Heterogeneous depressive symptom trajectories among women with type 2 diabetes: findings from the Women’s Interagency HIV Study

Abstract

Background

Depression affects 33% of women with type 2 diabetes (T2D) and leads to increased risks of premature mortality. Fluctuation and variation of depressive presentations can hinder clinical identification.

Purpose

We aimed to identify and examine subgroups characterized by distinct depressive symptom trajectories among women with T2D.

Methods

This retrospective analysis leveraged the Women’s Interagency HIV Study data to identify depressive symptom trajectories based on the Center for Epidemiological Studies Depression scores (2014-2019) among women with and without HIV. Descriptive statistics characterized sample demographics (eg, age, race, income), clinical indices (eg, hemoglobin A1C [HbA1c], BMI, HIV status), and psychosocial experiences (eg, discrimination, social support, anxiety, pain). We used growth mixture modeling to identify groups defined by distinct depressive symptom trajectories and parametric and non-parametric tests to examine demographic, clinical, and psychosocial differences across subgroups.

Results

Among the 630 women included, the mean age was 50.4 (SD = 8.3) years, 72.4% identified as Black and non-Hispanic, and 68.2% were living with HIV. Five subgroups were identified and distinguished by severity and symptom type. Participants with lower incomes (P = .01), lower employment (P < .0001), lower social support (P = .0001), and experiences of discrimination (P < .0001) showed greater membership in threshold, moderate, and severe depressive subgroups. Subgroup membership was not associated with metabolic indices (BMI, HbA1c) or HIV status. Anxiety, pain, and loneliness (all P = .0001) were worse in subgroups with higher depressive symptoms.

Conclusions

Among women with T2D, depressive symptom trajectories differ across clinical and social contexts. This study advances precision by delineating subgroups within a broad clinical category.

A machine-learning approach identified unique patterns of depression among women with diabetes, revealing how factors like income, social support, and discrimination are connected to depressive symptoms and trajectories.

Lay Summary

This study looked at how depression affects women with type 2 diabetes (T2D), especially those living with HIV. Depression is widespread among women with T2D and HIV, and it often makes these chronic conditions worse. However, depression can look very different from person to person, making early identification a challenge for healthcare providers. Researchers found women with T2D may experience different kinds and courses of depressive symptoms over time. By analyzing existing data from a large ongoing HIV study, they identified 5 distinct groups of women based on their depression severity and the type of symptoms they reported. Factors like lower income, less social support, and experiences of discrimination were linked to more severe depression groups, while other health indicators, like body mass index or HIV status, did not show a clear relationship. The findings highlight the importance of understanding mental health in various social and clinical contexts and the need to disentangle depression, which can be an overly broad term, to develop more personalized and effective therapies.

Introduction

Type 2 diabetes (T2D) is a leading contributor to early morbidity, mortality, and increased health costs due to complications including end-stage kidney disease, cardiovascular disease, and stroke.¹ Women and minoritized racial and ethnic groups are disproportionally affected by T2D—bearing higher T2D prevalence and 20%-40% higher rates of complications.^2,3 Like T2D, depression is a substantial yet poorly addressed public health burden that disproportionally affects women and is a leading cause of disability and reduced quality of life.⁴ Moreover, depression affects 33% of women with T2D, increasing their risk of diabetes-related mortality by up to 50%.^5,6 However, depression is under-recognized and therefore, under-addressed in 66% of patients with T2D and among women of color (WOC).^5,7

Depression is known for its heterogeneity, with over 1500 symptom combinations that meet the criteria for a depressive disorder.⁸ It has been argued that depression may be an overly broad clinical category in need of greater precision, as it may incorrectly lump distinct subpopulations with varied responses to treatment under the same diagnostic label.⁹ Depressive symptoms span affective, cognitive, somatic, and interpersonal domains and exist along a continuum of severity imparting significant health and economic consequences even when criteria for major depressive disorder (MDD) are not met and are therefore considered subclinical.¹⁰ In light of the impact of subclinical symptoms, we retain a diagnosis-agnostic perspective in referencing depression, inclusive of depressive disorders and subclinical presentations. Combined with the lack of diagnostic biomarkers, depression heterogeneity can contribute to its poor recognition, particularly in populations in which depression symptom profiles have been understudied (WOC and persons with comorbidities).¹¹ Closing this scientific gap will enable providers to better identify depressive symptoms and seize opportunities to mitigate depression’s harmful effects by offering appropriately tailored interventions.

In the context of T2D, co-occurrence of depressive symptoms is both common and consequential, as evidenced by diminished glycemic control, lower social and occupational functioning, and reduced quality of life among those with a higher burden of depression symptoms; as well as increased health costs, diabetes complications, and early mortality.⁶ T2D and depressive symptoms are known to increase the risk for and exacerbate one another through several biobehavioral pathways.¹² Self-care activities (eg, healthy diet, physical activity) are often limited by the presence of depressive symptoms, which may be worsened by demanding treatment regimens and the emotional burden of living with T2D.^12,13 Beyond behavioral links, T2D, and depressive symptoms are associated with inflammation and dysregulation of the hypothalamic-pituitary-adrenal (HPA) axis which corresponds with their co-occurrence in chronic inflammatory status such as HIV (10% and 40%, respectively).^13–15 These biological mechanisms may be differentially relevant to specific depressive phenotypes. Accordingly, an immunometabolic phenotype of depression has been proposed, in light of noted associations between atypical symptom presentations (eg, anhedonia, fatigue, increased sleep, and appetite), metabolic parameters (eg, higher BMI, central adiposity, dyslipidemia), and social factors (eg, socioeconomic disadvantage and early life stress).^16,17

In addition to the inflammatory contribution of HIV to depression in T2D, the social context of living with HIV is highly relevant to the characterization of depressive phenotypes, as women with or at risk for HIV in the United States commonly experience higher exposure to poverty, material needs deprivation, discrimination, and violence.¹⁸ Although many social derterminants of health (SDoH) are known risk factors for depression, their potential differential association with distinct depression phenotypes has been less studied. Similarly, T2D-related comorbidities such as obesity and dyslipidemia and concurrent psychological symptoms and experiences (eg, anxiety, pain, loneliness) have also been associated with depression,^19–21 but there has been a lack of attention to depressive symptom composition and variability over time.

