Association of 20-year longitudinal depressive symptoms with left ventricular geometry outcomes in the Coronary Artery Risk Development in Young Adults study: a role for androgens?

Abstract

Objective

Depression is a risk factor for coronary heart disease (CHD) and left ventricular hypertrophy (LVH) is a potent predictor of CHD events. Whether depression is associated with LVH has received limited investigation. This study assessed cross-sectional and 20-year longitudinal associations of depressive symptoms with LVH outcomes after accounting for important known confounders.

Methods

From 5,115 participants enrolled in 1985–86 in the Coronary Artery Risk Development in Young Adults Study, 2,533 had serial measures of depressive symptoms and subsequent echocardiography to measure normal LV geometry, concentric remodeling, and LVH. The primary exposure variable was trajectories of the Center for Epidemiologic Studies Depression (CES-D) Scale score from 1990–91 to 2010–11. Multivariable polytomous logistic regression was used to assess associations of trajectories with a composite LV geometry outcome created using echocardiogram data measured in 2010–11 and 2015–16. Sex-specific conflicting results led to exploratory models that examined potential importance of testosterone and sex hormone binding globulin (SHBG).

Results

Overall CES-D and Somatic subscale trajectories had significant associations with LVH for females only. Odds ratios for the subthreshold (mean CES-D≈14) and stable (mean CES-D≈19) groups were 1.49 (95% CI: 1.05–2.13) and 1.88 (95% CI: 1.16–3.04), respectively. For females, SHBG was inversely associated with LVH and for males, bioavailable testosterone was positively associated with concentric geometry.

Conclusions

Findings from cross-sectional and longitudinal regression models for females, but not males, and particularly for Somatic subscale trajectories suggested a plausible link among depression, androgens, and LVH. The role of androgens to the depression – LVH relation requires additional investigation in future studies.

INTRODUCTION

Depression is a risk factor for incident coronary heart disease (CHD). This statement is supported by several meta-analyses published in the early 21^st century (reviewed in Carney and Freedland(1)), by a 2018 position paper from the European Society of Cardiology(2), and by pooled analysis of more than 560,000 participants from the Emerging Risk Factors Collaboration and the UK Biobank(3). Depression also co-occurs with hypertension(4) and there appear to be mechanistic pathways, possibly bidirectional, between the two conditions(5–7). Left ventricular hypertrophy (LVH) is largely a pathological consequence of hypertension(8, 9) as well as other factors (i.e., excess adiposity, smoking, aging)(10) and it is a strong predictor of CHD. A few cross-sectional studies since 1990 have examined the association of depression with LVH(11–13), finding positive associations with LV concentric remodeling or concentric hypertrophy(12), as well as with subtle LV structural and functional changes(13). The authors of one study(11) could not exclude the possibility that depression could cause LVH directly, and not necessarily through hypertension.

Currently there appear to be no longitudinal studies examining the association of depression with adverse LV geometry, including LVH. Importantly, longitudinal studies that have observed temporal changes in adverse LV geometry(14) motivate further research. Because LV hypertrophy is a potent risk factor for CHD, whether depression is a risk factor for LVH, and potentially a modifiable one, is of interest. One aim of the present study is to assess the association of longitudinal depressive symptoms with LV geometry after accounting for known risk factors for abnormal LV geometry, including blood pressure – the most important driver - and body mass index (BMI). Because previous studies have observed a greater prevalence of LVH in females compared to males(15) and there are sex differences in the prevalence and pathophysiology of depression that are potentially accounted for by sex differences in depression-related gene expression, neuroplasticity, and immune dysregulation(16) we also examine whether the association differs by biological sex. We hypothesized that longitudinal patterns of elevated depressive symptoms would be associated with adverse LV geometry outcomes after 20 years of follow-up; and also that these associations would be present in both males and females. Some initial results that were inconsistent with the hypotheses led to additional exploratory analyses examining the potential importance of sex hormones in the association of depressive symptoms with LV geometry. Sex-related LV remodeling is driven by sex hormones(17). Sex hormones have also been investigated for a possible causal relation with depression in males and females(18–27). The presumed directions of these associations (i.e., sex hormones -> LV remodeling and sex hormones -> depression) would imply sex hormones may be a confounder as opposed to a mediator, which would require depression -> sex hormones and sex hormones -> LV remodeling.

METHODS

The CARDIA Study has provided NHLBI Data Repository Datasets at https://biolincc.nhlbi.nih.gov/home/.

Study participants

In 1985–1986 (Year 0), the Coronary Artery Risk Development in (Young) Adults (CARDIA) study recruited 5,115 Black or White males and females free of clinical CVD at baseline, 18 to 30 years of age, from four field centers. Follow-up examinations were completed in 1987–1988 (Year 2), 1990–1991 (Year 5), 1992–1993 (Year 7), 1995–1996 (Year 10), 2000–2001 (Year 15), 2005–2006 (Year 20), 2010–2011 (Year 25), and 2015–2016 (Year 30) after baseline (1985–1986: Year 0). Retention rates among surviving participants at each in-person examination were 91%, 86%, 81%, 79%, 74%, 72%, 72%, and 71%, respectively. Contact is maintained with participants via telephone, mail, or email every 6 months, with annual interim medical history ascertainment. Over the last 5 years, >90% of the surviving cohort members have been directly contacted, and follow-up for vital status is virtually complete through related contacts and intermittent National Death Index searches. The study was approved by institutional review boards at all sites, and participants gave informed consent.

