Seven-Year Trajectories of Depressive Symptoms and Their Predictors Among Older Americans

Abstract

Objective: This study examined the trajectories of depressive symptoms and associated factors in older adults using an enhanced group-based trajectory modeling. Method: The study sample consisted of 7,573 adults aged ≥65 years from the National Health and Aging Trends Study (Rounds 1-7). Depressive trajectories were estimated using a group-based trajectory modeling accounting for nonrandom attrition. Results: A four-trajectory model including “persistently low” (77.7%), “increasing” (7.9%), “declining,” (5.5%), and “persistently high” (8.9%) was the best fit using methods accounting for nonrandom attrition. In comparison, methods not accounting for attrition estimated that only 3.2% of older adults were on the “persistently high” trajectory. There were significant differences in depressive trajectories by age, race/ethnicity, sex, physical, and cognitive functioning, and social connections. Discussion: Persistently high depressive symptoms affected a larger proportion of older adults than previously estimated. Depression had a more long-term and increasing course in the oldest-old.

Keywords

group-based trajectory modeling nonrandom attrition late-life depression

Introduction

Depression is a common mental disorder in older adults that affects their health and quality of life. Major depression occurs in 2% of adults aged 60 years or older (Sjöberg et al., 2017), and despite being less common in late-life, has a more long-term course than in younger adults (Haigh, Bogucki, Sigmon, & Blazer, 2018). In addition, subclinical levels of depressive symptoms affect 10% to 15% of community-dwelling older adults (Laborde-Lahoz et al., 2015; Mohebbi et al., 2019) and increase substantially among older adults in health care and long-term care settings (Seitz, Purandare, & Conn, 2010). Subclinical depressive symptoms, even in the absence of major depression, are associated with functional impairments comparable to or worse than those with long-term physical conditions such as heart disease, arthritis, and diabetes (Wells & Burnam, 1990).

An increasing number of studies have suggested variability in the long-term trajectories of depressive symptoms, with most individuals experiencing consistently few or no depressive symptoms and a notable minority reporting persistent symptoms (Andreescu, Chang, Mulsant, & Ganguli, 2008; Kuchibhatla, Fillenbaum, Hybels, & Blazer, 2012; Liang, Xu, Quiñones, Bennett, & Ye, 2011; Saeed Mirza et al., 2018). As noted in a recent systematic review, the number of studies focused on the heterogeneity in depression trajectories has increased significantly over the past 15 years with the development of group-based trajectory modeling (Musliner, Munk-Olsen, Eaton, & Zandi, 2016). Most of these studies identified 3 or 4 trajectory groups with varying degree of severity and stability, supporting the idea that the long-term course of depressive symptoms is heterogeneous in the general population (Musliner et al., 2016). This heterogeneity has important implications for our understanding of the cause and the burden of depression, which can inform prevention and treatment efforts.

In addition to characterizing intrapersonal depressive symptoms over time, previous studies have also examined the predictors of depressive trajectories to identify interpersonal differences. Common risk factors of trajectories with greater symptoms included female sex, lower income and education, and non-White race/ethnicity from studies involving mixed-age samples (Musliner et al., 2016). Studies focused on older adults have identified additional predictors, including social network and social support (Byers et al., 2012; Kuchibhatla et al., 2012; Park, 2017; Rote, Chen, & Markides, 2015), functional impairment (Byers et al., 2012; Rote et al., 2015), mobility limitations (Carrière et al., 2017), and certain physical illnesses (Byers et al., 2012; Carrière et al., 2017; Liang et al., 2011; Montagnier et al., 2014; Rote et al., 2015; Tampubolon & Maharani, 2017). Moreover, baseline depressive symptom severity has been consistently linked to depressive trajectories in older adults (Andreescu et al., 2008; Cui, Lyness, Tang, Tu, & Conwell, 2008; Montagnier et al., 2014). In addition, several studies have reported that age influenced the size and shape of depressive trajectories (Liang et al., 2011; Montagnier et al., 2014; Tampubolon & Maharani, 2017). Results regarding the role of cognitive function on depressive trajectories are inconsistent, with both null (Byers et al., 2012; Carrière et al., 2017) and significant findings (Holmes et al., 2018; Tampubolon & Maharani, 2017) reported.

Despite these advances in research on depressive trajectories, most studies focused on younger adults (see Musliner et al., 2016, for a review). Studies that characterized the complex trajectories of depressive symptoms in older adults were limited, and only a few used population-based samples (Andreescu et al., 2008; Byers et al., 2012; Carrière et al., 2017; Kuchibhatla et al., 2012; Liang et al., 2011; Montagnier et al., 2014; Park, 2017; Saeed Mirza et al., 2018; Tampubolon & Maharani, 2017). Age differences in the diagnosis, cause, prevalence, and prognosis of depression have been documented (Haigh et al., 2018). Findings in younger adults, therefore, may not apply to older adults. Moreover, previous studies analyzing depressive trajectories relied on methods (e.g., growth-curve modeling and basic group-based trajectory modeling) that assumed missing at random. However, high levels of depressive symptoms are an independent risk factor for mortality among older adults (Schulz et al., 2000). Studies have also shown that the risk of premature mortality differs by depressive trajectory groups (Andreescu et al., 2008; Kuchibhatla et al., 2012; Liang et al., 2011; Saeed Mirza et al., 2018). Therefore, attrition due to mortality and other causes is unlikely to be independent of the depressive trajectory groups, violating the assumption of missing at random. Using methods that assume missing at random when attrition rates differ by groups can lead to biased estimates of trajectory group size (Haviland, Jones, & Nagin, 2011). A previous study found a higher risk of mortality in older adults with high and persistent depressive symptoms than those with few symptoms. (Andreescu et al., 2008). If this pattern holds, the proportion of older adults on the trajectory with high and persistent symptoms was likely underestimated in previous studies due to failure to account for nonrandom attrition.

The purpose of this study was to examine the 7-year trajectories of depressive symptoms in a nationally representative sample of community-dwelling older adults and to investigate the risk factors shaping the patterns of these trajectories. Depressive trajectories were hypothesized to be heterogenous over the 7-year period. This study used an enhanced group-based trajectory modeling that allows the joint estimation of depressive trajectories and probabilities of attrition, where attrition rates are specific to the trajectory group (Haviland et al., 2011). Use of this enhanced method may generate more accurate estimates of the proportions of older adults are on different depressive trajectories and may identify associated risk factors. This may influence efforts to reduce depression in older adults.