Most studies that examine depressive symptom co-occurrence in the context of T2D, HIV, or cardiometabolic conditions generally have done so cross-sectionally or using a single average trajectory for an entire study population.^5,11,22 The current study employs growth mixture modeling (GMM) to examine multiple subpopulations with distinct depressive symptom trajectories with differing social risk and clinical profiles, focusing on women, who bear greater social, mental health, and metabolic risks when compared with men.²³

We aimed to identify diagnosis-agnostic latent subgroups of women with T2D with distinct depressive symptom trajectories (inclusive of probable MDD and subclinical level symptoms) and to examine sociodemographic, clinical, and psychosocial variables across latent subgroups. We hypothesized that (1) there would be multiple trajectories differing in severity, stability, and composition of depressive symptoms corresponding to distinct subgroups and (2) trajectory subgroups would vary according to sociodemographic (eg, income, education) clinical (eg, hemoglobin A1C [HbA1c], HIV status), and psychosocial factors (eg, discrimination, loneliness, social support, anxiety). Specifically, we anticipated that 3-5 subgroups would emerge from the GMM and that subgroups of high depression severity would be characterized by unfavorable socioeconomic, clinical, and psychosocial factors based on prior research.^23–25

Methods

This retrospective analysis leveraged data from the Women’s Interagency HIV Study (WIHS) and was guided by an adapted conceptual framework of Bronfenbrenner’s socioecological model and the immunometabolic depression framework, described previously.²⁶ The WIHS is a multicenter prospective observational cohort study that examines the natural history of women with or at risk for HIV (enrolled in an approximate 3:1 ratio).²⁷ Since its establishment in 1994, the WIHS cohort has grown to over 2000 actively enrolled participants across centers in Atlanta, Birmingham, Brooklyn, Bronx, Raleigh, Chicago, Jackson, Los Angeles, Miami, San Francisco, and Washington, DC.²⁷ Data were collected every 6 months at study visits involving physical examinations, blood collection, and administration of surveys capturing information such as demographics, social factors, health history, and current symptoms. In 2019, the WIHS merged with the Multicenter AIDS Cohort (MACS), a natural history study of men with and at risk for HIV acquisition, to form the MACS/WIHS Combined Cohort Study (MWCCS).²⁸ In the current analysis, we included all WIHS participants with T2D and depressive symptom data. T2D was defined as fasting glucose ≥126 mg/L, HbA1C ≥6.5%, or self-reported diagnosis or diabetes medication.²⁹ Data from 2014 to 2019 (collected biannually) were used to assess the most contemporary pre-pandemic period of maximal length allowable for GMM.

Measures

Depressive symptoms

The Center for Epidemiological Studies Depression (CES-D) was used to measure the severity of depressive symptoms over the previous week. CES-D scoring ranges from 0 to 60 with a score of ≥16 indicating probable depression.³⁰ The validity and reliability of this self-reported 20-item measure are well established in women and populations with chronic illness.^30,31 In addition to the overall score, we also utilized four CES-D factor scores to examine symptom domains including negative affect (eg, depressed mood, guilt, increased crying), lack of positive affect (eg, anhedonia), somatic symptoms (eg, disruption in sleep or appetite, fatigue, psychomotor slowing, poor concentration), and interpersonal difficulties (eg, feeling disliked or disconnected from others).³⁰

Sociodemographic, clinical, and psychosocial variables

Data was leveraged from our analytic baseline, occurring in 2014. Sociodemographic data included baseline age, self-reported race and ethnicity, relationship status, education, annual household income, and current employment. Social support was assessed using the Medical Outcome Study Modified Social Support Survey (MSSS).³² The Major Experiences of Discrimination Scale assessed lifetime discrimination experiences³³ and was dichotomized (0/1 or more experiences), in line with other studies.³⁴ Health history surveys assessed medical comorbidities (eg, HIV, hepatitis C virus [HCV]) use of diabetes medication, and adherence to antiretroviral therapy (ART) for women with HIV. Physical examination included standardized blood pressure assessment and measurement of height and weight to calculate BMI. Laboratory data included HbA1C, fasting glucose, lipid panels, and CD4 cell counts. Perceived stress was assessed with the Perceived Stress Scale (PSS-10), anxiety symptoms with the Generalized Anxiety Disorder-7 (GAD-7), loneliness with the 3-item version of the R-UCLA Loneliness Scale, and quality of life with the Medical Outcome Study (MOS)-HIV tool.^35–38 Pain assessment was conducted via the Brief Pain Inventory short form (BPI-SF).³⁹ Across each psychosocial scale, higher scores indicate greater severity.