This study uses the same sample of 2,833 participants as a prior publication(28) from the CARDIA study. Derivation of the analytic sample is described therein. The 2,833 CARDIA participants attended both the Year 5 (1990–91, mean ages 23–35 years) and Year 30 (2015–16, mean ages 58–60 years) examinations, and completed echocardiograms at both visits. For participants who also attended the Year 25 (2010–11, mean ages 43–55 years) examination and had an echocardiogram, their data were also included. Female participants who were pregnant at any of the three examinations were excluded. The prevalence of adverse geometry (i.e., concentric remodeling, concentric hypertrophy, or eccentric hypertrophy) in this cohort was 10.5% at age 25 years and 36.6% at age 60 years(28). In the current study we examined the cross-sectional association of Year 5 depressive symptoms with prevalent adverse LV geometry and the longitudinal association of 20-year depressive symptom trends (Year 5 to Year 25) with adverse LV geometry at Year 25 and Year 30 (Y25/30). Sample sizes for regression models ranged from 2,739 to 2,533 due to 94 participants who could not be classified on the final outcome and 206 participants missing data on particular covariates.

Echocardiographic Measures

The echocardiographic protocol at Year 5 followed American Society of Echocardiography guidelines for study acquisition and measurement(29). The protocols at Year 25 and Year 30 followed the original Year 5 protocol for data acquisition and measurement and were consistent with more current American Society of Echocardiography guidelines.(30–32) LV geometry was classified into four categories using American Society of Echocardiography criteria(31): normal geometry, concentric remodeling (without LVH), LVH with concentric remodeling, and LVH with eccentric remodeling. LVH was defined as LV mass > 51 g/m^2.7 and concentric remodeling was defined as LV relative wall thickness ≥0.42 and eccentric remodeling as < 0.42. Additional details are provided in Perak et al.(28)

Exposure variables

Depressive symptoms were assessed at the years 5, 10, 15, 20 and 25 examinations using the 20-item Center for Epidemiologic Studies Depression Scale (CES-D scale; maximum score of 60)(33) which served as the primary exposure for analysis. Participants were asked to indicate how often they experienced each symptom in the past week with scores for the responses ranging from 0 to 3 points (0, rarely or none of the time; 1, some of the time; 2, much of the time; 3, most or all of the time). Examples of symptoms included are depressed mood, poor or excessive appetite, trouble concentrating, disturbed sleep, depressed mood, feeling disliked, talking less than usual, and inability to “get going.” In the past dozen years, depression research has been seeking to better understand diagnostic heterogeneity and delineate diagnostic subtypes that show a stronger relation to physiologic functioning. Although the CES-D scale traditionally has four subscales (Negative Affect, Positive Affect, Somatic Features, and Interpersonal Disturbances) that were determined from factor analytic studies(33), a 3-factor model that is consistent with contemporary diagnostic criteria for depressive episodes and excludes items that are unreliable or gender biased was developed by Carleton(34). The three Carleton-based subscales are Negative affect (CES-D items 3, 6, 14, and 18), Anhedonia (CES-D items 4, 8, 12, and 16), and Somatic (CES-D items 1, 2, 5, 7, 11, and 20). Trajectory groups for the three Carleton subscales were modeled as exploratory secondary exposures for the geometry group outcomes. The CES-D 20-year cumulative exposure for each participant was computed by summing the product of the average CES-D score and the time interval (in years) between 2 consecutive examinations over 20 years. The average cumulative CES-D years was determined for each trajectory group.

In previous CARDIA studies using the latent class modeling method(35, 36), PROC TRAJ in SAS, to fit trajectory groups to CES-D scores, five-group models were found to be optimal although some groups had very small numbers. For this reason we limited the number of trajectory groups in this study to a maximum of five for the full CES-D scale and all of the subscales. Maximum likelihood was used to derive trajectory parameters that defined the shape of each trajectory. We tested models with groups ranging from 2 to 5 with cubic polynomial function parameters. Final trajectory models were selected based on successful model convergence, the Bayesian information criterion (BIC), and visual inspection of trajectory plots for parsimony. The posterior predictive probabilities were used to assign a participant to a group.

Outcome assessment

For the outcome categories in the longitudinal analysis, a hierarchical composite LV geometry classification was created using the Year 25 and Year 30 LV geometry groupings. A participant was categorized first into the LVH group if s/he was classified with eccentric or concentric LVH at either Year 25 or Year 30 examination. Next, a participant was classified as having concentric remodeling if s/he had that classification at either Year 25 or Year 30 and did not have LVH at either year. The remaining participants were classified as having normal geometry at both Year 25 and Year 30.

Covariates

Demographic, anthropometric, and lifestyle measures were obtained at the Years 5, 7, 10, 15, 20, and 25 examinations. Previously defined risk factors for adverse LV geometry are life-course cumulative burdens of body mass index (BMI) and blood pressure(9), abdominal adiposity(37), physical activity(38), early onset diabetes(39), alcohol use(40), pack-years of smoking(41). These are also known correlates of depression.

Demographic, anthropometric, and lifestyle measures obtained at the Year 25 examination were used in these analyses. Height and weight were measured with the participant wearing light clothing with no shoes, and body mass index (BMI) was computed. Age, race, sex, years of education, and smoking status were self-reported at each examination. Waist circumference was classified as “elevated” if waist circumference exceeded the following thresholds for males and females, respectively(42): ≥90 cm and ≥80 cm for BMI 18.5 – 24.9 kg/m2, ≥ 100 cm and ≥ 90 cm for 25.0 – 29.9 kg/m2, ≥ 110 cm and ≥ 105 cm for 30.0 – 34.9 kg/m2, and ≥ 125 cm and ≥ 115 cm for BMI ≥35 kg/m2. A physical activity score was obtained from the CARDIA Physical Activity History, a modified version of the Minnesota Leisure Time Physical Activity Questionnaire(43). Diabetes was classified as early diabetes if diagnosed at Year 10 or earlier and as late if diagnosed after Year 10(39). Alcohol intake (ml/d) was computed from self-reported frequency of consumption of beer, wine, and liquor per week(44).