Methods

Participants

The National Health and Aging Trends Study (NHATS) is a nationally representative longitudinal study of people aged 65 and older who are also enrolled in Medicare (www.NHATS.org). The baseline sample interview was conducted in 2011 with 7,609 older Medicare beneficiaries who resided in the community and not in nursing homes. Annual follow-up interviews were scheduled for all eligible persons regardless of their residential status. A replenishment of the sample was added in 2015. NHATS methodology included the use of proxies familiar with the sample person’s health and daily routine when the sample person could not respond, typically due to an illness or impairment. In this study, Round 1 (2011) through Round 7 (2017) of the NHATS public use files from the baseline sample were analyzed. The final study sample consisted 7,573 individuals, after excluding 36 participants with missing data on depressive symptoms for all of the seven rounds of data collection.

Measures

Elevated depressive symptoms

The Patient Health Questionnaire-2 (PHQ-2) measures how often a person has been bothered by “little interest or pleasure in doing things” and “feeling down, depressed or hopeless” on a 4-point Likert-type scale: “not at all” (0), “several days” (1), “more than half the days” (2), and “nearly every day” (3) (Löwe, Kroenke, & Gräfe, 2005). NHATS used a reference time frame of “past 30 days” when administering the PHQ-2 to both proxy and self-respondents to screen for depressive symptoms of the sample persons. A cut-off score of 3 on the PHQ-2 has a sensitivity of 0.87 and a specificity of 0.78 for major depressive disorder, and has a sensitivity of 0.79 and specificity of 0.86 for any depressive disorder (Löwe et al., 2005). A dichotomous indicator of elevated depressive symptoms was created using the recommended cut-off score.

Predictors of depressive trajectories

Predictors of depressive trajectories were selected based on risk factors from previous studies. Final selection struck a balance between comprehensiveness and parsimony. Predictors were measured at baseline to reduce the possibility that these variables were influenced by the depressive trajectories themselves. Demographic factors included age, sex, and race/ethnicity. Socioeconomic status (SES) factors included educational attainment and family income, presented in quartiles. Social factors included an indicator of social isolation using six NHATS items for marriage/partnerships; contact with family and friends; church participation; and club participation (Pohl, Cochrane, Schepp, & Woods, 2017). A cutoff of ≥2 was used to create a dichotomous indicator of social isolation (Pohl et al., 2017). Health factor included a count of self-reported physician diagnoses of eight long-term conditions (hypertension, heart disease, arthritis, osteoporosis, diabetes, lung disease, stroke, and cancer). Cognitive functioning was measured by an NHATS classification for dementia status (no dementia, possible dementia, and probable dementia) based on a self-reported diagnosis of dementia or Alzheimer’s disease, an AD8 Dementia Screening Interview, and cognitive tests (Kasper, Freedman, & Spillman, 2013). Functioning factors included a count of functional limitations for activities of daily living (ADL) (eating, bathing, toileting, and dressing); for instrumental activities of daily living (IADL) (laundry, grocery shopping, meal preparation, and keeping track of medication); and a count of mobility-related limitations (going outside, getting around inside, and getting out of bed). A limitation in each activity was defined as (a) needing assistance with that activity (must be for health or functioning reasons for IADL activities) or (b) having difficulty performing the activity alone. An indicator of self-reported past-year hospitalization was also included. Mental health symptom severity was measured by the PHQ-4, combing the PHQ-2 and the two-item measure of generalized anxiety disorder (GAD-2) (Kroenke, Spitzer, Williams, & Lowe, 2009). In addition, an indicator of proxy respondent was included to adjust for the potential bias associated with proxy responses (Skolarus et al., 2010).

Statistical Analysis

Depressive trajectories were identified using a group-based trajectory modeling, enhanced to account for nonrandom attrition (Haviland et al., 2011). The basic group-based trajectory modeling is a specialized application of finite mixture modeling designed to identify clusters of individuals following similar outcomes over time (Nagin, 2005). The basic modeling assumes independence of probabilities of group membership and attrition (i.e., missing at random), like other methods for estimating developmental trajectories such as growth-curve modeling. However, simulations showed that methods with missing at random assumption led to biased estimates of trajectory group size (Haviland et al., 2011), particularly in studies on topics related to disability (Zimmer, Martin, Nagin, & Jones, 2012). Given the close relationship between depression and disability, and given prevalent attrition in longitudinal studies of older adults, using methods that assume missing at random may lead to biased estimates of trajectory groups. The enhanced modeling estimates attrition as a function of time before dropping out using a logit distribution simultaneously with the trajectory group. Estimated probabilities of dropping out are specific to each depressive trajectory group.

Depressive trajectories were estimated using the Proc Traj plug-in (Jones & Nagin, 2007) through an iterative process of adding one trajectory group at a time and changing the polynomial type of time for each group. All-cause attrition (i.e., death or other causes of loss to follow-up) was modeled along with the depressive trajectories to account for unequal probabilities of attrition among different depressive trajectories. A small amount of missingness (<1%) occurred due to no-response or incomplete responses to the depression assessments; these missing data were treated as missing at random. The optional trajectory group number was determined by a combination of four criteria: (a) a comparison of the change in Bayesian Information Criteria (BIC), (b) the average posterior probability of group assignment (≥0.7), (c) group size such that no less than 5% of the study sample are assigned to one trajectory group, and (d) conceptual considerations of group distinctiveness and interpretability (Nagin, 2005).

Each participant was assigned to the trajectory group for which they had the highest posterior probabilities of group membership. Subsequently, multinomial logistic regression was used to identify risk factors associated with depressive trajectories. Analyses were conducted using Stata 15.1 (Stata Corp., College Station, TX) and adjusted for the baseline complex survey design of NHATS using Taylor linearization for variance estimation.

Sensitivity analysis

A basic group-based trajectory model was estimated without accounting for attrition to examine the influence of unequal probabilities of attrition on group assignments (N = 7,573). The main analysis did not distinguish between mortality and other causes of attrition because they could not be modeled as separate trajectories along with the depressive trajectories. An enhanced model accounting for mortality only (N = 4,765) was estimated to check the robustness of the main findings against different causes of attrition. Finally, a group-based trajectory model was estimated with completers only (N = 2,822) to examine the influence of excluding incomplete data on group assignments.