Data analysis

Descriptive statistics including mean (SD) and median (IQR), when appropriate, were calculated at the analytic baseline (2014 visit). GMM was conducted to identify latent classes (ie, subgroups of participants) with distinct and divergent depressive symptom trajectories. As a latent variable modeling approach, GMM allows for the identification of unobserved subgroups (of individuals with similar response patterns) within a sample by leveraging observed variables over time which it uses to model heterogeneous trajectories for these subgroups.⁴⁰ We used the CES-D scores from 11-time points across 2014-2019 and a summary score indicating the number of instances participants had a CES-D score over 16 as the model indicators.⁴¹ Data met the missing at random assumptions (non-monotone pattern of missingness); therefore, full information maximum likelihood (FIML) was used for estimation, reducing the likelihood of bias associated with list-wise deletion and imputation approaches.^42,43 Maximum likelihood assumptions of conditional independence for repeated measures were used. We assessed and compared models iteratively fitting C + 1 larger number of subgroups up to 10, as beyond 10 subgroups would be less clinically useful. To avoid convergence to local maxima, 100 starting values were used.

We determined the optimal number of classes based on model convergence and several fit information criteria including Bayesian Information Criteria (BIC), Akaike Information Criteria (AIC), and sample size adjusted BIC (aBIC)—in which better model fit is indicated by lower values. Entropy (>0.80) and significance (P < .05) of likelihood ratio tests (LRT) (Vuong-Lo-Mendell-Rubin LRT, Lo-Mendell-Rubin LRT, and Bootstrap LRT) were considered. Parsimony, theoretical justification, and the potential for clinical usefulness (eg, rejection of models with classes smaller than 5%) were considered in the final model selection.⁴³ Following selection, descriptive labels were assigned, and, we assessed depressive symptom composition (ie, what share of CES-D scores were attributable to negative affect, lack of positive affect, somatic symptoms, and interpersonal difficulties) within each subgroup.

Parametric and non-parametric tests examined significant differences in baseline characteristics across latent subgroups. Categorical variables (Fisher’s exact) and continuous variables (ANOVA or Kruskal–Wallis testing, when appropriate) were compared between the subgroups. This GMM analysis was conducted in Mplus version 8.7⁴⁴ and StataSE 17.⁴⁵ Results from the GMM analysis were presented per the Guidelines for Reporting on Latent Trajectory Studies (GRoLTS).⁴⁶ Significance was defined as P < .05.

Results

As displayed in Table 1, women in this analysis (n = 630) were on average 50.4 (SD = 8.3) years old, most identified as Black and non-Hispanic (72.4%), a majority were not partnered (69.3%), and three-quarters had annual household incomes under $18 000 (75%). Mean depressive symptoms scores at baseline were 13.5 (SD = 11.8). Two-thirds of the women included were living with HIV (68.2%) with a mean CD4 count of 251.5 (SD = 197.3). The average BMI was 34.8 (SD = 9.1) kg/m², HbA1c was 7.1 (SD = 1.8) mg/dL, and glucose was 139.5 (SD = 64.3) mg/dL.

Table 1.

Baseline sociodemographic, clinical, and symptom characteristics.

	M (SD) or n (%)
Sociodemographic factors
Age	50.3 (8.3)
Race and Ethnicity
Non-Hispanic Black	456 (72.3%)
Non-Hispanic-White	49 (7.7%)
Hispanic	105 (16.6%)
Asian, Native American, Pacific Islander, Multi-racial	20 (3.1%)
Partnership status
Married, cohabitating	180 (30.6%)
Single, divorced, separated, widowed	407 (69.3%)
Household income ≤ $18 000/year	437 (75.0%)
Education: some college, college degree	191 (32.0%)
Employed	153 (25.7%)
Discrimination (1 + Major Experiences of Discrimination)	238 (45.4%)
Tangible Support (MSSS Scores)a	71.8 (23.4)
Emotional Support (MSSS Scores)a	73.5 (20.4)
Clinical factors
HIV comorbidity	412 (68.2%)
HVC comorbidity	457 (75.7%)
Taking diabetes medication	306 (50.7%)
Smoking
Current	254 (42.6%)
Former	185 (31.0%)
Never	157 (26.3%)
Alcohol (drinks/week)	2.1 (5.0)
Substances (no. used)	3 (0.5)
ART adherenceb	1.6 (0.56)
CD4 countb	251.5 (197.3)
BMI	34.8 (9.1)
Glucose	139.5 (64.3)
Hemoglobin A1C	7.1 (1.8)
Triglycerides	153.9 (112.3)
Total cholesterol	181.1 (41.3)
HDL	50.9 (17.2)
LDL	100.8 (34.4)
Psychological factors and symptoms
Depressive Symptoms (CES-D Scores)	13.5 (11.8)
Anxiety (GAD-7 Scores)b	4.4 (4.6)
Loneliness (R-UCLA Loneliness Scores)b	3.7 (1.9)
Pain (BPI-SF scores)	63.8 (22.8)
Loneliness (R-UCLA Loneliness Scores)b	3.7 (1.9)
Stress (PSS score)	14.5 (7.8)
Quality of Life (MOS-HIV Score)	63.6 (18.1)

^aHigher score indicates a higher level.

^bMean score per visit was used as these measures were only collected at visits 46, 48, and 50.

Five classes were identified and labeled as euthymic, subclinical, threshold, moderate depression, and severe depression, based on their depressive symptom trajectories. The 5 trajectories appeared similar in stability but were distinguished mainly by the severity of depressive symptoms (Figure 1D). The threshold subgroup was characterized by participants with scores hovering near the clinical cut-off for probable depression (CES-D ~16), while moderate and severe trajectory subgroups appeared well above the CES-D screening cut-off at all time points. The 5-class model was selected for its (1) optimal fit indices, (2) class size adequacy, (3) clinical interpretation, and (4) parsimony. All tested models performed better (smaller AIC and BIC values) than the unconditional baseline module, (ie, 1-class model) (Table 2). Models with 2 through 8 classes demonstrated acceptable entropy (>0.80). BIC reached a nadir at the 7-class model. LRT comparing models revealed better performance for 3 and greater class models. Trajectory plots of models with three to seven classes were mapped (Figure 1).