Seated blood pressure (BP) was measured using a Hawksley random zero sphygmomanometer (Hawksley, Sussex, United Kingdom) for the Year 0 to Year 15 examinations and an Omron model HEM907XL for Year 20 to Year 30. To eliminate machine bias, a calibration study was conducted and values standardized to the sphygmomanometric measures were used for the Year 20, 25, and 30 BP measurements. The BP was measured on the right arm at three 1-minute intervals in a quiet room after a 5-minute rest; the average of the second and third measurements was used. A 30-second pulse was taken at the radial artery by palpation and counting after the BP cuff had been applied and prior to the start of BP measurement. Twenty-year cumulative exposures of BMI and SBP were computed by summing the product of the average exposure value and the time interval (in years) between 2 consecutive examinations over 20 years.

Statistical analysis

Cross-sectional and longitudinal models were used to assess associations of depressive symptoms and their trajectories with LV geometry outcomes. For the cross-sectional model, the Year 5 CES-D score was log transformed and modeled as a continuous exposure for the polytomous LV geometry groups (normal geometry, concentric remodeling, and LV hypertrophy) at Year 5. The longitudinal model used the 20-year depressive symptom trajectories as the exposure. The concentric and eccentric LVH categories were pooled for both the cross-sectional and longitudinal models due to small numbers in each category.

A sequence of regression models was run to examine the change in the odds ratios for adverse LV geometry outcomes as accounted for by demographic factors, baseline geometry, health and lifestyle factors, cumulative BMI and elevated waist circumference, and SBP. Interactions of sex with depressive symptoms were assessed by including the product of an indicator variable for sex with indicator variables for the trajectory groups.

After noting interactions of depressive symptoms with sex in associations with LV geometry outcomes, in exploratory analyses we examined whether sex hormones confound the depressive symptoms – adverse LV geometry associations. We included total testosterone, bioavailable testosterone, and sex hormone binding globulin (SHBG) one-at-a-time in the sex-restricted final models. These hormones were originally measured in the CARDIA Male Hormone Study (CMHS)(45) at the Year 2, Year 7, and Year 10 exams and in the CARDIA Women’s Study (CWS)(46) in the Year 2, Year 10, and Year 16 exams. Details regarding hormone measurements are given in Gapstur et al.(45) and Calderon-Margalit et al.(46) We used the mean of each participants sex hormone measures in these analyses to determine the statistical significance of the hormones in the final models and the “change-in-estimate” approach(47) to assess the percentage change in the estimate of the ORs for the CES-D trajectory groups. Bootstrap resampling using 1000 samples was used to estimate the median change-in-estimate and the 95% confidence limits. A 10% change-in-estimate is typically used as a threshold to decide important confounding. The hormones were assessed as log-transformed continuous variables and also categorized into quartiles. Statistical analyses were conducted using SAS for Windows, release 9.4 (SAS Institute Inc., Cary, NC, USA).

Results

Plot 1 in Figure 1 shows the five CES-D trajectory groups. We named the groups as follows: trajectory 1 is “No depression”; trajectory 2, “Subthreshold Depression”; trajectory 3, “Stable Depression”; trajectory 4, “Increasing to High Depression”; and trajectory 5, “Stable High Depression”. Other than for Group 4, CES-D scores tended to track over time and the mean±sd cumulative CES-D years of exposure for Groups 1, 2, 3, and 5 increased monotonically: 98±43, 226 ±48, 400 ±56, and 617±81 CESD-years, respectively. Group 4 had mean 329±81, falling between Groups 2 and 3. In statistical analysis models trajectory group 1, No Depression, was taken to be the referent group.

Figure 1. Scale trajectories in CARDIA study from examination Years 5 to 25.

Plot 1. CES-D scale trajectories. The individual trajectories were named No Depression (Red: n=1358; 47.9%), Subthreshold Depression (Blue: n=1071; 37.8%), Stable Depression (Green: n=290; 10.2%), Increasing to High Depression (Black: n=54; 1.9%), Stable High depression (Yellow: n=60; 2.1%).

Plot 2. CES-D Somatic subscale trajectories. The individual trajectories were named Low (Red: n=454; 16.0%), Moderately Low (Blue: n=1633; 57.6%), Moderate (Green: n=618; 21.8%), High (Black: n=128; 4.5%).

Plot 3. CES-D Anhedonia subscale trajectories. The individual trajectories were named Low (Red: n=665; 23.5%), Moderate (Blue: n=1598; 56.4%), High (Green: n=570; 20.1%).

Plot 4. CES-D Negative Affect subscale trajectories. The individual trajectories were named Low (Red: n=1383; 48.8%), Moderate (Blue: n=1322; 46.7%), High (Green: n=128; 4.5%).

CES-D = Center for Epidemiologic Studies Depression;

CARDIA = Coronary Artery Risk Development in Young Adults.

Table 1 shows Year 5 and Year 25 characteristics of the five trajectory groups for males and females. For both males and females, mean age at Year 5 was approximately 30 years in all groups, trajectory 1 had the highest mean education, trajectory 4 had the highest proportion of late onset diabetes, and trajectory 5 had the highest proportion of Black participants. Year 5 mean BMI was overweight in all 5 trajectory groups and at Year 25 all mean BMIs were obese or nearly obese for females whereas most mean BMIs for males were just under 30 kg/m² cutoff for obese.

Table 1.

Demographic and clinical characteristics of 1237 male and 1596 female CARDIA participants by CES-D trajectory group at 1990–91 and 2010–11.

