Sex Differences in Hemoglobin A1c Levels Related to the Comorbidity of Obesity and Depression

Abstract

Background: Obesity (OB) and major depressive disorder (MDD) are chronic conditions associated with disease burden, and their comorbidity appears more common among women. Mechanisms linking these conditions may involve inflammatory and metabolic pathways. The goal of this study was to evaluate the impact of MDD on relationships between OB and cardiometabolic function, and sex differences therein.

Materials and Methods: Adult offspring from the New England Family Studies (NEFS) were assessed at ages 39–50, including anthropometry, cardiometabolic profile assays, and metabolic syndrome. Individuals were grouped by body mass index (BMI) and MDD status: healthy weight with (n = 50) or without MDD (n = 95) and obese with (n = 79) or without MDD (n = 131). The interaction of (recurrent) MDD and BMI on cardiometabolic markers was tested using quantile regression models.

Results: Participants with MDD exhibited significantly higher hemoglobin A1c (HbA1c) than those without MDD (5.60% vs. 5.35%, p < 0.05). Women with comorbid recurrent MDD and OB had higher HbA1c levels compared to obese women without MDD (5.75% vs. 5.44%, p < 0.05); an interaction between MDD and BMI status was not observed among men.

Conclusions: We demonstrated sex differences in the interaction between BMI and recurrent MDD status on a primary biomarker for diabetes risk, suggesting a common metabolic pathway predisposing women to these comorbid conditions. Further investigation is needed to identify mechanisms that may lead to more effective, sex-dependent screening and therapies.

Introduction

Among US adults, depression has a 12-month prevalence rate of 10.4% and a lifetime prevalence of 20.6% (of which 40% are recurrent¹),² and is a leading cause of disability worldwide.³ Similarly, rates of adult overweight and obesity (OB) have increased dramatically in recent decades,⁴ such that over 70% of the adult US population are now considered either overweight or obese.⁵

In characterizing OB phenotypes, studies report a strong positive association between body mass index (BMI) and major depressive disorder (MDD).^6,7 Demonstrating the compounding impact of this comorbidity, depression is more severe among obese compared with nonobese individuals with MDD.⁸ Furthermore, among those taking antidepressants for MDD, individuals who are overweight demonstrate worse outcomes than healthy weight counterparts.⁹ Reciprocally, among overweight and obese individuals undergoing weight loss treatment, comorbid MDD and high BMI are associated with poorer outcomes.¹⁰

Recent investigations suggest a potential sex difference, with the association stronger in women.^6,7 Women with depression are more likely than those not depressed to develop OB,^11,12 while obese women may be at a 30%–48% increased risk of being diagnosed with MDD.^6,12 Given evidence of comorbidity between MDD and OB, particularly among women, increases in the prevalence of OB (occurring at a faster pace for women than men)⁴ enhance the significant disease burden associated with MDD, OB, and related diseases.¹³ These trends tend to predict parallel increases in other comorbid conditions, such as diabetes and cardiovascular disease (CVD), which also show sex differences in incidence and prevalence. Illumination of potential shared mechanisms underlying the comorbidity between MDD and OB may lead to more effective, sex-dependent treatments.

MDD has been associated with CVD risk factors. Hypertension has been reported as being twice as likely to occur among those who are depressed¹⁴; and women have twice the risk of this comorbidity than men. Cholesterol levels have been associated with MDD, although, inconsistently¹⁵; and type 2 diabetes has been observed to be more prevalent among depressed than nondepressed individuals.¹⁶ The relationship between type 2 diabetes and depression has been well documented, with hemoglobin A1c (HbA1c) levels showing a positive, bidirectional relationship with depressive symptoms.¹⁷

In longitudinal studies, high HbA1c and elevated fasting glucose levels have been associated with increased risk for depression,¹⁸ while insulin resistance and diabetes have been associated with increases in several inflammatory markers, including C-reactive protein (CRP) levels, which are elevated in patients with MDD.¹⁹ Based on recent National Health and Nutrition Survey data, elevated CRP levels among depressed individuals were significantly associated with higher waist circumference, BMI, insulin, and 2-hour glucose tolerance than those not depressed.²⁰ Furthermore, biomarkers of subclinical inflammation (e.g., CRP and adiponectin) have been found to be associated with depressive symptoms in patients with recently diagnosed type 1 and type 2 diabetes, and independent of BMI status.²¹ These studies suggest that individuals with depression suffer a significant inflammatory and metabolic burden that may be further exacerbated by comorbid OB.

Common metabolic and inflammation-related pathways may be implicated in the comorbidity of OB and MDD. Given that OB and MDD are both independent risk factors for CVD, it is important to characterize the clinical characteristics underlying their comorbidity and determine sex differences among CVD biomarkers, which may enable us to identify high-risk individuals. Thus, the first objective of this study was to examine differences in the cardiometabolic profiles between individuals with and without MDD in a large, population-level cohort (the New England Family Studies, NEFS), hypothesizing worse cardiometabolic functioning in those with a history of MDD compared to individuals without MDD.