Figure 1.

Trajectory plots of growth mixture models solutions.

Table 2.

Fit indices and likelihood ratio tests of growth mixture models.

No. of classes	AIC	BIC	aBIC	VLMR-LRT (P-value)	LMR-LRT (P-value)	BLRT (P-value)	Entropy
1	45 975	46 322	45 975
2	44 355	44 760	44 471	<.001	<.001	<.001	0.91
3	44 123	44 585	44 255	.11	.12	<.001	0.88
4	43 998	44 518	44 146	.13	.13	<.001	0.90
5	43 896	44 474	44 061	.15	.16	<.001	0.86
6	43 837	44 472	44 018	.78	.78	<.001	0.88
7	43 768	44 461	43 966	.63	.63	<.001	0.88
8	43724	44476	43939	.29	.29	<.001	0.86

Abbreviations: aBIC, sample size adjusted BIC; AIC, Akaike Information Criteria; BIC, Bayesian Information Criteria; BLRT, Bootstrap Likelihood Ratio Test; LMR-LRT, Lo-Mendell-Rubin Likelihood Ratio Test; VLMR-LRT, Vuong-Lo-Mendell-Rubin Likelihood Ratio Test.

Table 3 shows the number of instances participants in the total sample and the 5 subgroups had a CES-D score of 16 or greater, indicative of a probable depressive episode. Predictably, no participants within the euthymic subgroup ever scored above 16, but over 90% of those in the subclinical subgroup scored above 16 at least 1 out of 11 times. Among the threshold subgroup, 73% experienced between 3 and 7 occurrences of probable depression. A substantial number (88%) of the moderate depression subgroup experienced depressive instances 6 or more times, and 75% of the severe depression subgroup score was above this clinical threshold 8 or more times.

Table 3.

Depressive instances by latent class over 5 years.

No. of depressive instancesa	Class 1 Euthymic n (%)	Class 2 Subclinical n (%)	Class 3 Threshold n (%)	Class 4 Moderate n (%)	Class 5 Severe n (%)	Total sample n (%)
0	177 (100%)	13 (7.51%)	—	—	—	190 (30.16%)
1	—	91 (52.60%)	2 (1.54%)	—	1 (3.12%)	94 (14.92%)
2	—	57 (32.95%)	12 (9.23%)	1 (0.85%)	1 (3.12%)	71 (11.27%)
3	—	9 (5.20%)	24 (18.46%)	2 (1.69%)	1 (3.12%)	36 (5.71%)
4	—	3 (1.73%)	30 (23.08%)	3 (2.54%)	4 (12.50%)	40 (6.35%)
5	—	—	16 (12.31%)	8 (6.78%)	1 (3.12%)	25 (3.97%)
6	—	—	26 (20.0%)	10 (8.47%)	—	36 (5.71%)
7	—	—	12 (9.23%)	16 (13.56%)	—	28 (4.44%)
8	—	—	7 (5.38%)	18 (15.25%)	2 (6.25%)	27 (4.29%)
9	—	—	1 (0.77%)	27 (22.88%)	11 (34.38%)	39 (6.19%)
10	—	—	—	21 (17.80%)	7 (21.88%)	28 (4.44%)
11	—	—	—	12 (10.17%)	4 (12.50%)	16 (2.54%)

“—” indicates no instances. Percentages in parentheses reflect instances within each class.

^aInstances of probable depression were defined as the number of times participants had a CES-D score ≥16.

The proportion of each domain score to the total CES-D score was significantly different across subgroups (Table 4). Supplementary Figure 1 shows the symptom domain item averages for all time points for each subgroup. In the euthymic subgroup, the majority of the total CES-D score was driven by somatic symptoms, and in each subgroup of increasing severity, the somatic proportion of the score decreased. However, in all subgroups, at least 33% of the CES-D scores were comprised of somatic symptoms. Lack of positive affect symptoms was an increasingly prominent score driver starting in the subclinical subgroup, whereas, in the severe depression subgroup, negative affect symptoms eclipsed somatic symptoms as the most prominent domain. The proportion of negative affect symptoms increased from euthymic, subclinical, and threshold subgroups and was proportionally similar in the threshold, moderate depression, and severe depression subgroups. In all subgroups, interpersonal difficulty symptoms comprised less than 10% of the total score but proportions increased in each group of rising severity.

Table 4.

Baseline clinical and symptom characteristics by latent subgroup.