	Group 1 (No depression)	Group 2 (Subthreshold depression)	Group 3 (Stable depression)	Group 4 (Increasing to high depression)	Group 5 (Stable high depression)
Males
Baseline (Y5), N	640	462	100	18	17
Age, years	30.2 (3.5)	30.2 (3.6)	29.3 (3.3)	29.4 (3.8)	30.4 (4.3)
Black race, N (%)	219 (34)	216 (47)	63 (63)	8 (44)	11 (65)
Educational attainment, years
Y5 (1990–91)	15.1 (2.5)	14.4 (2.4)	13.7 (2.4)	13.4 (3.0)	12.9 (2.8)
Y25 (2010–11)	15.6 (2.6)	14.7 (2.9)	13.9 (2.5)	13.9 (2.9)	13.2 (2.8)
Diabetes, N (%)
None	570 (89)	373 (85)	84 (84)	13 (72)	14 (82)
Late onset	56 (9)	54 (12)	13 (13)	3 (17)	2 (12)
Early onset	14 (2)	15 (3)	3 (3)	2 (11)	1 (6)
Systolic BP, mmHg
Y5	111 (10)	111 (11)	113 (13)	113 (11)	116 (10)
Y25	119 (13)	122 (15)	123 (15)	121 (8)	125 (16)
Diastolic BP, mmHg
Y5	71 (9)	71 (10)	73 (11)	71 (8)	77 (12)
Y25	74 (10)	76 (11)	76 (11)	76 (8)	75 (12)
BP medication, N (%)
Y5	5 (1)	5 (1)	1 (1)	0 (0)	0 (0)
Y25	121 (19)	124 (27)	26 (26)	6 (33)	5 (29)
Heart rate (beats /60 sec)
Y5	66 (10)	65 (10)	66 (10)	68 (7)	65 (12)
Y25	65 (10)	66 (10)	67 (12)	67 (9)	67 (10)
Body mass index, Kg/m2
Y5	26.2 (4.7)	25.8 (4.2)	25.9 (5.1)	25.1 (4.6)	25.4 (5.3)
Y25	29.4 (5.7)	29.6 (6.0)	29.0 (6.7)	30.2 (5.6)	28.0 (5.0)
Waist circumference, cm
Y5	86.7 (10.3)	86.2 (10.0)	86.1 (11.3)	85.5 (11.9)	85.6 (11.4)
Y25	98.2 (13.1)	99.2 (14.5)	98.6 (15.4)	100.3 (15.2)	96.4 (11.5)
Elevated waist, N (%)
Y5	19 (3)	13 (3)	3 (3)	2 (11)	1 (6)
Y25	108 (18)	86 (20)	25 (28)	3 (18)	4 (25)
Cigarettes/day
Y5	2.7 (6.6)	4.1 (8.0)	6.4 (10.7)	6.4 (9.0)	5.8 (11.5)
Y25	1.3 (4.4)	2.6 (6.7)	4.4 (8.2)	6.7 (9.5)	1.4 (3.4)
Alcohol intake, mL/day
Y5	12.5 (18.6)	17.8 (28.1)	25.4 (37.2)	21.4 (23.0)	16.8 (26.2)
Y25	14.5 (20.4)	18.4 (36.4)	20.4 (43.3)	19.1 (38.4)	25.0 (36.8)
Physical activity score, Activity units
Y5	496 (323)	460 (312)	440 (303)	319 (227)	362 (368)
Y25	458 (311)	362 (60)	332 (253)	422 (326)	288 (319)
Y5 LV Geometry, N (%)
Normal	537 (87)	378 (84)	72 (76)	14 (82)	12 (75)
Concentric remodeling	52 (8)	44 (10)	17 (18)	2 (12)	2 (13)
Eccentric or concentric hypertrophy	28 (5)	26 (6)	6 (6)	1 (6)	2 (13)
Mean Y2, Y7, and Y10 Androgens, N	499	359	75	14	12
Total testosterone, ng/mL	6.00 (1.60)	6.06 (1.65)	6.20 (1.70)	5.79 (1.65)	6.89 (1.22)
Bioavailable testosterone, ng/mL	3.02 (0.82)	3.04 (0.85)	3.13 (0.82)	3.07 (0.95)	3.42 (0.99)
Sex hormone binding globulin, ng/mL	30.2 (11.7)	30.0 (11.9)	30.0 (9.6)	27.8 (13.0)	33.5 (11.1)
Females
Baseline (Y5), N	718	609	190	36	43
Age, years	30.4 (3.5)	30.0 (3.6)	29.9 (3.6)	30.6 (3.7)	29.0 (4.2)
Black race, N (%)	282 (39)	307 (50)	129 (68)	18 (50)	35 (81)
Educational attainment, years
Y5 (1990–91)	15.1 (2.1)	14.5 (2.4)	13.6 (2.0)	13.8 (2.4)	12.3 (2.0)
Y25 (2010–11)	15.7 (2.5)	15.3 (2.6)	14.3 (2.5)	14.4 (2.6)	13.0 (1.9)
Diabetes, N (%)
None	642 (89)	526 (86)	161 (85)	27 (75)	35 (81)
Late onset	61 (9)	67 (11)	21 (11)	8 (22)	6 (14)
Early onset	15 (2)	16 (3)	8 (4)	1 (3)	2 (5)
Systolic BP, mmHg
Y5	103 (10)	104 (10)	105 (11)	102 (11)	107 (11)
Y25	115 (15)	116 (15)	121 (19)	113 (15)	118 (17)
Diastolic BP, mmHg
Y5	67 (10)	67 (9)	68 (10)	65 (10)	68 (10)
Y25	71 (11)	73 (11)	77 (12)	71 (11)	75 (11)
BP medication, N (%)
Y5	9 (1)	7 (1)	4 (2)	0 (0)	1 (2)
Y25	147 (20)	157 (26)	73 (38)	10 (28)	14 (33)
Heart rate (beats /60 sec)
Y5	72 (11)	71 (10)	71 (9)	71 (12)	74 (17)
Y25	65 (9)	67 (9)	68 (11)	72 (23)	68 (11)
Body mass index, Kg/m2
Y5	25.4 (6.2)	26.1 (6.4)	27.0 (6.8)	25.0 (5.7)	30.2 (7.9)
Y25	29.4 (7.5)	31.0 (8.3)	32.8 (8.7)	29.5 (7.3)	34.4 (8.7)
Waist circumference, cm
Y5	76.3 (11.6)	78.0 (12.7)	80.4 (14.4)	77.2 (12.3)	87.1 (15.1)
Y25	88.2 (15.3)	91.6 (16.5)	96.2 (18.6)	91.8 (17.3)	98.4 (16.6)
Elevated waist, N (%)
Y5	20 (3)	21 (3)	13 (7)	3 (8)	5 (12)
Y25	131 (20)	128 (23)	41 (23)	9 (26)	10 (26)
Cigarettes/day
Y5	2.0 (5.3)	3.3 (6.9)	4.3 (7.4)	3.6 (6.1)	3.8 (6.3)
Y25	1.0 (3.1)	1.5 (4.2)	2.1 (4.8)	1.5 (4.6)	4.5 (7.0)
Alcohol intake, mL/day
Y5	4.7 (9.6)	6.6 (12.1)	8.8 (37.3)	10.9 (23.52	13.1 (27.5)
Y25	7.7 (12.0)	7.7 (13.8)	8.2 (16.7)	7.9 (12.4)	10.0 (32.6)
Physical activity score, Activity units
Y5	324 (262)	321 (241)	272 (233)	197 (184)	253 (259)
Y25	322 (255)	276 (228)	215 (229)	207 (263)	228 (215)
Y5 LV Geometry, N (%)
Normal	641 (92)	517 (87)	153 (82)	34 (97)	33 (85)
Concentric remodeling	36 (5)	50 (8)	16 (9)	0 (0)	2 (5)
Eccentric or concentric hypertrophy	21 (3)	26 (4)	17 (9)	1 (3)	4 (10)
Mean Y2, Y10, and Y16 Androgens, N	570	440	145	27	31
Total testosterone, ng/dL	32.5 (20.3)	35.8 (38.6)	36.9 (30.9)	30.8 (19.8)	45.5 (20.5)
Free testosterone, ng/dL	0.23 (0.19)	0.26 (0.35)	0.26 (0.18)	0.25 (0.20)	0.36 (0.21)
Sex hormone binding globulin, nmol/liter	32.5 (12.6)	31.6 (12.4)	30.3 (12.6)	29.9 (16.1)	28.1 (10.2)

