations, such as those with cancer or enrolled in hospice. These studies demonstrate high psychological symptom burden, with 44% of patients with cancer reporting depression in the last week of life,¹² and 42% to 70% of cancer patients in hospice reporting depression and associated signs, such as sadness (50%), worry (43%), and nervousness (42%).^13,14

Epidemiologic information is critical to drive population-level strategies for treatment of depressive symptoms for individuals nearing end of life. Patients with serious illness can respond to different pharmacological and nonpharmacological interventions, including tricyclic antidepressants, selective serotonin reuptake inhibitors, or psychosocial interventions, such as cognitive behavioral therapy or mindfulness-based therapy.^15–19 However, ideal modalities may be driven by prognosis. For example, many pharmacotherapies require months to titrate to ideal effect, while newer therapies, such as ketamine, demonstrate effects within days.^20,21 Traditional cognitive-behavioral therapy requires 12 to 20 sessions over as many weeks,²² but abbreviated therapies can be administered in under 8 sessions.²³ Furthermore, most palliative care teams do not employ a dedicated mental health professional.²⁴ Assessing the magnitude of psychological symptoms for those nearing end of life may be helpful to demonstrate where specialist support would be most useful.

One reason why our understanding of depressive symptoms at the end of life is limited is the methodological challenges in identifying an appropriate cohort. Prognosticating who is in the last month of life, much less the last 6 or 12 months, is challenging.^25–27 To address this, prior studies limited the scope to only hospice settings or asked raters or family members to estimate, postmortem, levels of depression, which introduces significant bias.^28–31 We utilized the Health and Retirement Study (HRS), a nationally representative study of adults older than 50 years, which assessed respondent-reported levels of depressive symptoms according to the eight-item Center for Epidemiologic Studies Depression Scale (CESD-8). Given that the time between biennial survey and death is random, but the intervals of interviews are fixed, we present 12 cross-sectional cohorts of population-based, respondent-reported information on the prevalence and correlates of depressive symptoms in monthly intervals over the last year of life.

METHODS

Study Population

We used data from the HRS, a nationally representative panel study of older adults in the United States of more than 37 000 individuals in 23 000 households.^32,33 The HRS surveys adults older than 50 years every 2 years on questions around function and health. Initial response rates for new cohorts recruited into the HRS range from 69.9% to 81.6%, with follow-up rates ranging from 85.4% to 92.3%.³² If participants are unable to complete the interview, a proxy respondent is used. The HRS also provides survey weights for analyses to account for the complex survey design and sampling strategy.³³ The HRS is sponsored by the National Institute on Aging (U01AG009740) and is conducted by the University of Michigan.

Our study uses both HRS survey data and claims and National Death Index–derived death dates. The study was approved by the Icahn School of Medicine at Mount Sinai’s Institutional Review Board, the HRS Data Confidentiality Committee, and the Centers for Medicare and Medicaid Services Privacy Board.

We identified HRS respondents who died within 12 months of a biennial interview. Deaths were determined through the HRS interviewers as well as National Death Index data linked to respondents. Given that the timing between death and HRS survey is random, this allowed us to observe trends in the last year of life using methods similar to prior studies assessing trajectories of pain at the end of life.³⁴ We identified 12 different cohorts based on the number of months between survey completion and death to include in our cross-sectional analyses. Because the HRS protocol does not assess depression when respondents rely on a proxy reporter, we excluded those with proxies. The rate of proxy reliance over the year before death increased from 24.7% 12 months before death to 57.8% 1 month before death. The proportion with no CESD-8 measure, regardless if due to proxy, increased from 26.21% 12 months before death to 59.01% 1 month before death (Supplementary Table S1).

Measures

Depressive Symptoms

Depressive symptoms were measured using the CESD-8.³⁵ This scale has been validated in adults with cancer³⁶ and older adults,³⁷ and it has been used in epidemiological studies of depression.³⁵ The CESD-8 is reliable in interview studies similar to the HRS.^38,39 The CESD-8 in the HRS was adapted from a 20-item scale that was developed for the Established Populations for Epidemiologic Study of the Elderly Survey and asks adults to rate if they experienced the following symptoms over the past week: depression, restless sleep, sadness, feeling like everything takes effort, unhappiness, lack of motivation, and loneliness. The range of CESD-8 scores in the HRS is 0 to 8. A CESD-8 score of greater than 3 has been used as a positive screen for depression, as it has a 71% sensitivity and 79% specificity related to a fully structured interview for diagnosing depression.⁴⁰ We report the percentages of individuals who screened positive for depression at different time points throughout the study, as the CESD-8 is best used as a depression screen rather than a depression severity measure.