	Subgroup 1: Euthymic (n = 177)	Subgroup 2: Mild Symptoms (n = 173)	Subgroup3: Border Depression (n = 130)	Subgroup 4: Moderate Depression (n = 118)	Subgroup 5: Severe Depression (n = 32)	P
Clinical factors
HIV status						.35
Seronegative	56 (31.64%)	45 (26.01%)	37 (28.46%)	41 (34.75%)	13 (40.63%)
Seropositive	115 (64.97%)	121 (69.94%)	84 (64.62%)	75 (63.56%)	17 (53.13%)
HCV status						.009
HVC positive	42 (23.73%)	26 (15.03 %)	37(28.46%)	29 (24.58%)	12 (37.50%)
HCV negative	129 (72.88%)	139 (80.35%)	84 (64.42%)	87 (73.73%)	18 (56.25%)
Diabetes medication						.63
Yes	84 (47.46%)	95 (54.91%)	70 (53.85%)	58 (49.15%)	15 (46.88%)
No	93 (52.54%)	78 (45.09%)	59 (45.38%)	60 (50.85%)	17 (53.13%)
Smoking						.004
Current	65 (36.72%)	55 (31.79%)	55 (42.31%)	57 (48.31%)	22 (68.75%)
Former	55 (31.07%)	57 (32.95%)	39 (30.00%)	30 (25.42%)	4 (12.50%)
Never	48 (27.12%)	52 (30.06%)	25 (19.23%)	28 (23.73%)	4 (12.50%)
Alcohol (drinks/week)	0.09 (0.00-1.11)	0.13 (0.00-1.39)	0.11 (0.00-1.90)	0.16 (0.00-2.13)	0.25 (0.00-1.91)	.06
Substances (no. used)	0.00 (0.00-0.11)	0.00 (0.00-0.29)	0.00 (0.00-0.54)	0.09 (0.00-0.73)	0.08 (0.00-0.89)	<.001
ART adherence	1.33 (1.00-2.16)	1.42 (1.10-2.00)	1.56 (1.17-2.00)	1.70 (1.38-2.17)	1.80 (1.37-2.75)	.0002
CD4 Nadir	234.00 (89.00-344.00)	206.00 (88.00-341.00)	221.00 (112.00-356.00)	272.00 (130.00-404.00)	206.00 (101-347)	.49
BMI	34.13 (29.23-39.74)	34.88 (28.58- 41.39)	33.04 (27.18-38.93)	34.60 (27.73-42.62)	32.17 (28.07-36.91)	.59
Glucose	115.00 (98.50-159.40)	112.15 (94.40-152.75)	124.63 (96.42-163.33)	116.50 (96.83-190.00)	106.50 (97.00-150.00)	.25
Hemoglobin A1C	6.43 (5.98-7.82)	6.40 (5.88-7.58)	6.68 (5.88-8.08)	6.33 (5.88-8.33)	6.21 (5.90-7.25)	.78
Insulin	15.89 (9.03-23.72)	15.87 (9.48-23.17)	15.11 (9.52-22.06)	17.06 (8.98-27.13)	15.40 (9.00-23.80)	.77
Triglycerides	119.50 (90.00-169.00)	137.00 (90.5-183.00)	119.00 (92.00-186.50)	134.00 (98.00-177.00)	131.00(76.00-195.00)	.61
Total cholesterol	173.50 (152.00-198.0)	182.00 (158.50-214.00)	181.00 (151.00-214.00)	178.00 (154.00-208.00)	165.00(132.00-206.0)	.34
HDL	48.00 (38.00-61.00)	47.00 (39.00-62.00)	48.50 (41.50-59.00)	48.00 (39.00-60.00)	46.50 (33.00-55.00)	.75
LDL	96.00 (77.00-122.00)	105.50 (79.50-122.00)	102.50 (75.00-124.00)	98.50 (78.00-116.00)	78.00 (66.00-120.00)	.75
Symptoms Psychosocial factors
CESD total	3.00 (1.00-6.00)	3.00 (1.00-6.00)	18.00 (13.00-26.00)	23.00 (17.00-29.50)	34.50 (26.00-48.00)
Lack of positive affectb	0.12 (0.02-0.29)	0.18 (0.10-0.31)	0.21 (0.15-0.27)	0.21 (0.17-0.25)	0.22 (0.20-0.23)	<.001
Negative affectb	0.11 (0.03-0.22)	0.26 (0.19-0.32)	0.30 (0.26-0.35)	0.34 (0.32-0.37)	0.37 (0.35-0.38)	<.001
Somatic symptomsb	0.67 (0.45-0.83)	0.49 (0.39-0.59)	0.42 (0.34-0.49)	0.38 (0.33-0.43)	0.35 (0.32-0.38)	<.001
Interpersonalb	0.00 (0.00-0.03)	0.02 (0.00-0.05)	0.05 (0.02-0.08)	0.06 (0.03-0.09)	0.07 (0.04-0.08)	<.001
Paina	79.17 (63.33-90.00)	70.00 (53.33-87.08)	62.92 (50.83-80.00)	47.50 (37.50-62.92)	36.88 (26.82-50.00)	<.001
Quality of life	78.55 (69.54-86.75)	69.42 (56.35-79.18)	59.07 (50.39-70.47)	48.24 (38.34-56.87)	34.88 (26.15-40.77)	<.001
Loneliness	3.00 (2.00-3.33)	3.33 (3.00-4.33)	4.00 (3.00-5.00)	5.17 (4.00-6.33)	6.83 (3.17-7.83)	<.001
Stress	7.42 (3.50-12.00)	13.00 (8.50-16.50)	16.83 (13.75-19.50)	21.67 (18.00-24.67)	27.75 (24.00-31.00)	<.001
Anxiety	0.33 (0.00-1.67)	2.33 (0.67-4.67)	4.67 (2.00-7.67)	8.00 (5.33-11.67)	11.83 (6.33-15.67)	<.001

^aFrom SF-36, lower scores indicate worse pain.

^bDomain proportion of total CESD scores at all time points; column percentage totals may be less than 100 due to missing data. Bolded p-values indicated statitical significance.

As shown in Tables 4 and 5, significant differences in baseline sociodemographic, clinical, and psychosocial factors were identified among the 5 subgroups. Specifically, participants with incomes ≤$18 000, who were unemployed, who reported discrimination, and who reported less social support were overrepresented in the threshold, moderate depression, and severe depression subgroups. Race and ethnicity also varied significantly across the subgroups. Participants living with HCV, current smokers, and those with greater use of alcohol demonstrated higher membership in the threshold, moderate depression, and severe depression subgroups. Subgroup membership was not associated with baseline metabolic indices (BMI, HbA1c, glucose, lipids) or HIV status. Each of the co-occurring symptoms and psychosocial experiences examined (ie, pain, quality of life, stress, anxiety, and loneliness) were significantly different in terms of severity across subgroups. Among the women with HIV in this sample, subgroups with greater depressive severity reported less ART adherence, but CD4 cell count did not vary in the depressive subgroups.