Values reported are mean (SD) or number (percent).

Y, Year

Cross-sectional analysis

In the multivariable adjusted Year 5 cross-sectional analysis, the log(CES-D) score was significantly associated with Year 5 LV geometry category. Compared to normal geometry, a one-standard deviation higher log(CES-D) score had an adjusted OR=1.25 (95%CI: 1.06, 1.47) for concentric remodeling and OR=1.28 (95%CI: 1.03, 1.58) for LV hypertrophy. The p-value for a sex interaction added to the model was 0.34 and the ORs for both males and females were all greater than 1, though the magnitudes were higher in females (ORs 95% CIs: 1.25 (1.00, 1.57) and 1.48 (1.10, 2.01) for females for concentric remodeling and LVH, respectively; ORs and 95% CIs: 1.24 (0.98, 1.58) and 1.08 (0.80, 1.47) for males.

Longitudinal analysis

In the longitudinal polytomous regression modeling of the main CES-D trajectory groups without any covariate adjustment (Model 1, Table S1, Supplemental Digital Content), trajectories 2 (Subthreshold), 3 (Stable) and 5 (Stable high) had elevated and statistically significant odds ratios (ORs), 1.31 (95%CI: 1.06, 1.61), 2.28 (95%CI: 1.68, 3.09), and 1.98 (95%CI: 1.06, 3.67), respectively, for the LVH outcome compared to normal. The addition of demographic factors to the model (Model 2) resulted in attenuation of all ORs leaving only the trajectory 3 OR for LVH outcome significant (OR=1.59, 95% CI: 1.14–2.22). The addition of baseline LV geometry groups in model 3 attenuated that OR further to 1.50 (95%CI: 1.07, 2.09). In model 4, adding health and lifestyle factors, none of the ORs was statistically significant. However, adding in cumulative BMI, elevated waist, and cumulative SBP into model 5 returned the magnitude and significance of the trajectory 3 (Stable) OR for the LVH outcome (1.49, 95%CI: 1.02, 2.19). Finally, in Model 5 some covariates were nonsignificant and so were excluded from the final model (Model 6). The excluded covariates were percent of visits taking cholesterol-lowering medications, percent of visits taking BP-lowering medications, alcohol consumption, and total physical activity. Because trajectory groups 4 and 5 were small, an additional sensitivity analysis pooling these groups was performed and found null. However, it is noted that trajectory group 4 had elevated effect estimates for its association with concentric LV geometry that were not attenuated with any adjustment, though none of these achieved statistical significance in light of the small number of participants in this group and the resultant wide CIs.

The test of interaction of sex with CES-D trajectory groups (p=0.25) was not significant at commonly accepted levels, but as noted earlier, previous studies(16, 17) provide rationale for more thorough examination.