Results

Seven-Year Depressive Trajectories Accounting for Unequal Probabilities of Attrition

A logit model with four trajectory groups was the best fit for the data based on considerations of changes in BIC (Table 1), group distinctiveness and interpretability, group size, and the average posterior probability of group assignment (p = .82, 0.72, 0.70, and 0.72, for groups 14, respectively). As shown in Figure 1, Group 1 (“persistently low”) had a very low risk of experiencing elevated depressive symptoms throughout the 7-year period, best presenting 77.7% of the weighted sample (95% confidence interval [CI]: 76.2, 79.2). Group 2 (“increasing”) had a steady increase in the risk of elevated depressive symptoms, best representing 7.9% of the weighted sample (95% CI: 7.2, 8.7). Group 3 (“declining”) captured participants whose depressive risk declined over time until most of them stopped reporting elevated depressive symptoms, best representing 5.5% of the weighted sample (95% CI: 4.8, 6.3). Group 4 (“persistently high”) represented a group of participants who maintained elevated depressive symptoms throughout the study period, best representing 8.9% of the weighted sample (95% CI: 8.1, 9.7).

Table 1.

Tabulated BIC and 2∆BIC.

Number of groups	BIC	Null model	2 (∆BIC)
1	−21,379.31
2	−20,245.88	1	2,266.86
3	−20,195.15	2	101.46
4	−20,160.72	3	68.86
5	−20,171.86	4	−22.28
6	−20,190.15	5	−36.58

Note. N = 7,573 for all models. BIC = Bayesian Information Criterion.

Figure 1.

Depressive trajectories over 7 years jointly modeled with attrition. It shows depressive trajectories with estimated probability of experiencing elevated depressive symptoms at each study round and the weighted proportions (95% confidence intervals in parentheses) of the study population following each trajectory.

Sample Characteristics by Depressive Trajectories

Depressive trajectory groups differed significantly in demographic backgrounds, SES, social status, physical health, functioning, health care utilization, and initial mental health symptom severity (Table 2). The “persistently low” group had the highest share of men, non-Hispanic Whites, persons with a college degree, no social isolation, no dementia; and had the lowest long-term disease count, functional limitations, mobility limitations, rate of past-year hospitalization, and PHQ-4 score.

Table 2.

Baseline Sample Characteristics by Depressive Trajectory Group.

	Persistently low	Increasing	Declining	Persistently high
Age groups (%)
65-69 years	83.2 [80.5, 85.6]	4.9 [3.8, 6.3]	4.6 [3.4, 6.1]	7.3 [5.6, 9.4]
70-74 years	79.8 [77.5, 82.0]	7.9 [6.6, 9.5]	4.8 [3.6, 6.4]	7.4 [6.1, 9.1]
75-79 years	75.9 [73.0, 78.5]	9.3 [7.6, 11.3]	5.9 [4.7, 7.5]	8.9 [7.3, 10.9]
80-84 years	74.0 [71.4, 76.4]	8.8 [7.3, 10.7]	6.3 [5.0, 7.9]	10.9 [9.3, 12.8]
85-89 years	71.3 [68.0, 74.3]	9.4 [7.6, 11.5]	8.7 [6.9, 10.8]	10.7 [8.6, 13.1]
90 years or above	63.9 [59.7, 67.9]	14.4 [11.7, 17.6]	4.7 [3.2, 6.8]	17.0 [13.9, 20.5]
Sex (%)
Female	76.4 [74.5, 78.1]	8.3 [7.4, 9.3]	6.0 [5.2, 7.0]	9.3 [8.3, 10.4]
Male	79.5 [77.5, 81.3]	7.4 [6.4, 8.6]	4.8 [4.0, 5.8]	8.3 [7.1, 9.7]
Race/ethnicity (%)
White, non-Hispanic	80.1 [78.4, 81.7]	7.0 [6.3, 7.8]	5.0 [4.2, 6.0]	7.9 [7.0, 8.8]
Black, non-Hispanic	67.7 [64.5, 70.8]	12.7 [10.8, 14.8]	7.9 [6.8, 9.1]	11.7 [10.2, 13.5]
Hispanic	62.7 [56.4, 68.7]	11.7 [8.4, 16.1]	8.6 [6.6, 11.2]	17.0 [12.5, 22.7]
Other	75.6 [69.8, 80.7]	9.1 [6.7, 12.4]	5.3 [3.2, 8.7]	9.9 [7.0, 13.8]
Education (%)
Less than high school	62.8 [59.2, 66.3]	12.9 [11.2, 14.8]	8.6 [6.9, 10.6]	15.7 [13.8, 17.9]
High school	76.7 [74.2, 78.9]	7.7 [6.4, 9.4]	6.1 [5.1, 7.4]	9.5 [8.1, 11.0]
Some college, no degree	81.4 [78.8, 8.4]	7.3 [6.0, 8.8]	4.6 [3.5, 6.0]	6.8 [5.5, 8.4]
College graduate	87.4 [85.7, 88.9]	4.9 [4.0, 6.0]	3.4 [2.5, 4.6]	4.4 [3.6, 5.2]
Family income in 2011 ($) (M)	64,266 [57,806, 70,726]	34,155 [28,856, 39,455]	33,313 [29,162, 37,466]	41,148 [21,228, 61,068]
Socially isolation status (%)
Not isolated	90.9 [89.0, 92.5]	3.9 [3.0, 5.1]	2.2 [1.5, 3.3]	3.0 [2.1, 4.1]
Isolated	74.6 [72.9, 76.2]	8.8 [8.0, 9.7]	6.3 [5.5, 7.2]	10.3 [9.4, 11.3]
long-term disease count (M)	2.22 [2.17, 2.27]	2.67 [2.55, 2.79]	2.81 [2.64, 2.98]	3.26 [3.15, 3.37]
Dementia status (%)
No dementia	82.5 [80.9, 83.9]	6.6 [5.9, 7.4]	4.8 [4.0, 5.7]	6.2 [5.5, 7.0]
Possible dementia	69.2 [65.5, 72.8]	11.0 [8.9, 13.5]	7.3 [5.7, 9.2]	12.5 [10.1, 15.5]
Probable dementia	49.2 [46.0, 52.3]	14.8 [12.5, 17.4]	9.6 [7.7, 11.8]	26.5 [23.0, 30.3]
Functional limitations (M)	0.9 [0.8, 0.9]	1.9 [1.7, 2.1]	2.3 [2.0, 2.6]	3.5 [3.2, 3.7]
Mobility limitations (M)	0.4 [0.4, 0.4]	0.9 [0.8, 1.0]	1.1 [1.0, 1.2]	1.5 [1.4, 1.6]
Hospitalized in past year
No	80.6 [78.9, 82.2]	7.3 [6.5, 8.2]	4.8 [4.0, 5.7]	7.3 [6.5, 8.3]
Yes	67.1 [64.8, 69.3]	10.1 [8.7, 11.6]	8.1 [6.8, 9.6]	14.8 [13.1, 16.6]
PHQ-4 score (M)	1.1 [1.1, 1.2]	2.4 [2.2, 2.6]	5.9 [4.8, 5.4]	6.3 [6.0, 6.5]
Proxy status (unweighted, %)
Self-respondent	76.3	9.1	5.9	8.6
Proxy respondent	49.4	12.2	8.6	29.8
Attrition (unweighted %)
Completers	79.6	10.8	4.3	5.3
Attrition, due to mortality	66.1	8.1	9.7	16.1
Attrition, all other causes	60.7	13.3	8.7	17.3
Sample size	5,628	707	465	773
Estimated proportion in the population (%)	77.7 [76.2, 79.2]	7.9 [7.2, 8.7]	5.5 [4.8, 6.3]	8.9 [8.1, 9.7]

Note. All bivariate comparisons were statistically significant at p < .05. Unless otherwise noted, weighted estimates were presented applying NHATS complex survey design features using Taylor linearization for variance estimation. PHQ = Patient Health Questionnaire; NHATS = National Health and Aging Trends Study.