Table 5.

Baseline sociodemographic and social characteristics by Latent subgroup.

	Subgroup 1: Euthymic (n = 177)	Subgroup 2: Subclinical (n = 173)	Subgroup3: Threshold (n = 130)	Subgroup 4: Moderate Depression (n = 118)	Subgroup 5: Severe Depression (n = 32)	P
Agea mean (SD)	51.96 (8.44)	49.91 (7.97)	50.14 (8.49)	49.67 (8.80)	48.66 (7.54)	.13
Race and Ethnicity, n (%)						.04
Non-Hispanic Black	135 (76.27%)	128 (73.99%)	95 (73.08%)	79 (66.95%)	19 (59.38%)
Non-Hispanic-White	5 (2.82%)	14 (8.09%)	10 (7.69%)	16 (13.56%)	4 (12.50%)
Hispanic	31 (17.51%)	24 (13.87%)	23 (17.69%)	20 (16.95%)	7 (21.88%)
Asian, Native American, Pacific Islander, or Multi-racialc	6 (3.39%)	7 (4.05%)	2 (1.54%)	3 (2.54%)	2 (6.25%)
Partnership Status, n (%)						.72
Single, divorced, separated, widowed	111 (62.71%)	111 (64.16%)	86 (66.15%)	76 (64.41%)	23 (71.88%)
Married, cohabitating	52 (29.38%)	50 (28.90%)	32 (24.62%)	39 (33.05%)	7 (21.88%)
Household income, n (%)						.01
≤$18 000/year	110 (62.15%)	115 (66.47%)	92 (70.77%)	92 (77.97%)	28 (87.50%)
≥$18 000/year	52 (29.38%)	43 (24.86%)	25 (19.23%)	23 (19.49%)	2 (6.25%)
College n (%)						.09
Yes	56 (31.64%)	55 (31.79%)	29 (22.31%)	36 (30.51%)	15 (46.88%)
No	112 (63.28%)	109 (63.01%)	90 (69.23%)	79 (66.95%)	15 (46.88%)
Employment, n (%)						<.0001
Yes	60 (33.90%)	51 (29.48%)	22 (16.92%)	18 (15.25%)	2 (6.25%)
No	108 (61.02%)	113 (65.32%)	97 (74.62%)	97 (82.20%)	28 (87.50%)
Health insurance, n (%)						.42
Yes	158 (89.27%)	154 (89.02%)	111 (85.38%)	102 (86.44%)	27 (84.38%)
No	10 (5.65%)	10 (5.78%)	8 (6.15%)	13 (11.02%)	3 (9.38%)
Discrimination, n (%)						<.0001
Yes	62 (35.03%)	52 (30.06%)	54 (41.54%)	54 (45.76%)	16 (50.00%)
No	97 (54.80%)	91 (52.60%)	44 (33.85%)	49 (41.53%)	5 (15.63%)
Tangible Supportb Mdn (IQR)	87.50 (73.85-97.58)	80.00 (64.06-93.75)	70.31 (53.12-87.5)	65.90 (51.56-79.68)	33.33 (27.6-56.25)	.0001
Emotional Supportb Mdn (IQR)	88.47 (38.75-96.22)	80.63 (67.01-91.67)	69.79 (55.47-83.33)	64.79 (49.66-78.13)	47.27 (34.07-54.43)	.0001

Analysis of performed with Fisher’s exact testing except when indicated otherwise. Bolded p-values indicate statitical signifiance.

^aANOVA.

^bKruskal–Wallis.

^cExcluded from subgroup comparison due to small group size.

Discussion

Few studies provide a detailed characterization of depressive symptom trajectories among women with T2D, especially among samples with significant representation of WOC. The present study used a GMM approach to explore change over a 5-year pre-pandemic period in CES-D scores. Among 630 women with T2D from the WIHS cohort, we identified 5 distinct and divergent trajectory subgroups. These trajectories were distinguished by severity, stability, and symptom composition—a novel contribution to the literature as depressive symptom composition over time has not been thoroughly examined. Further, descriptive analysis illuminated the diverse social, clinical, and symptom experiences of predominantly middle-aged and older adult women living with T2D according to depressive subgroup membership.

Depression in women with T2D

In line with prior conservative estimates, we identified 24% of the sample of this study in subgroups clearly above the CES-D cut-off of 16 at most time points (ie, moderate and severe depression subgroups), but the presence of threshold and subclinical subgroups suggests that depression in women with T2D is more nuanced, especially considering the frequency these participants still experienced depressive instances over the 5 year period. Our finding of 5 distinct and relatively stable trajectory subgroups is consistent with similar studies using GMM of depressive symptoms in mid-life and older adults²³ and within populations with T2D,⁴⁷ HIV,⁴⁸ or other chronic conditions⁴⁹ that identified between 3 and 5 subgroups. Differing lengths and intervals of time studied, depression symptom measures, and the characteristics of each population likely contribute to the variance of findings. Studies that explored depressive trajectories relative to a significant clinical event (eg, cancer treatment) identified more synchronous dynamic variation across subgroups.^47,49 Since it is unlikely that participants in our sample would experience increased depression at the same time points, the lack of clear dynamic variability may be a smoothing effect of the GMM, and our findings (Table 3) showing the number of instances participants within each subgroup scored above a CES-D of 16 add to a more nuanced understanding of these subgroups.