Table 2 shows the sex-specific ORs for the final model 6 as well as for the secondary depressive symptom trajectory models. For the main CES-D trajectory group model, no ORs were statistically significant for males. For females in trajectories 2 (Subthreshold) and 3 (Stable), the ORs for LVH compared to normal LV geometry were 1.49 (95%CI: 1.05, 2.03) and 1.88 (95%CI: 1.16, 3.04), respectively. For the secondary depressive symptom trajectory groups, the interaction of sex and CES-D trajectory group had p=0.14 for the Somatic scale trajectories and p=0.076 for the Anhedonia scale trajectories. For the Somatic scale trajectories for females, trajectories 3 and 4 were significantly positively associated with LVH. Both ORs were approximately 2.1. Males had no significant associations for the Somatic trajectories. On the other hand, for males for the Anhedonia subscale trajectory analysis, the ORs for the concentric remodeling outcomes for trajectories 2 and 3 were approximately 0.6 compared to normal geometry. Females had no significant associations for the Anhedonia trajectories.

Table 2.

Tests of trajectory group × sex interaction and sex-specific polytomous logistic regression models for main and secondary CES-D trajectory groups. Trajectory 1 is the reference group.

	Trajectory 2 Outcomes			Trajectory 3 Outcomes			Trajectory 4 Outcomes			Trajectory 5 Outcomes
	Normal	Concentric geometry vs. normal	LVH vs. normal	Normal	Concentric geometry vs. normal	LVH vs. normal	Normal	Concentric geometry vs. normal	LVH vs. normal	Normal	Concentric geometry vs. normal	LVH vs. normal
	--	OR	OR	--	OR	OR	--	OR	OR	--	OR	OR
	--	(95% CI)	(95% CI)	--	(95% CI)	(95% CI)	--	(95% CI)	(95% CI)	--	(95% CI)	(95% CI)
	(N)	(N)	(N)	(N)	(N)	(N)	(N)	(N)	(N)	(N)	(N)	(N)
CES-D trajectory groups, P=0.25a
Males	--	0.80	0.72	--	0.70	1.16	--	2.28	1.15	--	0.86	0.67
	--	(0.57, 1.11)	(0.50, 1.04)	--	(0.37, 1.31)	(0.63, 2.16)	--	(0.71, 7.36)	(0.27, 4.89)	--	(0.25, 2.92)	(0.17, 2.66)
	(255)	(89)	(93)	(49)	(18)	(27)	(7)	(6)	(4)	(10)	(4)	(3)
Females	--	0.97	1.49	--	1.16	1.88	--	1.68	0.70	--	1.18	1.12
	--	(0.72, 1.31)	(1.05, 2.13)	--	(0.74, 1.82)	(1.16, 3.04)	--	(0.75, 3.77)	(0.18, 2.66)	--	(0.49, 2.85)	(0.45, 2.81)
	(329)	(121)	(142)	(81)	(37)	(64)	(19)	(11)	(3)	(18)	(9)	(14)
Somatic trajectory groups, P=0.14a
Males	--	0.96	0.73	--	0.75	0.88	-	0.84	0.43
	--	(0.64, 1.43)	(0.47, 1.12)	--	(0.44, 1.26)	(0.51,1.50)	-	(0.32, 2.22)	(0.13, 1.42)
	(415)	(151)	(151)	(118)	(45)	(57)	(20)	(8)	(6)
Females	--	1.01	1.20	--	1.27	2.07	-	1.11	2.05
	--	(0.69, 1.48)	(0.73, 1.99)	--	(0.82, 1.97)	(1.21, 3.54)	-	(0.56, 2.20)	(0.99, 4.24)
	(531)	(182)	(154)	(171)	(82)	(112)	(38)	(18)	(34)
Negative trajectory groups, P=0.74a
Males	--	0.90	0.96	--	0.76	0.80
	--	(0.65, 1.23)	(0.68, 1.35)	--	(0.31, 1.87)	(0.33, 1.98)
	(283)	(104)	(113)	(21)	(9)	(9)
Females	--	0.94	1.31	--	1.07	0.98
	--	(0.72, 1.24)	(0.94, 1.81)	--	(0.57,2.00)	(0.50, 1.95)
	(431)	(155)	(185)	(38)	(18)	(27)
Anhedonia trajectory groups, P=0.076a
Males	--	0.56	0.72	--	0.64	0.78
	--	(0.39, 0.79)	(0.49, 1.08)	--	(0.40, 1.00)	(0.47, 1.28)
	(392)	(123)	(144)	(131)	(48)	(53)
Females	--	1.03	0.89	--	1.26	1.33
	--	(0.74, 1.43)	(0.60,1.33)	--	(0.83, 1.91)	(0.83, 2.11)
	(519)	(192)	(173)	(152)	(70)	(91)

^ap-value for trajectory groupXsex interaction.

Exploratory analysis

Table S2, Supplemental Digital Content, shows the exploratory analyses for males and females separately. For the males, both the log-transformed bioavailable testosterone and the quartile categorized version were significant and indicated that high testosterone was associated with LV concentric geometry. The Akaike information criterion (AIC) is a measure of statistical model quality, and a lower absolute value indicates better fit. From best to worst, the AICs were 1548.4 , 1551.6, and 1552.0 for the model with log(bioavailable testosterone), the model with bioavailable testosterone in quartiles, and the model without bioavailable testosterone, respectively. Pearson correlations of log(bioavailable testosterone) with the separate CES-D measures from Year 5 through Year 25 were not significant. For females, it was log-transformed SHBG and the quartile categorized SHBG that were significant, indicating high SHBG associated with lower odds of LVH. The best to worst AICs were 1920.1, 1926.1, and 1927.4, respectively, for the model with SHBG quartiles, the model using log(SHBG), and the model without SHBG. Pearson correlations of log (SHBG) with the separate CES-D measures were inverse and statistically significant. Table S3, Supplemental Digital Content, shows the results using 1000 bootstrap samples to determine the median change-in-estimate in the ORs for the CES-D trajectory groups after including the sex hormones in the model. Most ORs changed less than an absolute median of 6.7%. Five had changes ranging from 7.0% to 11.4%. All of these occurred for Trajectory groups 4 or 5, the groups with worst CES-D scores at Year 25. Also, many confidence limits in Trajectory 4 and 5 were wide, frequently with both endpoints exceeding an absolute 10%.