Risk Factors Shaping Depressive Trajectories

Table 3 presents results from the multinomial logistic regression using the “persistently low” trajectory group as the reference category. The relative risks of being on the three depressive trajectories (i.e., “increasing,” “declining,” and “persistently high) were consistently higher for non-Hispanic Blacks than non-Hispanic Whites (relative risk ratios [RRRs] ranged from 1.47 to 1.53, p < .05 for all), socially isolated persons (RRRs ranged from 1.45 to 1.89, p < .05 for all), and persons with higher baseline PHQ-4 scores (RRRs ranged from 1.33 to 2.20, p < .001 for all). Compared with the young-old (65-69 years), the oldest-old (≥90 years) had twice the risk of being on the “increasing” (RRR = 2.11, 95% CI: 1.39, 3.22, p = .001) and “persistently high” trajectories (RRR = 2.07, 95% CI: 1.36, 3.15, p = .001), and the middle-old (75-79 years) had a higher risk of being on the “increasing” trajectory (RRR = 1.68, 95% CI: 1.15, 2.45, p = .008). The old-old (85-89 years) had a higher risk of being on the “declining” trajectory (RRR = 1.80, 95% CI: 1.13, 2.86, p = .014). College graduates had a lower risk of being on the “increasing” (RRR = 0.66, 95% CI; 0.46, 0.93, p = .019) and the “persistently high” trajectories (RRR = 0.68, 95% CI: 0.48, 0.96, p = .028) compared to those without a high-school diploma. Persons living in high-income households had a lower risk of being on the “increasing” and the “declining” trajectories. Having probable dementia (RRR = 1.69, 95% CI: 1.27, 2.25, p = .001) and mobility limitations (RRR = 1.17, 95% CI: 1.02, 1.35, p = .030) were unique risk factors of being on the “increasing” trajectory. Some factors uniquely predicted the “persistently high” trajectory membership. The risk of being on the “persistently high” trajectory was higher in men than in women (RRR = 1.54, 95% CI: 1.24, 1.92, p < .001), in persons with a greater number of long-term diseases (RRR = 1.13, 95% CI: 1.03, 1.23, p = .009), and in persons with more functional limitations (RRR = 1.12, 95% CI: 1.03, 1.20, p = .006).

Table 3.

Multivariable Multinomial Logistic Regression Predicting Depressive Trajectory Group as a Function of Base Characteristics.

Baseline characteristics	Increasing versus persistently low		Declining versus persistently low		Persistently high versus persistently low
Baseline characteristics	RRR	p	RRR	p	RRR	p
Age groups
65-69 years	Reference		Reference		Reference
70-74 years	1.45 [0.99, 2.13]	.057	1.01 [0.68, 1.50]	.970	0.94 [0.58, 1.55]	.816
75-79 years	1.68 [1.15, 2.45]	.008	1.47 [0.96, 2.24]	.076	1.30 [0.89, 1.91]	.170
80-84 years	1.43 [0.99, 2.07]	.056	1.34 [0.89, 2.00]	.157	1.21 [0.77, 1.89]	.400
85-89 years	1.41 [0.94, 2.10]	.093	1.80 [1.13, 2.86]	.014	0.97 [0.60, 1.57]	.905
90 years or above	2.11 [1.39, 3.22]	.001	1.25 [0.66, 2.36]	.486	2.07 [1.36, 3.15]	.001
Male (reference: female)	1.23 [0.99, 1.53]	.067	1.23 [0.99, 1.52]	.066	1.54 [1.24, 1.92]	.000
Race/ethnicity
White, non-Hispanic	Reference		Reference		Reference
Black, non-Hispanic	1.52 [1.23, 1.90]	.000	1.47 [1.09, 1.99]	.012	1.53 [1.19, 1.95]	.001
Hispanic	1.25 [0.85, 1.84]	.254	1.12 [0.79, 1.60]	.521	1.49 [0.88, 2.51]	.133
Other	1.23 [0.66, 2.28]	.506	1.00 [0.48, 2.10]	.999	0.92 [0.52, 1.62]	.760
Education
Less than high school	Reference		Reference		Reference
High school	0.74 [0.55, 1.00]	.048	0.84 [0.58, 1.21]	.335	0.84 [0.62, 1.14]	.257
Some college, no degree	0.79 [0.55, 1.12]	.177	0.75 [0.50, 1.12]	.156	0.77 [0.52, 1.14]	.190
College graduate	0.66 [0.46, 0.93]	.019	0.73 [0.48, 1.10]	.131	0.68 [0.48, 0.96]	.028
Family income in 2011
First Quartile	Reference		Reference		Reference
Second Quartile	0.79 [0.63, 0.98]	.035	0.59 [0.43, 0.81]	.002	1.02 [0.76, 1.37]	.904
Third Quartile	0.58 [0.40, 0.83]	.004	0.48 [0.33, 0.70]	.000	0.84 [0.61, 1.16]	.295
Fourth Quartile	0.46 [0.31, 0.69]	.000	0.60 [0.38, 0.93]	.024	0.73 [0.49, 1.09]	.123
Socially isolated	1.45 [1.06, 1.99]	.021	1.89 [1.15, 3.09]	.012	1.58 [1.04, 2.39]	.033
Long-term disease count	1.05 [0.98, 1.13]	.170	0.97 [0.88, 1.07]	.563	1.13 [1.03, 1.23]	.009
Dementia status
No dementia	Reference		Reference		Reference
Possible dementia	1.23 [0.96, 1.59]	.105	1.02 [0.70, 1.50]	.901	1.19 [0.79, 1.79]	.390
Probable dementia	1.69 [1.27, 2.25]	.001	1.18 [0.77, 1.81]	.440	1.32 [0.89, 1.95]	.165
Functional limitations	1.01 [0.95, 1.09]	.679	1.01 [0.93, 1.11]	.754	1.12 [1.03, 1.20]	.006
Mobility limitations	1.17 [1.02, 1.35]	.030	1.11 [0.95, 1.29]	.193	1.06 [0.92, 1.22]	.431
Past-year hospitalization	1.05 [0.84, 1.32]	.661	1.06 [0.80, 1.42]	.677	0.89 [0.69, 1.15]	.364
PHQ-4 score	1.33 [1.27, 1.40]	.000	2.00 [1.88, 2.13]	.000	2.20 [2.06, 2.34]	.000
Proxy respondents	0.81 [0.54, 1.21]	.297	0.85 [0.52, 1.38]	.497	1.38 [0.85, 2.25]	.188