Symptom composition variation

The predominance of somatic symptoms in this sample is consistent with prior studies among populations with a high comorbidity burden and samples of Black U.S. women (72% of our sample).^11,50 The threshold subgroup was similar to subgroups with higher levels of severity in terms of pain, anxiety, stress, and quality of life—suggesting that this group may benefit from intervention yet be susceptible to missed care. Lack of positive affect symptoms were of similar proportions among threshold, moderate depression, and severe depression subgroups in this sample. These symptoms include worthlessness, hopelessness, and anhedonia—the last of which may also have particular relevance for Black women in the United States, studies of which show more endorsement of anhedonia and atypical symptoms over melancholic symptoms.¹¹

Race and discrimination

The significant variations in race and ethnicity across the latent depressive subgroup identified in this study add to the ongoing debate about the stress paradigm in mental health disparities. Specifically, non-Hispanic Black and Hispanic women comprised larger shares of the euthymic subgroups and lower shares of the severe depression groups than did non-Hispanic White women. However, the percentages were similar within the subclinical and threshold subgroups. Racial correlates of depression are largely inconsistent, with studies finding increased levels of depression among Black populations in the United States,^51,52 and others finding no association,⁵³ but comparison is often complicated by differences in the sample demographics and the long-standing precedent of over-representation of White populations. Heterogeneity within Black and Hispanic populations in the United States should be considered, as nativity status and duration of time living in a majority-White U.S. racial context is associated with depression,⁵⁴ supporting racism as the contributing factor. Our finding that discrimination was higher among subgroups of greater depressive severity is consistent with this idea and prior studies.^55,56

Social and economic factors

Consistent with our findings, a long-standing association between poverty and depression exists.^57,58 In an analysis of depression trajectories in the American Changing Lives (ACL) Study of 3617 adults, those with lower SES were more likely to be in a persistently depressed trajectory,⁵¹ and greater risk of chronicity has also been demonstrated in women with lower incomes.⁵³ Relatedly, place of birth and rural residency have also been associated with more severe depression trajectories,²⁴ suggesting the economic effects of depression are multifaceted. Our findings related to the role of social support are consistent with similar studies showing that women with lower social support were more likely to experience severe depressive symptoms over time²⁵ and add to the literature by differentiating emotional and tangible social support.

Clinical factors

Contrary to our hypothesis, baseline clinical metabolic and immunological factors did not significantly vary across the depressive subgroups. Although depressive symptoms are known to be associated with poorer metabolic health, it is possible that an effect that may have been detectable within a full metabolic distribution was masked when restricted to a sample with T2D. It is also possible that the diagnosis of T2D overshadowed the risk effects that HIV comorbidity has on depression. In fact, among men with and without HIV, the co-occurrence of T2D and depressive symptoms did not differ by HIV serostatus, consistent with our findings.²² Depressive symptom trajectories were significantly associated with HCV, and this may be related to the shared behavioral risk factors of alcohol and substance use, which also significantly varied by increasing severity across the subgroups.

HIV-related factors

Women with HIV comprised over 3 quarters of our total sample. There was no significant variation across subgroups in terms of CD4 count, in contrast to a previous study that found lower CD4 counts and a higher risk of mortality among women with HIV who had chronic and intermittent depressive symptoms.⁵⁹ The lower ART adherence across the depressive subgroups, in step with the increasing subgroup severity, underscores the importance of addressing depression among women living with HIV to improve HIV-related outcomes and is aligned with prior studies.^60,61

Strengths and limitations

This study’s strengths include using a data-driven modeling approach to identify subpopulations of depressive phenotypes, a construct known for heterogeneity. Moreover, we took a diagnosis-agnostic approach, allowing us to consider the full distribution of depressive symptoms in light of the negative health effects of subclinical symptoms. Additionally, the sample studied in this analysis had a substantial representation of women most likely to be affected by depression owing to the combination of socioeconomic disadvantage and clinical comorbidities.

Limitations of our study must also be noted. First, due to data limitations, we were unable to discern which participants had a history or current diagnosis of MDD or other mental health disorders or were currently receiving psychotropic medications and/or psychotherapy. This would have been especially informative concerning the threshold depressive subgroup because it may have shed light on potential gaps in care. Second, our analysis was not timed to an event such as the onset of T2D, developmental milestone, or historical/environmental event, and other trajectories may have been identified from this standpoint. However, our analysis is more consistent with the reality of clinical care in which assessment of depressive symptoms may begin at varying time points across the lifespan. Our reporting of the number of depressive instances over the analyzed period presented in Table 3 is consistent with the notion of depression as dynamic and episodic. Third, our study is limited by the length of follow-up in predominantly mid-life women. Depression is known to follow a U-shaped trajectory with increases expected with older age.⁶² Fourth, our comparisons of characteristics across the 5 subgroups were performed with factors measured at baseline only. Although many sociodemographic factors (eg, race, ethnicity) are stable, our results should be interpreted by considering that many behaviors and clinical factors (eg, smoking, cholesterol, pain) can vary. Finally, further analysis, such as regression or post hoc analysis would have augmented our findings by enabling us to identify subgroup predictors or examine differences between specific subgroups. As our primary aim was to robustly identify and characterize these distinct subgroups, this was beyond the scope of this paper. Replicating these depressive subgroups in future studies would provide a firm foundation for such analysis going forward.