Furthermore, most studies examining the pathophysiology of comorbid MDD and OB utilize BMI as a continuous variable to examine the linear relationship between depressive symptom severity and BMI level rather than investigate the modifying effect of BMI as a qualitative variable and the interaction between MDD status and BMI groups (healthy weight and obese). Following this, the second objective was to characterize the cardiometabolic profile in individuals with comorbid OB and major depression, compared with that of relevant “control” groups, that is, individuals with OB, but without MDD; healthy weight individuals with MDD; and healthy weight individuals without MDD.

We hypothesized that MDD would enhance the metabolic dysregulation seen in obese versus healthy-weight individuals, such that cardiometabolic markers (i.e., glucose and HbA1c) would increase with greater BMI levels, with the elevated levels in comorbid MDD and OB, particularly among those with recurrent MDD. Finally, given the dearth of research focused on elucidating potential sex differences, despite variable prevalence rates of both conditions by sex, the third objective was to investigate sex differences in these relationships, hypothesizing that metabolic disturbance observed among subjects with comorbid OB and depression would be more pronounced among women than men.

Materials and Methods

Study sample

Study participants were adult offspring born to a community-based population of pregnant women enrolled in the New England cohorts of the Collaborative Perinatal Project (CPP) between 1959 and 1966.^22,23 As part of the CPP, these women were prospectively followed through delivery and their offspring up to age 7 years. In the New England CPP cohort, there were 17,741 pregnancies with 17,921 offspring [NEFS²⁴].

Approximately 4,000 NEFS offspring were located, consented, and reassessed as adults.²³ Among the pool of NEFS offspring, we identified 525 adult singleton offspring for this study, who came from a previously selected subset of NEFS offspring for whom comprehensive cardiometabolic data and diagnostic data for MDD were available. Offspring were between 39 and 50 years of age at adult follow-up and had participated in at least one of three previous NEFS adult follow-up studies (R01MH074679: JMG, PI; RC2AG036666: EL, SB, MPIs; and R01AG023397: SB, PI). All three of these NEFS studies used similar strategies to locate, recruit, and evaluate subjects among the original CPP offspring, described elsewhere. Overall, demographic and clinical characteristics of the participants did not significantly differ between the three substudies.

Individuals with a history of MDD or current MDD symptomatology (n = 179) and controls without MDD (n = 346) were identified among the NEFS offspring included in this study. The higher frequency of MDD (34%) was due to the fact that over half the sample (n = 259) came from a separate, prior NEFS follow-up study on MDD. For 415 of the 525 subjects in the overall sample (79%), diagnoses were based on systematic, semistructured clinical assessments completed with expert diagnosticians providing consensus diagnoses based on the Structured Clinical Interview for Diagnoses [SCID DSM-IV version²⁵] and verified using medical records (when available). Recurrent cases of depression (MDD-R) were identified from the SCID interviews based on participants' reports of two or more lifetime episodes of MDD.²⁶

For 110 of the 525 subjects (21%), a score ≥16 on the Center of Epidemiological Studies Depression (CES-D) scale, a semistructured assessment tool based on the DSM-IV,²⁷ was used to define cases of mild depression.^28,29 As the CES-D measures current symptomatology (within the past week), recurrent/past history of depression was not ascertained on these 110 subjects. Participants with lifetime diagnoses of psychotic and bipolar disorders were excluded from the analytic sample.

Data ascertainment

In accordance with the protocols of prior NEFS follow-up studies (R01MH074679; RC2AG036666; and R01AG023397), subjects completed relevant study procedures. For those subjects evaluated with the SCID, diagnostic evaluation of MDD was ascertained either through a de novo SCID interview (if no interview had previously been conducted as part of an earlier NEFS study) or through a follow-up SCID interview with a clinician (if an SCID had previously been done as part of an earlier NEFS study), which focused on obtaining updated psychiatric symptomatology since the last interview. The de novo SCID, SCID update, and CES-D evaluations were completed by the time of the follow-up study (R01MH074679; RC2AG036666; and R01AG023397) and were collected in parallel with that of anthropometric measures and biomarker assays. The study protocols were approved by institutional review boards at Partners Healthcare, Brown University, and Memorial Hospital of Rhode Island.

Anthropometry measures

Weight and height were measured and used to calculate BMI. We used standard definitions to categorize normal weight (BMI: 18.5–24.9) and obese (≥30.0 BMI) individuals. Waist circumference was measured from the level of the iliac crests and around the trunk in a horizontal plane.

Biomarker assays

Biomarkers included a lipid profile, glucose, HbA1c, CRP, and metabolic syndrome criteria. Lipid profile: the determination of total cholesterol, triglycerides, and high-density lipoprotein cholesterol (HDL-C) concentrations was performed on the Hitachi 911 analyzer using reagents and calibrators from Roche Diagnostics (Indianapolis, IN).