Note. Effective sample size was 7,413 after listwise deletion of missing data on X variables. Reference category was the “persistent low” group. NHATS complex survey features were applied to generate weighted estimates using Taylor linearization for variance estimation. RRR = relative risk ratios; PHQ = patient health questionnaire; NHATS = National Health and Aging Trends Study.

To further explore the difference between the “persistently high” and the “declining” group, the multinomial logistic model described above was rerun using the “persistently high” group as the reference category. Persons with more long-term diseases (RRR = 0.86, 95% CI: 0.78, 0.95, p = .003) and functional limitations (RRR = 0.90, 95% CI: 0.83, 0.99, p = .029) were less likely to be on the “declining” trajectory. (Results not shown in tables.)

Sensitivity Analysis

Compared to the trajectory model accounting for attrition (i.e., main analysis), the basic trajectory model without accounting for attrition (N = 7,573) and the model accounting for mortality only underestimated the proportion of persons on the “persistently high” trajectory (3.2% and 4.5%, respectively) and overestimated that of persons on the “increasing” trajectory (12.6% for both). The trajectory model with completers only (N = 2,822) overestimated the proportion of persons on the “persistently low” trajectory (83.1%) and underestimated that of persons on the “persistently high” trajectory (4.1%).

Discussion

This study expanded research on depressive trajectories in older adults by applying a group-based trajectory model enhanced to account for nonrandom attrition and by examining a rather comprehensive list of predictors shaping the trajectories. Four distinct depressive trajectories over a 7-year period were identified. Most community-dwelling older adults had persistent no or low depressive symptoms. A nontrivial proportion (8.9%) had persistent symptoms, representing 3.4 million older adults in the population. Nearly 3 million (7.9%) older adults in the population experienced increasing depressive symptoms, whereas 2.1 million (5.5%) experienced declining symptoms. Depressive trajectories were shaped by demographics, SES, social contact, illness and functioning, as well as initial symptom severity.

The number of distinct trajectories identified in this study is in line with findings from previous studies, most of which identified 3 or 4 trajectory groups with most individuals experiencing consistently few or no depressive symptoms and a minority reporting persistent symptoms (Musliner et al., 2016). Nevertheless, the proportion of older adults on the trajectory with high and persistent symptoms in this study (8.9%) is considerably higher than that from previous studies, where three reported around 3% (Byers et al., 2012; Liang et al., 2011; Saeed Mirza et al., 2018) and one reported 5.4% (Kuchibhatla et al., 2012). The sensitivity analysis involving the basic group-based trajectory model not accounting for attrition estimated that 3.2% of older adults were on the persistently high trajectory, which is similar to reports from previous studies. These patterns provide evidence to support the previously stated hypothesis that methods not accounting for nonrandom attrition may underestimate the proportion of older adults on the trajectory with a high burden of depression. The number of older adults with a persistently high burden of depression may be larger than what was previously thought.

Age difference in depressive trajectories suggested that the oldest-old, referring to people aged 90 or above in this study, were at a much higher risk of having persistent or increasing depressive symptoms, even after adjusting for baseline symptom severity and other covariates. Existing evidence suggests that the oldest-old is a subgroup with the highest depression rates among older adults, possibly due to a higher proportion of women, more functional limitations, higher cognitive impairment, lower SES, and more limited social networks and social support (Blazer, 2003). The prevalence rates of depressive disorders in adults aged 90 or above were 30% to 50% higher than those aged 75 to 79 (Luppa et al., 2012) and twice as high as those aged 85 to 89 in a population-based study in Sweden (Bergdahl et al., 2005). This study adds to previous research by showing that depression had a more long-term and increasing course in the oldest-old, emphasizing the high burden of depression in this understudied but growing population (Haigh et al., 2018). The old-old, referring to people aged 85 to 89 years, were more likely to be on the declining trajectory than the young-old (65-59 years) after adjusting for covariates. This may partially reflect the higher level of depressive symptoms in the old-old than the young-old as reported in previous research (Luppa et al., 2012) despite adjusting for baseline symptom severity, because persons on the declining trajectory on average had more depressive symptoms than those on the persistently low trajectory. However, the exact mechanisms underlying these age differences are unclear and warrant future research, as age difference in the presentation of depressive symptoms is debated and not well understood, particularly in adults aged 85 and older (Haigh et al., 2018).

Depressive trajectories were also shaped by race/ethnicity such that non-Hispanic Blacks, compared to their White counterparts, were more likely to be on the three trajectories characterized by medium to high depression than on the persistently low symptoms trajectory. This finding echoes findings from Liang et al. (2011), where African Americans were more likely than White Americans to be in the low, moderate, decreasing and increasing trajectory classes relative to the low/no depressive symptoms class. Another study reported that older African Americans had 60% more depressive symptoms than older White Americans at baseline, and the Black–White difference increased slightly over time (Skarupski et al., 2005). Liang et al. (2011) suggested the Black–White difference in depressive trajectories were likely a result of intrapersonal changes and interpersonal differences in SES and health. Uneven access to treatment may be another reason. In a nationally representative sample of older Medicare beneficiaries, African Americans were half as likely as non-Hispanic Whites to receive a depression diagnosis from a health care provider; those with a depression diagnosis were half as likely to be treated (Akincigil et al., 2012). Further research elucidating the mechanisms underlying the racial/ethnic difference in depressive trajectories in older adults is warranted.

Consistent with research synthesis from a systematic review (Musliner et al., 2016), lower education and income were significant predictors of trajectories with greater symptom burden. The SES-health gradient has been frequently observed across time periods and age groups. Compelling evidence on SES inequality related to the prevalence and incidence of depression has also been presented in meta-analysis (Lorant et al., 2003). Possible mechanisms underlying socioeconomic inequality in late-life depression include psychosocial resources (e.g., coping style, social support, and stress), physical health, health behaviors, access to treatment, and neighborhood-level factors (Koster et al., 2006; Muramatsu, 2003). Findings from this study added to these studies by showing that SES shaped the long-term trajectories of depressive symptoms over a long period among older adults.