Covariates

To assess the clinical and sociodemographic characteristics of respondents, we included information on age, race, sex, net worth (as assessed by a survey of assets on the HRS⁴¹), and education, as reported on the biennial surveys, as well as self-reported information about three common serious illnesses with different trajectories⁴² (cancer, lung disease, and heart disease) and activities of daily living (ADLs) (requiring help with bathing, dressing, toileting, eating, transferring in/out of bed, and walking indoors).

STATISTICAL ANALYSIS

To examine the temporal trajectory of depression over the last year of life, the mean CESD-8 score was plotted for each cohort, ranging from 12 months before death to 1 month before death. Given that a cutoff of 3 in the CESD-8 indicates a positive depression screen,⁴⁰ we repeated this analysis for the proportion of each cohort with CESD-8 of 3 or greater. We also examined the rates of each of the 8 CESD-8 items in each cohort to capture differences in somatic (eg, difficulty sleeping) vs psychological (eg, feeling sad) symptoms. For each individual item, we used a logistic regression model to assess the association between reporting a symptom and months before death (as a continuous variable).

We compared the association between depression and time before death using both linear and spline models to allow for the possibility that the slope of depression over time changes in the months approaching death. The linear regression model and spline regression models with an internal knot at 2, 3, 4, 6, and 9 months prior to death were compared using the AIC and BIC statistics to select the model with best fit.⁴³ As a sensitivity test, we additionally compared models with a second knot at 2, 3, 4, 6, and 9 months prior to death. To examine if depression trajectories varied by demographic and clinical characteristics, we used multivariate models that included each of these characteristics separately with an interaction term between the characteristic and time. All the multivariate models additionally adjusted for age, sex, and race. This allowed us to estimate the predicted CESD-8 score among subgroups at 12, 6, 3, and 1 month prior to death for distinct subgroups. We used survey weights to adjust all analyses for HRS survey design and sampling approach.

RESULTS

Sample

We identified 3274 individuals who died within 12 months of an HRS survey interview and did not have a proxy respondent. The number of respondents in each cohort ranged from 132 (1 month before death) to 335 (12 months before death); Supplementary Table S1 provides full details on sample size by month. Select characteristics are displayed in Table 1. Cohorts were evenly distributed over four age groups, with 20.9% younger than 65 years and 19.1% older than 85 years; nearly half were female, and 14% were nonwhite. Rates of self-reported conditions ranged from 10.8% for congestive heart failure to 33.0% for cancer. One quarter reported an ADL impairment. The average CESD-8 score was relatively stable in the cohorts from 12 to 4 months prior to death, with rising rates in the cohorts 4 months (average score = 2.09) to 1 month (average score = 3.31) prior to death. Similarly, the proportion in each cohort with a score of 3 or greater increased approaching death, with the greatest increase between the cohort 2 months prior to death (42%) and 1 month prior to death (59%) (Figure 1). This compares to the general HRS population, of whom 23.3% have a CESD-8 score of 3 or greater.⁴⁴ Supplementary Tables S2 and S3 provide full details on depression scores by month.

Table 1.

Characteristics of study population (n = 3274)

Characteristic	% of Population
Age, y
≤65	20.9
66–75	27.5
76–85	32.5
>85	19.1
Female	47.5
Nonwhite race	14.0
Net worth in lowest quartilea	30.9
Did not complete high school	14.7
Self-reported conditions
Cancer	33.0
Lung disease	24.4
Congestive heart failure	10.8
ADL impairmentb	25.0

Note: Health and Retirement Study, 1998 to 2014. All proportions are adjusted for complex survey design and weighting.

Abbreviation: ADL, activity of daily living.

^aQuartiles defined for entire Health and Retirement Study population, not this specific cohort.

^bDefined as the need to rely on help for one or more of the following: eating, bathing, toileting, dressing, transferring, and walking inside.

Figure 1.

Trends in depression before death. Note. Data source: Health and Retirement Study, 1998 to 2014. Error bars indicate the 95% confidence interval. Sample size 1 month prior to death was 132; from 2 to 12 months prior to death, it ranged from 204 to 335. All estimates are adjusted by survey weights to account for survey design and sampling approach. CESD-8 indicates eight-item Center for Epidemiologic Studies Depression Scale (range = 1–8), and cutoff for depression is score of 3.