Discussion

The cross-sectional analysis revealed positive associations of the CES-D score with Year 5 concentric remodeling and LV hypertrophy and did not show evidence of modification by sex via the interaction p-value=0.34. However, the effect size of the OR for LVH relative to normal geometry was notably greater for females than males. Of greater interest and potential importance, the associations of the 20-year depression trajectories with the Year 25/30 geometric remodeling outcomes in this study were variable and differed somewhat between males and females. In females, we found global depressive symptoms were associated with LV hypertrophy in the largest trajectory groups after adjusting for cumulative blood pressure, cumulative BMI exposure, elevated waist circumference, and other important confounding factors. Trajectory groups 4 (Increasing to High) and 5 (Stable High) were too small to contribute evidence for a positive or inverse association. Although these two groups each accounted for less than 5% of the sample, we kept them separate in the tables because of their highly contrasting trends. The results for males for the main CES-D trajectories provide no evidence of an adverse association between depressive symptoms and LV geometry. On the contrary, in the polytomous logistic regression analysis for the Anhedonia subscale trajectories, the ORs for concentric geometry indicated lower odds compared to those for normal geometry. In summary, with exceptions for the small sample sizes for trajectories 4 and 5, the results for females were consistent with our hypothesis, whereas the results for males were inconsistent.

A few previous cross-sectional studies have examined the depression – LVH association without separate consideration for sex differences. A study using 346 males from the Japan Self-Defense Forces was likely the first study to examine the depression – LVH association(11). LVH was identified by electrocardiography and ORs for LVH ranged from 2.22 for an unadjusted model to 2.00 for BMI, SBP, and DBP adjusted models. Another relatively small study with 370 males and females reported that depressive symptoms were significantly associated with a 2.1 higher odds of concentric remodeling or concentric LVH that was independent of BP levels and partially mediated by increased arterial stiffness(12). In the Prescription Use, Lifestyle, and Stress Evaluation (PULSE) study(48) with 769 males and females with non-ST elevation acute coronary syndrome (ACS), those with depressive symptoms had higher odds of multivariable adjusted Cornell product-LVH (OR=1.95, 95%CI: 1.19, 3.18). These models adjusted for risk factors including reduced LV ejection fraction, reduced eGFR, BMI, ACS type, and hypertension. In these studies, the authors identified possible mechanisms for the depression – LVH association, including the high levels of psychological demand and low levels of perceived control that comprise psychosocial job strain(11, 49) (mediated by elevated catecholamines and cortisol(50, 51)), arterial stiffness(12), and elevated sympathetic activity(48). Other possible mechanisms include lifestyle factors, dysregulated hypothalamic-pituitary-adrenal axis activity, and altered immune/inflammation responses(13).

It is important to consider what possible confounders of the depression – LV geometry association might be missing from the present study. Our exploratory analyses assessed sex hormones as potential confounders. There are precedents for androgen associations with LV mass and hypertrophy(52–54). We found bioavailable testosterone significantly and positively associated with concentric LV geometry. The relation of testosterone to adverse LV geometry is possibly causal(17, 55). Testosterone has roles in three systems that are involved in the development of hypertension and CVD and activate differentially in males and females(55): i) testosterone increases sympathetic nervous system activity; ii) testosterone acts in the renin-angiotensin-aldosterone system by shifting the balance of vasodilatory and vasoconstrictor pathways to activation of angiotensin II receptor type 1. A consequence is greater vasoconstriction and cardiac hypertrophy as well as well as sodium retention by the kidneys(55); and iii) testosterone acts in the immune system. Colafella and Denton(55) explain the differential actions of CD4+ T cells between males and females on the production of pro-inflammatory and anti-inflammatory markers.

Another action of testosterone that contributes to hypertrophy is through its effect on cardiac myocytes(56). Marsh et al.(56) have determined that i) cardiac myocytes express the androgen receptor gene, and this is necessary for cardiac muscle to be regulated by androgen steroids; and ii) both testosterone and dihydrotestosterone may induce a hypertrophic reaction in myocytes through receptor-specific mechanisms, increasing amino acid incorporation into protein.

While bioavailable testosterone was significantly associated with adverse LV outcomes in the exploratory model, the sex hormones in males were measured in the first 10 years of the CARDIA study whereas the CES-D measures were recorded every 5 years from Year 5 up through Year 25. It might be expected that hormone measures made contemporaneously with the 20-year sequence of CES-D measures could provide a better joint fit with the CES-D trajectories in predicting Y25/30 LV outcomes and producing less biased estimates for the CES-D trajectory group coefficients. Although none of the estimated median percentage change-in-estimates of ORs for males (Table S3, Supplemental Digital Content) exceeded 10%, the range of the confidence limits for the change-in-estimates for Trajectories 4 and 5 suggest a potential for greater bias by excluding the hormone. This requires future research.

Although several research studies have found testosterone to be inversely associated with depressive symptoms in males (18–22, 57), others have found the relation depends on the length of the CAG repeat sequence in the Androgen Receptor Gene(58–61). Still, other studies have found no association of testosterone with depressive symptoms in males or females(23).