This study contributed to the literature by revealing potential mechanisms underlying the sex difference in depressive trajectories. Contrary to research that identified female sex as a predictor of depressive trajectories with greater symptom burden (Musliner et al., 2016), this study found that men had a higher risk being on the persistently high symptoms trajectory than women after adjusting for covariates. Additional regression analyses were conducted to better understand this discrepancy by removing one covariate at a time. These additional analyses suggested suppression effects: the sex difference in depressive trajectories disappeared when functional limitations and baseline PHQ-4 scores were both removed from the regression model and re-emerged when either variable was added back. In this regard, functional limitations and baseline symptom severity are suppressor variables—a variable that magnifies the regression coefficient of another variable when included in a regression equation (MacKinnon, Krull, & Lockwood, 2000). Another way to interpret this finding is within the context of inconsistent mediation, where a suppression effect occurs when the direct and indirect effects have opposite signs (MacKinnon et al., 2000). Further analyses identified inconsistent mediation in this study: male sex was negatively associated with functional limitations and baseline PHQ-4 scores, both of which were positively associated with being persistently high symptom, producing a negative indirect effect; whereas, the direct effect of male sex was positive, as indicated by an OR >1 (Table 3). In the multinomial regression model including only the basic demographic variables (age, sex, and race), men, compared to women, tended to have a lower risk of being on the three depressive trajectories than the persistently low symptoms trajectory. In this regard, this study did not contradict previous findings on the gendered trajectories of depressive symptoms in older adults (Carrière et al., 2017; Hsu, 2012) but rather revealed the mechanisms underlying the gender difference. Future studies should use different data sources to confirm the findings of this study. Studies aimed at further exploring gender difference in depressive trajectories should be aware of the potential suppression effects of functional limitations and baseline symptoms severity in model specification and interpretation of findings.

In addition, this study contributed to the literature on depressive trajectories by examining the role of social isolation, which emerged as an important risk factor of depressive trajectories with mid to high burden of depression. Social isolation and limited social support has been consistently linked to depression and depressive symptoms (Cacioppo, Hawkley, & Thisted, 2010; Golden et al., 2009; Teo, Choi, & Valenstein, 2013). Although studies that specifically examined the role of social isolation on depressive trajectories among older adults were lacking, several studies that examined related concepts including social network, social support, and social participation produced converging results (Byers et al., 2012; Hsu, 2012; Kuchibhatla et al., 2012; Park, 2017; Rote et al., 2015). In a 20-year population-based study of older women, a small social network, assessed using the Lubben Social Network Scale (Lubben, 1988), was associated with three to six times the risk of increasing or persistently high depressive symptoms (Byers et al., 2012). In another study, based in North Carolina, a larger social network was associated with higher odds of decreasing symptoms and lower odds of high depressive symptoms (Kuchibhatla et al., 2012). In this study, a low cutoff of 2 was applied to create the binary indicator of social isolation, capturing both somewhat isolated and very isolated older adults (Pohl et al., 2017). In this regard, this study suggested that even partial social isolation could shape the long-term trajectories of depressive symptoms, highlighting the importance of addressing the public health issue of social isolation. Future studies that investigate the dual trajectories of social isolation and depressive symptoms will further our understanding regarding their relationship.

Finally, this study expanded the literature by suggesting that physical illness, functional and mobility limitations, and cognitive impairment may subtly play different roles in shaping depressive trajectories. Physical illness, functional limitations, and mobility limitations have been associated with depressive trajectories with moderate to high depression (Byers et al., 2012; Carrière et al., 2017; Liang et al., 2011; Montagnier et al., 2014; Rote et al., 2015; Tampubolon & Maharani, 2017). Cognitive impairment has also been shown to influence depressive trajectories in some studies (Holmes et al., 2018; Tampubolon & Maharani, 2017). In this study, not only did physical illness and ADL/IADL limitations increase the risk of having persistently high symptoms versus low symptoms, they also lowered the chance of recovery from moderate to high baseline symptoms, as they were less likely to be on the declining trajectory than the persistently high trajectory. Mobility limitations and dementia, on the other hand, increased the risk of rapidly increasing symptoms versus persistently low symptoms, while also exerting little influence on the declining and persistently high trajectories. Future studies should further clarify the role of different types of impairment and illnesses in shaping depressive trajectories. If the patterns found in this study hold, depression prevention efforts should target older adults with mobility limitations and dementia who present with low depressive symptoms, as they are at an elevated risk of developing clinically significant symptoms over time. Treatment efforts to reduce the burden of late-life depression should focus on improving access to and use of evidence-based treatments among older adults with functional limitations and long-term diseases, as they are at a higher risk of experiencing persistent symptoms and are less likely to achieve remission naturally.

This study has several limitations. NHATS used proxies when sample persons were unavailable. This included answering questions on the PHQ-4 to rate symptoms of the sample person. As shown in a previous study, proxies tend to report more dysfunction symptoms than self-responders (Williams et al., 2006), which could overestimate prevalence of depression, cognitive impairment, and functional limitations. Nevertheless, an indicator of proxy response was included in the multinomial logistic regression model, a method recommended to adjust for proxy bias (Skolarus et al., 2010). Moreover, despite having good psychometric properties, PHQ-2 is not a diagnostic test, is prone to ceiling effects, and requires follow-up with PHQ-9 to further probe depression severity. In addition, a low cutoff of 2 was applied to create the binary indicator of social isolation (Pohl et al., 2017). Although this low cutoff helped distinguish depressive trajectories, this cutoff produced a higher rate of social isolation compared to results from previous studies (Cudjoe et al., 2018; Valtorta & Hanratty, 2012). Furthermore, simple counts of self-reported difficulties with activities in ADL, IADL, and mobility domains were included in the regression model. This approach does not capture the hierarchies of activity limitations nor does it consider the complex interactions among the activity domains and how these hierarchies and interactions affect depressive trajectories.

Conclusion

Changes in clinically significant depressive symptoms among community-dwelling older Americans follow four distinct trajectories over a 7-year period. The number of older adults with a persistently high burden of depression may be larger than previously thought. Demographics, SES, social contact, illness, and functioning all play a role in shaping depressive trajectories. Older adults with mobility limitations and dementia who present low depressive symptoms are prime candidates for depression prevention efforts. Additional efforts for reducing the burden of depression should focus on improving access to and use of evidence-based treatments among the oldest-old, socially isolated older adults, and older adults with functional limitations, and long-term physical illnesses. Future research is needed to elucidate the mechanisms underlying the differences in depressive trajectories: age, race/ethnicity, and gender. Future research should also further examine the dynamic relationship between social isolation and long-term depression.