CESD-8 Individual Items

Figure 2 displays the rates of individual CESD-9 items by month before death. The two somatic CESD-8 items (restless sleep and activities took effort) were high and increased more gradually in the cohorts closer to death. The item assessing loneliness, sadness, and feeling unhappy had the greatest increases in the cohorts just prior to death. In the cohort 1 month prior to death, 42% report feeling lonely, 40% report feeling sad, and 34% report they did not feel happy. In a regression model to check for a linear time trend in individual symptoms, there was a significant (P < .05) increase in symptoms over the year prior to death for sadness, activities taking effort, not happy, did not enjoy life, and unmotivated.

Figure 2.

Proportion of individuals reporting positive eight-item Center for Epidemiologic Studies Depression Scale depression measures over the 12 months prior to death. Note. Data source: Health and Retirement Study, 1998 to 2014. The x axis indicates the number of months before death. Error bars indicate the 95% confidence interval. All estimates are adjusted by survey weights to account for survey design and sampling approach. *P < .05, a significant association with the depression measure and months before death in a logistic regression model.

Trajectory Model Fit

Compared to the linear regression and spline models with knots at 2, 3, 6, and 9 months prior to death, the spline model with an internal knot at 4 months prior to death demonstrated the best fit. Models with a second knot did not improve over the single-knot model. This indicates that depression did not just increase over the last month of life, but over the last 4 months of life. In this model, from 12 to 4 months, CESD-8 score increased on average 0.05 points/month (95% confidence interval [CI] = 0.01–0.08 points/month) as the cohorts approached death. Starting at 4 months prior to death, the CESD-8 score increased an average of 0.29 points/month (95% CI = 0.16–0.39 points/month) as the cohorts approached death. This indicates that, on average, nearly one of three people developed a new depression symptom each of the last 4 months of life.

CESD-8 Trajectory by Demographic Subgroup

Table 2 demonstrates how the trajectories of CESD-8 scores varied by different subgroups. While there was a significant difference between the CESD-8 score of men and women 12 months prior to death (1.99 for men vs 2.90 for women; P < .01), for both groups the score increased so that there was no longer a significant difference 1 month prior to death, although the score for men was still lower than that for women (3.11 for men vs 3.54 for women; P = .49). Similarly, the difference between age and net worth groups became less pronounced approaching death, although those of the youngest age and with the lowest net worth groups had higher CESD-8 scores across all time points. While the difference at 1 month prior to death was not statistically significant for nonwhite compared to white individuals (CESD-8 of 4.69 vs 3.16; P = .17), the trend between these groups appeared to be a widening gap between depression scores as the cohorts approached death, with nonwhite respondents having increasingly high depression scores. While differences in CESD-8 by education were not significant, those with less than a high school education at 1 month before death had the highest CESD-8 score seen of any subgroup examined (4.71).

Table 2.