Gender stereotypic roles(62) and what has been termed “gendered responding”(63) may play a part in this study’s inability to tie depressive symptoms in males to LV geometry, although we could tie bioavailable testosterone in males to LV geometry. Men with traditional symptoms of depression, as defined in the DSM-IV-TR, are perceived to be less masculine and more feminine than men with no symptoms or with “male-type” depression(64), characterized by externalizing symptoms such as anger, irritability, somatic symptoms, and substance abuse(63). Price et al. studied the relation of scores on externalizing and internalizing symptoms from the Masculine Depression Scale(65) to measures of typical depressive symptoms on the CES-D scale in four groups (younger men and women with mean age=20.0 years and older men and women with mean age=72.6 years). Masculine traits were positively related to the externalizing scores and were inversely related to CES-D scores in all groups. One of the conclusions from Price et al. was that endorsement of masculine traits is associated with inhibited reporting of typical symptoms of depression. Pertinent to the current study, the CES-D scale may not have captured “male-type” depressive symptoms in CARDIA, and hence the absence of hypothesized associations of CES-D with LV geometry outcomes in males.

In contrast to the findings for males, we did observe a positive depression – adverse LV geometry relation in females. Additionally, in the exploratory analysis we found log(SHBG) inversely associated with adverse LV geometry, and log(SHBG) was also inversely associated with depressive symptoms, suggesting a potential role as confounder. Other studies support this possibility. The MESA study found an androgenic sex hormone profile (i.e., higher levels of free testosterone and lower levels of SHBG) was associated with higher risk for concentric remodeling in females(53). A recent study in postmenopausal hypertensive females showed that free androgen index was positively correlated and SHBG was inversely correlated with LVH(66). While we must acknowledge that studies assessing sex hormone associations with depression in females have had varying findings(21, 25–27, 67, 68), we add that one report found the associations may depend on the length of the time frame post-menopause(27).

The statistical significance of the Somatic subscale trajectory for females is notable. The items on this subscale (i.e., poor appetite, restless sleep, trouble keeping mind on what it was doing, could not get going) are consistently linked to cardiovascular dysfunction(69). Somatic symptoms have been associated with dysregulations of autonomic function(70) and hypothalamic-pituitary-adrenal (HPA) axis(71, 72). The HPA axis has a complex biological interaction with the hypothalamic-pituitary-gonadal (HPG) axis(73, 74). In rodent studies(73, 74) researchers have examined the interactions of these axes on the role of gonadal hormones (i.e., testosterone and estradiol) in the HPA axis response to stress and depressive behavior. Taken together, our findings suggest an underlying link of depression and sex hormones with LVH for females.

There are limitations in this study to consider. Although it is well known that omitted confounders may bias beta estimates and result in null or contrary results in a study, it is not yet widely known that the modern causal inference definition of confounding, versus the traditional associational definition(75) (p. 91), requires that all covariates adjusted for be causally related to the exposure (in this study depressive symptoms), and to the outcome (in this study LVH). In our analyses we did make those presumptions. However, as noted in the Introduction, the direction of the association between hypertension and depression is not certain(5–7), and directional relationships of other pairs of variables, such as smoking and depression(76), could also be called into question. Indeed, given the current states of knowledge in the psychosocial and cardiovascular epidemiology fields, multiple causal structures linking depression and cardiovascular dysfunction are plausible(77) . While we used the best available evidence to guide our analyses, in future studies, considering alternative directed acyclic graphs which account for directions of paths, common causes and common effects of variables, colliders and unmeasured confounders, will inform the adjustment of statistical models.

An additional limitation is we had to pool eccentric LVH and concentric LVH due to small numbers. Studies with larger numbers of eccentric LVH and concentric LVH cases should try to investigate differences in risk factors for these two types of LVH. We did not have measures of female sex hormones (i.e., estradiol) in males or females to include in the exploratory analyses. Estrogen has been reported to have an important role in inhibiting hypertrophy development(78) and is worthy of future study.

Although the statistical tests of sex interaction were not significant at the α=0.05 level, tests for interaction in epidemiologic studies tend to have low power(79) and we presented the results stratified by sex owing to external studies that point to the need for better understanding of sex differences in myocardial pathophysiology(80), LV remodeling(17), and depression(16). Because so few (if any) previous studies have examined the depression – LVH association with consideration of sex differences, additional studies that can confirm or refute our findings are needed.

In summary, we have found depressive symptoms over 20 years were positively associated with LVH in the models for the main CES-D trajectories and the Somatic subscale trajectories for females. The Somatic subscale findings suggest an underlying link among gonadal hormones, depression, and LVH. For males, there was no association with the main CES-D trajectories and an inverse association for concentric geometry with the Anhedonia subscale trajectories. As already noted, additional insight into whether depression is a cause of adverse LV geometry could be obtained upon application of modern causal inference methodology(75). Use of depressive symptom questionnaires that include externalizing symptoms may be informative. Finally, if it can be concluded that depression is causally related to LVH, future studies might test if treating the depression lessens the degree of LVH.

Abbreviations:

LVH

left ventricular hypertrophy

CHD

coronary heart disease

CARDIA

Coronary Artery Risk development in Young Adults

CES-D

Center for Epidemiologic Studies Depression

SHBG

sex hormone binding globulin

BMI

body mass index

BIC

Bayesian Information Criterion

SBP

systolic blood pressure

CMHS

CARDIA Male Hormone Study

CWS

CARDIA Women’s Study

OR

odds ratio

CI

confidence interval

AIC

Akaike information criterion

PULSE

Prescription Use, Lifestyle, and Stress Evaluation

ACS

acute coronary syndrome

eGFR

estimated glomerular filtration rate

MESA

Multi-ethnic Study of Atherosclerosis

HPA

hypothalamic-pituitary-adrenal

HPG

hypothalamic-pituitary-gonadal