Footnotes

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by a grant from the National Institutes of Health, University of Michigan Older Americans Independence Center Research Education Core (grant no.: AG024824).

ORCID iD

Xiaoling Xiang

References

Akincigil

Olfson

Siegel

Zurlo

K. A.

Walkup

J. T.

Crystal

(2012). Racial and ethnic disparities in depression care in community-dwelling elderly in the United States. American Journal of Public Health, 102, 319-328. doi:10.2105/AJPH.2011.300349

Andreescu

Chang

C. C. H.

Mulsant

B. H.

Ganguli

(2008). Twelve-year depressive symptom trajectories and their predictors in a community sample of older adults. International Psychogeriatrics, 20, 221-236. doi:10.1017/S1041610207006667

Bergdahl

Gustavsson

J. M. C.

Kallin

Wågert

P. V. H.

Lundman

Bucht

Gustafson

(2005). Depression among the oldest old: The Umeå 85+ study. International Psychogeriatrics, 17, 557-575. doi:10.1017/S1041610205002267

Blazer

D. G.

(2003). Depression in late life: Review and commentary. The Journals of Gerontology: Series A, Biological Sciences and Medical Sciences, 58, 249-265. doi:10.1093/gerona/58.3.M249

Byers

A. L.

Vittinghoff

Lui

L. Y.

Hoang

Blazer

D. G.

Covinsky

K. E.

. . . Yaffe

(2012). Twenty-year depressive trajectories among older women. Archives of General Psychiatry, 69, 1073-1079. doi:10.1001/archgenpsychiatry.2012.43

Cacioppo

J. T.

Hawkley

L. C.

Thisted

R. A.

(2010). Perceived social isolation makes me sad: 5-year cross-lagged analyses of loneliness and depressive symptomatology in the Chicago health, aging, and social relations study. Psychology and Aging, 25, 453-463. doi:10.1037/a0017216

Carrière

Farré

Proust-Lima

Ryan

Ancelin

M. L.

Ritchie

(2017). Chronic and remitting trajectories of depressive symptoms in the elderly. characterisation and risk factors. Epidemiology and Psychiatric Sciences, 26, 146-156. doi:10.1017/S2045796015001122

Cudjoe

T. K. M.

Roth

D. L.

Szanton

S. L.

Wolff

J. L.

Boyd

C. M.

Thorpe

R. J.

(2018). The epidemiology of social isolation: National Health and Aging Trends Study. The Journals of Gerontology: Series B, Psychological Sciences and Social Sciences. doi:10.1093/geronb/gby037

Cui

Lyness

J. M.

Tang

Conwell

(2008). Outcomes and predictors of late-life depression trajectories in older primary care patients. American Journal of Geriatric Psychiatry, 16, 406-415. doi:10.1097/01.JGP.0000308881.22956.27

10.

Golden

Conroy

R. M.

Bruce

Denihan

Greene

Kirby

Lawlor

B. A.

(2009). Loneliness, social support networks, mood and wellbeing in community-dwelling elderly. International Journal of Geriatric Psychiatry, 24, 694-700. doi:10.1002/gps.2181

11.

Haigh

E. A. P.

Bogucki

O. E.

Sigmon

S. T.

Blazer

D. G.

(2018). Depression among older adults: A 20-year update on five common myths and misconceptions. American Journal of Geriatric Psychiatry, 26, 107-122. doi:10.1016/j.jagp.2017.06.011

12.

Haviland

A. M.

Jones

B. L.

Nagin

D. S.

(2011). Group-based trajectory modeling extended to account for nonrandom participant attrition. Sociological Methods and Research, 40, 367-390. doi:10.1177/0049124111400041

13.

Holmes

S. E.

Esterlis

Mazure

C. M.

Lim

Y. Y.

Ames

Rainey-Smith

. . . Maruff

(2018). Trajectories of depressive and anxiety symptoms in older adults: A 6-year prospective cohort study. International Journal of Geriatric Psychiatry, 33, 405-413. doi:10.1002/gps.4761

14.

Hsu

H. C.

(2012). Group-based trajectories of depressive symptoms and the predictors in the older population. International Journal of Geriatric Psychiatry, 27, 854-862. doi:10.1002/gps.2796

15.

Jones

B. L.

Nagin

D. S.

(2007). Advances in group-based trajectory modeling and an SAS procedure for estimating them. Sociological Methods and Research, 35, 542-571. doi:10.1177/0049124106292364

16.

Kasper

J. D.

Freedman

V. A.

Spillman

B. C

. (2013). Classification of persons by dementia status in the National Health and Aging Trends Study. Technical Paper, 5. Baltimore: Johns Hopkins University School of Public Health. Retrieved from www.NHATS.org.

17.

Koster

Bosma

Kempen

G. I. J. M.

Penninx

B. W. J. H.

Beekman

A. T. F.

Deeg

D. J. H.

van Eijk

J. T. M.

(2006). Socioeconomic differences in incident depression in older adults: The role of psychosocial factors, physical health status, and behavioral factors. Journal of Psychosomatic Research, 61, 619-627. doi:10.1016/j.jpsychores.2006.05.009

18.

Kroenke

Spitzer

R. L.

Williams

J. B. W.

Lowe

(2009). An ultra-brief screening scale for anxiety and depression: The PHQ-4. Psychosomatics, 50, 613-621. doi:10.1016/S0033-3182(09)70864-3

19.

Kuchibhatla

M. N.

Fillenbaum

G. G.

Hybels

C. F.

Blazer

D. G.

(2012). Trajectory classes of depressive symptoms in a community sample of older adults. Acta Psychiatrica Scandinavica, 125, 492-501. doi:10.1111/j.1600-0447.2011.01801.x

20.

Laborde-Lahoz

El-Gabalawy

Kinley

Kirwin

P. D.

Sareen

Pietrzak

R. H.

(2015). Subsyndromal depression among older adults in the USA: Prevalence, comorbidity, and risk for new-onset psychiatric disorders in late life. International Journal of Geriatric Psychiatry, 30, 677-685. doi:10.1002/gps.4204

21.

Liang

Quiñones

A. R.

Bennett

J. M.

(2011). Multiple trajectories of depressive symptoms in middle and late life: Racial/ethnic variations. Psychology and Aging, 26, 761-777. doi:10.1037/a0023945

22.