Implications and conclusions

Our findings have several important implications for clinical practice and future research. Not all depressive symptoms in women with T2D follow the same trajectory. Further, a substantial number of women with T2D will present with subclinical or threshold-level symptoms that may be overlooked in the time limitations of a clinical encounter. Language is currently the primary tool for the diagnosis of depression, and more research is needed to identify objective biomarkers with the potential to offer diagnostic and treatment-decision support. Many atypical and somatic symptoms may not be recognized as such or may be normalized as a part of having a chronic illness. Providers should seek and seize opportunities to speak with patients about the array of depression symptoms beyond sadness and encourage self-monitoring, especially among women with or at risk for HIV. Our findings underscore the necessity of holistic assessment for psychological distress as greater anxiety, stress, and loneliness were associated with subgroups of greater depression severity. Further, the acceptability of expressing “depression” can vary by culture and patient-provider concordance,⁶³ particularly among communities of color impacted by mental health stigma and repeated breaches of trust by the health and scientific community.

Even within a demographically and clinically similar cohort of women with T2D, trajectories of depressive symptoms differ. Over a 5-year period, women with T2D in our study followed distinct depressive symptom trajectories ranging from nearly no symptoms to severe and chronic symptoms. A substantial proportion of women in this study experienced some degree of depressive symptoms, highlighting the importance of careful screening and the urgency of effective and accessible evidenced-based therapies for this population. Our study provides a clearer picture of several subpopulations of women with divergent depression experiences, a vital step toward disentangling depression and moving toward greater specificity in research and precision mental health care. More research is needed to understand how trajectories differ in populations with newly diagnosed T2D or those transitioning to older adulthood and including sex or gender-specific factors, to examine what social and clinical factors may be predictive of risk, and to ascertain what if any care is offered to those symptoms near or below a diagnostic threshold. Such information is needed to move mental health care toward greater precision and equity.

Author contributions

Nicole Beaulieu Perez (Conceptualization [lead], Data curation [lead], Formal analysis [lead], Funding acquisition [lead], Methodology [lead], Writing – original draft [lead], Writing – review & editing [lead]), Gail D’Eramo Melkus (Conceptualization [supporting], Funding acquisition [supporting], Supervision [supporting], Writing – review & editing [supporting]), Jason Fletcher (Formal analysis [supporting], Methodology [supporting], Writing – review & editing [supporting]), Kristen Allen-Watts (Writing – review & editing [supporting]), Deborah Jones (Writing – review & editing [supporting]), Laura Collins (Writing – review & editing [supporting]), Catalina Ramirez (Writing – review & editing [supporting]), Amanda Long (Writing – review & editing [supporting]), Mardge Cohen (Writing – review & editing [supporting]), Daniel Merenstein (Writing – review & editing [supporting]), Tracy Wilson (Writing – review & editing [supporting]), Anjali Sharma (Writing – review & editing [supporting]), and Brad Aouizerat (Conceptualization [supporting], Funding acquisition [supporting], Methodology [supporting], Supervision [supporting], Writing – original draft [supporting], Writing – review & editing [supporting])

Funding

Support was provided by the NYU P20 Center for Precision Health in Diverse Populations through a grant from NINR (P20NR018075). Data in this manuscript were collected by The Women’s Interagency HIV Study (WIHS), now the MACS/WIHS Combined Cohort Study (MWCCS) is sponsored by the NIH. MWCCS (Principal Investigators): Atlanta CRS (Ighovwerha Ofotokun, Anandi Sheth, and Gina Wingood), U01-HL146241; Bronx CRS (Kathryn Anastos, David Hanna, and Anjali Sharma), U01-HL146204; Brooklyn CRS (Deborah Gustafson and Tracey Wilson), U01-HL146202; Data Analysis and Coordination Center (Gypsyamber D’Souza, Stephen Gange and Elizabeth Topper), U01-HL146193; Chicago-Cook County CRS (Mardge Cohen, Audrey French, and Ryan Ross), U01-HL146245; Northern California CRS (Bradley Aouizerat, Jennifer Price, and Phyllis Tien), U01-HL146242; Los Angeles CRS (Roger Detels and Matthew Mimiaga), U01-HL146333; Metropolitan Washington CRS (Seble Kassaye and Daniel Merenstein), U01-HL146205; Miami CRS (Maria Alcaide, Margaret Fischl, and Deborah Jones), U01-HL146203; UAB-MS CRS (Mirjam-Colette Kempf, Jodie Dionne-Odom, James B. Brock, and Deborah Konkle-Parker), U01-HL146192; UNC CRS (M. Bradley Drummond and Michelle Floris-Moore), U01-HL146194. MWCCS data collection is also supported by UL1-TR000004 (UCSF CTSA), UL1-TR003098 (JHU ICTR), UL1-TR001881 (UCLA CTSI), P30-AI-050409 (Atlanta CFAR), P30-AI-073961 (Miami CFAR), P30-AI-050410 (UNC CFAR), P30-AI-027767 (UAB CFAR), P30-MH-116867 (Miami CHARM), UL1-TR001409 (DC CTSA), KL2-TR001432 (DC CTSA), and TL1-TR001431 (DC CTSA).

Conflicts of interest

We have no known conflicts of interest to disclose.

Transparency statements

This study was not formally registered. The analysis plan was not formally pre-registered; however, the analytic plan was completed and submitted to the MWCCS for formal review before the initiation of data analysis. De-identified data from this study are available in a protected archive maintained by the MWCCS DACC. Data can be obtained by submitting a data request at https://statepi.jhsph.edu/mwccs/work-with-us/. The analytic code used to conduct the analysis presented in this study is not available in a public archive. They may be available by emailing the corresponding author upon reasonable request. Materials used to conduct this study are not publicly available, but all data collection forms can be viewed at https://statepi.jhsph.edu/mwccs/data-collection-forms/.