Predicted CESD-8 score among subgroups at distinct time points before death

Variable	Time prior to death
	12 mo			6 mo			3 mo			1 mo
	Mean	95% CI	P value	Mean	95% CI	P value	Mean	95% CI	P value	Mean	95% CI	P value
Sex
Male	1.99	1.63–2.35	<.01	2.08	1.66–2.49	.05	2.13	1.51–2.75	.06	3.11	2.46–3.75	.49
Female	2.90	2.50–3.30		2.72	2.27–3.18		2.99	2.33–3.64		3.54	2.61–4.48
Age, y
≤65	3.21	1.96–4.45	.07	2.52	1.51–3.53	.26	1.96	0.83–3.09	.27	3.99	2.34–5.64	.26
66–75	2.58	1.94–3.21		2.30	1.70–2.90		2.58	1.62–3.55		3.14	1.96–4.32
76–85	2.30	1.81–2.78		2.70	2.05–3.35		3.06	2.29–3.83		3.62	2.75–4.49
≥86	1.91	1.50–2.32		1.84	1.32–2.36		2.73	1.73–3.72		2.70	1.46–3.94
Race
White	2.40	2.05–2.76	.44	2.32	2.01–2.64	.54	2.56	2.02–3.10	.8	3.16	2.69–3.64	.17
Nonwhite	2.78	1.97–3.59		2.72	1.51–2.94		2.4	1.39–3.41		4.69	2.61–6.77
Lowest quartile net worth
No	2.11	1.77–2.46	.02	2.23	1.88–2.58	.11	2.62	2.09–3.15	.6	3.16	2.59–3.74	.24
Yes	3.14	2.44–3.85		2.83	2.20–3.45		2.34	1.42–3.26		3.97	2.83–5.10
Less than high school education
No	2.41	2.06–2.76	.55	2.38	2.04–2.72	.89	2.50	2.06–2.95	.94	3.27	2.69–3.85	.21
Yes	2.74	1.76–3.73		2.33	1.66–3.00		2.45	0.89–4.00		4.71	2.65–6.77
Cancer
No	2.49	2.14–2.85	.67	2.24	1.83–2.65	.29	2.67	2.03–3.31	.53	2.44	1.89–2.99	.01
Yes	2.30	1.57–3.04		2.66	2.08–3.25		2.23	1.15–3.30		3.90	3.06–4.74
Lung disease
No	2.16	1.83–2.49	.01	2.13	1.74–2.53	.015	2.36	1.86–2.86	.06	3.15	2.60–3.70	.29
Yes	3.16	2.55–3.76		2.98	2.48–3.48		3.41	2.50–4.33		3.86	2.78–4.95
Heart disease
No	2.48	2.16–2.80	.61	2.31	1.98–2.63	.26	2.41	1.93–2.90	.12	3.41	2.87–3.94	.41
Yes	2.22	1.31–3.13		3.11	1.78–4.44		3.39	2.23–4.56		2.68	1.14–4.23
ADL impairment
No	2.30	2.01–2.58	.15	2.17	1.8–2.53	.08	2.24	1.74–2.74	.09	2.73	2.16–3.30	<.01
Yes	2.93	2.12–3.74		2.91	2.22–3.61		3.50	2.21–4.78		3.96	3.27–4.64

Abbreviations: ADL, activity of daily living; CESD-8, eight-item Center for Epidemiologic Studies Depression Scale; CI, confidence interval.

CESD-8 Trajectory by Illness

Clinical characteristics appeared to shift the trajectories of depression before death. Those with cancer vs without had similar CESD-8 scores until just before death, where the average CESD-8 rises to 3.90 for those with cancer and 2.44 for those without cancer (P = .01). This contrasted with those with lung disease, who had significantly higher CESD-8 scores along all time points other than 1 month prior to death compared to all other respondents. Even at 1 month prior to death, the mean CESD-8 score for people with lung disease (3.86) was similar to that for those with cancer (3.90). Similar to lung disease, ADL impairment was associated with persistently high CESD-8 scores along the entire 12 months prior to death. At 1 month prior to death, those with ADL impairment had an average CESD-8 score of 3.96, which was the highest seen in any of the clinical subgroups.

DISCUSSION

To our knowledge, this is the first nationally representative study to assess the epidemiology of self-reported depression in the last year of life. This study revealed that rates of depressive symptoms increased over the last year of life and, in particular, the last 4 months of life. By the last month of life, most respondents (59%) were above the threshold for a positive depression screen. However, trajectories of depressive symptoms at the end of life varied considerably by demographic, socioeconomic, and clinical characteristics. These results have implications for strategies to reduce depressive symptoms and differences in depressive symptoms at the end of life.

Over the last year of life, depression scores for all subgroups increased, but women, younger adults, and those with the lowest net worth had significantly higher scores 12 months before death. This indicates a greater longitudinal burden of depressive symptoms over the course of the last year of life for these populations, which could potentially be mitigated through proactive depression screening and treatment.

There are several reasons why sex, age, and socioeconomic status may be associated with more depressive symptoms over the last year of life. Regarding sex, women report higher rates of depression than men across the lifespan, so it is not surprising that this effect holds in the last year of life.^45,46 Regarding age, younger adults with diagnoses of serious illness may have more difficulty accepting death or illness because it is perceived as “untimely” when younger adults are diagnosed with terminal illness.⁴⁷ Regarding socioeconomic status, poverty, in general, is associated with higher chronic stress contributing to depression risk, which may be compounded by the financial challenges faced in the last year of life. For example, treatments for serious illness and household financial stressors of missed work due to both illness and family caregiving would likely increase rates of stress and impact symptoms of depression.