Lorant

Deliège

Eaton

Robert

Philippot

Ansseau

(2003). Socioeconomic inequalities in depression: A meta-analysis. American Journal of Epidemiology, 157, 98-112. doi:10.1093/aje/kwf182

23.

Löwe

Kroenke

Gräfe

(2005). Detecting and monitoring depression with a two-item questionnaire (PHQ-2). Journal of Psychosomatic Research, 58, 163-171. doi:10.1016/j.jpsychores.2004.09.006

24.

Lubben

J. E.

(1988). Assessing social networks among elderly populations. Family and Community Health, 11(3), 42-52. doi:10.1097/00003727-198811000-00008

25.

Luppa

Sikorski

Luck

Ehreke

Konnopka

Wiese

. . . Riedel-Heller

S. G.

(2012).Age- and gender-specific prevalence of depression in latest-life—Systematic review and meta-analysis. Journal of Affective Disorders, 136, 212-221. doi:10.1016/j.jad.2010.11.033

26.

MacKinnon

D. P.

Krull

J. L.

Lockwood

C. M.

(2000). Equivalence of the mediation, confounding and suppression effect. Prevention Science, 1, 173-181. doi:10.1023/A:1026595011371

27.

Mohebbi

Agustini

Woods

R. L.

McNeil

J. J.

Nelson

M. R.

Shah

R. C.

. . . Berk

(2019). Prevalence of depressive symptoms and its associated factors among healthy community-dwelling older adults living in Australia and the United States. International Journal of Geriatric Psychiatry. Advance online publication. doi:10.1002/gps.5119

28.

Montagnier

Dartigues

J. F.

Rouillon

Pérès

Falissard

Onen

(2014). Ageing and trajectories of depressive symptoms in community-dwelling men and women. International Journal of Geriatric Psychiatry, 29, 720-729. doi:10.1002/gps.4054

29.

Muramatsu

(2003). County-level income inequality and depression among older Americans. Health Services Research, 38, 1863-1884. doi:10.1111/j.1475-6773.2003.00206.x

30.

Musliner

K. L.

Munk-Olsen

Eaton

W. W.

Zandi

P. P.

(2016). Heterogeneity in long-term trajectories of depressive symptoms: Patterns, predictors and outcomes. Journal of Affective Disorders, 192, 199-211. doi:10.1016/j.jad.2015.12.030

31.

Nagin

(2005). Group-based modeling of development. Cambridge, MA: Harvard University Press.

32.

Park

M. J.

(2017). Impact of social capital on depression trajectories of older women in Korea. Aging and Mental Health, 21, 354-361. doi:10.1080/13607863.2015.1088511

33.

Pohl

J. S.

Cochrane

B. B.

Schepp

K. G.

Woods

N. F.

(2017). Measuring social isolation in the National Health and Aging Trends Study. Research in Gerontological Nursing, 10, 277-287. doi:10.3928/19404921-20171002-01

34.

Rote

Chen

N. W.

Markides

(2015). Trajectories of depressive symptoms in elderly Mexican Americans. Journal of the American Geriatrics Society, 63, 1324-1330. doi:10.1111/jgs.13480

35.

Saeed Mirza

Ikram

M. A.

Freak-Poli

Hofman

Rizopoulos

Tiemeier

. (2018). 12 year trajectories of depressive symptoms in community-dwelling older adults and the subsequent risk of death over 13 years. Journals of Gerontology: Series A, Biological Sciences and Medical Sciences, 73, 820-827. doi:10.1093/gerona/glx215

36.

Schulz

Beach

S. R.

Ives

D. G.

Martire

L. M.

Ariyo

A. A.

Kop

W. J.

(2000). Association between depression and mortality in older adults: The Cardiovascular Health Study. Archives of Internal Medicine, 160, 1761-1768. doi:10.1001/archinte.160.12.1761

37.

Seitz

Purandare

Conn

(2010). Prevalence of psychiatric disorders among older adults in long-term care homes: A systematic review. International Psychogeriatrics, 22, 1025-1039. doi:10.1017/S1041610210000608

38.

Sjöberg

Karlsson

Atti

A. R.

Skoog

Fratiglioni

Wang

H. X.

(2017). Prevalence of depression: Comparisons of different depression definitions in population-based samples of older adults. Journal of Affective Disorders, 221, 123-131. doi:10.1016/j.jad.2017.06.011

39.

Skarupski

K. A.

Mendes De Leon

C. F.

Bienias

J. L.

Barnes

L. L.

Everson-Rose

S. A.

Wilson

R. S.

Evans

D. A.

(2005). Black-white differences in depressive symptoms among older adults over time. Journals of Gerontology: Series B, Psychological Sciences and Social Sciences, 60, 136-142. doi:10.1093/geronb/60.3.P136

40.

Skolarus

L. E.

Sánchez

B. N.

Morgenstern

L. B.

Garcia

N. M.

Smith

M. A.

Brown

D. L.

Lisabeth

L. D.

(2010). Validity of proxies and correction for proxy use when evaluating social determinants of health in stroke patients. Stroke, 41, 510-515. doi:10.1161/STROKEAHA.109.571703

41.

Tampubolon

Maharani

(2017). When did old age stop being depressing? Depression trajectories of older Americans and Britons 2002–2012. American Journal of Geriatric Psychiatry, 25, 1187-1195. doi:10.1016/j.jagp.2017.06.006

42.

Teo

A. R.

Choi

H. J.

Valenstein

(2013). Social relationships and depression: Ten-year follow-up from a nationally representative study. PLoS ONE, 8, e62396. doi:10.1371/journal.pone.0062396

43.

Valtorta

Hanratty

(2012). Loneliness, isolation and the health of older adults: Do we need a new research agenda? Journal of the Royal Society of Medicine, 105, 518-522. doi:10.1258/jrsm.2012.120128

44.

Wells

K. B.

Burnam

M. A.

(1990). Caring for depression in America: Lessons learned from early findings of the medical outcomes study. General Hospital Psychiatric, 13, Article 361. doi:10.1016/0163-8343(91)90056-3

45.

Williams

L. S.

Bakas

Brizendine

Plue

Hendrie

Kroenke

(2006). How valid are family proxy assessments of stroke patients’ health-related quality of life? Stroke, 37, 2081-2085. doi:10.1161/01.STR.0000230583.10311.9f

46.

Zimmer

Martin

L. G.

Nagin

D. S.

Jones

B. L.

(2012). Modeling disability trajectories and mortality of the oldest-old in China. Demography, 49, 291-314. doi:10.1007/s13524-011-0075-7