In contrast, the gap between depression rates in respondents of white vs nonwhite race trended toward a pattern of widening over time, with nonwhite participants reporting 1.5 additional CESD-8 items in the last month of life compared to their age-and sex-matched white counterpart. This racial difference in end-of-life depression may be driven by differences in the end-of-life care received by different racial groups, given that nonwhite populations receive less hospice, receive more aggressive medical treatments, and report poorer communication than white older adults at the end of life.^48–50

Trajectories of depressive symptoms over the last year of life also varied by clinical factors. Individuals with cancer diagnoses reported steep increases in levels of depressive symptoms in the last month of life compared to decedents who did not have cancer. This may be because the clinical trajectory of cancer is more likely to demonstrate a steep increase in disease burden at the end of life, as opposed to conditions with more gradual or fluctuating changes over time.⁵¹ In contrast, individuals with lung disease had higher rates of depressive symptoms over the entire 12 months prior to death. This likely points to the long-term burden of symptoms individuals with lung disease have that affect their mental health and quality of life. Finally, individuals with ADL impairment, regardless of diagnosis, had greater depressive symptoms at all time points and the highest level of depression of all subgroups in the month prior to death. This is consistent with prior research on the relationship between depression and physical impairment and highlights the importance of recognizing changes in functional status to identify unmet needs, including psychological support.^52–54

Differences between clinical populations in the trajectory of depressive symptoms at the end of life also have implications for potential therapeutic modalities. Patients with cancer may not become depressed until the final weeks or months of life and, thus, not have time for multisession psychotherapy and antidepressant treatments, which take months to see full effect. They may benefit more from preventative treatments and earlier on-boarding of supportive services prior to the onset of depressive symptoms. Additionally, alternative drug treatments, such ketamine, are under investigation^20,21; and they may be beneficial for patients with depression and limited life expectancies. In contrast, those with advanced lung disease and functional disability may have time for therapies better suited to longer prognoses if their depression is promptly identified.

Limitations

Due to HRS protocol, CESD-8 was not collected when a proxy respondent was used. While the rates of proxy respondents increased closer to death, this likely created a bias toward underreporting depressive symptoms at the end of life. Those with proxies likely had worse symptom burden and, thus, higher depression, although they may not be able to express it. It is likely that our data underrepresent the amount of depressive symptoms patients experience approaching end of life.

Another possible limitation is using the CESD-8 as a measure for depressive symptoms at end of life. While the full CESD has been validated in several populations, the shortened version has not undergone rigorous psychometric evaluation in populations nearing end of life. The CESD-8 includes two items that assess somatic symptoms of depression (difficulty sleeping and low energy), which may conflate serious illness adverse effects with depression symptoms. However, the CESD-8 is the best tool available in the HRS, and we examined individual items to ensure that increases in CESD-8 scores were not purely due to somatic items. We saw an increase in nonsomatic items (sadness and loneliness) approaching death as well.

In addition, the proportion of nonwhite respondents in the weighted sample (demonstrated in Table 1) was low at 14.0%. However, HRS oversamples black and Hispanic older adults to better capture the experience of these important population groups,³² so the nonweighted proportion of nonwhite respondents in our survey is higher than demonstrated.

Finally, we used cross-sectional data to estimate prevalence and correlates of symptoms of depression in the months leading up to death. Ideally, we would follow patients longitudinally to determine trajectories and potential causes of depression symptoms in the last year of life. Given the unknown nature of when death will occur, a longitudinal study of patients in the last year of life with frequent assessments is likely impossible. In all cross-sectional studies, biases of nonresponse can occur, but we believe that nonresponse bias in this particular sample likely underestimated the prevalence of depression in the last year of life.

CONCLUSION

This work revealed high rates of depressive symptoms in the last year of life that escalated in the last 4 months before death. Furthermore, there are differences in the burden of depressive symptoms experienced by those in different sociodemographic and clinical populations. Given the range of options to treat depression, unaddressed depressive symptoms in the last year of life must be a focus of both quality measurement and improvement. This will require research, health system, and policy efforts to support routine depression screening throughout the course of serious illness, integrated psychological and psychiatric expertise, and evidence-based treatments that are tailored to anticipated prognosis. While depressive symptoms at the end of life are common, they are treatable and must be proactively addressed to reduce distress and ensure that everyone has the opportunity to experience a “good death,” free of depressive symptoms.