Glucose and HbA1c: glucose was measured on the Hitachi 911 analyzer using Roche Diagnostics reagents. HbA1c was measured using the Roche P Modular system based on turbidimetric immunoinhibition with hemolyzed whole blood or packed red cells (Roche Diagnostics). The reported result is a calculation of the %HbA1c in the total hemoglobin. Day-to-day variability at %HbA1c values of 5.5 and 9.1 are 1.9% and 3.0%, respectively. Glucose data were not collected in R01MH074679, and HbA1c data were not collected as part of RC2AG036666/R01AG023397, resulting in missing data on n = 156 (30%) for glucose and n = 110 (21%) for HbA1c.

CRP: serum concentration of CRP was analyzed using a high-sensitivity immunoturbidimetric assay on a Hitachi 917 analyzer (Roche Diagnostics), with reagents and calibrators from DiaSorin (Stillwater, MN). Metabolic syndrome: blood pressure (BP) was measured with a Dinamap automated device (Dinamap Monitor Pro 100). Metabolic syndrome was defined using modified Adult Treatment Panel III (ATPIII) criteria, requiring presence of ≥3 of these criteria: waist circumference >40 inches (men) and >35 inches (women); HDL-C <40 mg/dL (men) and <50 mg/dL (women), or prescription drug treatment for low HDL; triglycerides ≥150 mg/dL, or prescription drug treatment for elevated triglycerides; high BP: systolic BP ≥130 mm Hg and diastolic BP ≥85 mm Hg, or on antihypertensive treatment; and fasting glucose ≥100 mg/dL, or on diabetes treatment.³⁰

Statistical analysis

Demographic characteristics and clinical measures of the sample were compared by MDD status using chi-square and t-test statistics for qualitative and quantitative data, respectively. Continuous variables with skewed distributions (>|0.8|) were normalized using natural logarithm (ln) transformation for analysis. Given that distributions for glucose and HbA1c levels were highly skewed, and remained so even after ln transformation, comparisons based on these measures were performed using nonparametric (Wilcoxon rank sum) methods. For those clinical continuous measurements that differed significantly by MDD status, we created binary variables based on clinical thresholds of these measures and compared rates by MDD status using chi-square tests. Clinically relevant levels of HbA1c were identified as follows: ≥5.7%³¹; of glucose, fasting plasma levels ≥100 mg/dL³¹; and CRP ≥3 mg/L.³²

For clinical measurements that differed significantly by MDD status, we evaluated whether the association was modified by categorical BMI status (normal weight vs. obese) by adding an interaction between MDD and BMI status into a regression model that included main effects for MDD and BMI status. Provided the positive skew in the distributions for glucose, HBA1c, and CRP, we used median regression to estimate the conditional median of the response variable [STATA 15 QREG2 (ref. QREG2: Stata module to perform quantile regression with robust and clustered standard errors)]. For the analyses of skewed data, means are not an adequate measure of central tendency because they are sensitive to outliers and medians are better measures of central tendency for such skewed distributions.³³

Analytic models were adjusted for intrafamilial correlation among siblings and relevant demographic and clinical factors identified as potential confounders. Using an iterative process, potential confounders were identified among demographic and clinical factors associated with both the outcome (Wald p-value <0.2 in final model) and the exposure of interest (accounting for ≥10% change in the exposure estimate when added to the model). These potential confounders were identified as fetal growth restriction (for outcomes HbA1c and CRP) and metabolic syndrome (for outcomes glucose and HbA1c). Sample sizes for these models varied according to available data: HbA1c unadjusted: n = 262 total; HbA1c adjusted: n = 259 total; CRP unadjusted: n = 307 total; and CRP adjusted: n = 305 total.

Models were run for the overall sample as well as stratified by sex. For models where the p-value of the interaction was <0.20, we ran additional models using recurrent MDD status to determine whether the interaction held in more diagnostically stringent analyses. To quantify the magnitude of the impact of MDD status on cardiometabolic outcomes exhibiting significant interactions, effect sizes were derived by calculating the mean difference in HbA1c between those with recurrent MDD (MDD-R) versus without OB and divided this difference by the overall standard deviation.

Sensitivity analyses were conducted to examine change in effect modification by BMI status on the association between MDD/MDD-R status and %HbA1c (1) with the addition of subjects meeting overweight BMI status [categorized as overweight (BMI: 25.0–29.9) versus normal weight (BMI: 18.5–24.9) and as overweight/obese (BMI ≥25.0) versus normal weight (BMI: 18.5–24.9)], and (2) when individuals on diabetes medications were excluded. An additional sensitivity analysis replaced BMI status with waist circumference status (as a marker of central adiposity); high waist circumference was defined as ≥102 cm in men and ≥88 cm in women.

Results

Of the 525 subjects in the study, 179 (34%) had a history of MDD or current MDD symptomatology. Although 34% is higher than the US prevalence of ∼20%, we oversampled MDDs to assess comorbidity and MDD alone. The remaining 346 without MDD were analyzed as controls (Table 1). The prevalence of OB in our cohort was 40% (n = 210), which is representative of the US general population.³⁴ Overall, the study sample was on average 44 years of age, female, Caucasian, nonsmoking, obese, and did not meet criteria for metabolic syndrome.

Table 1.

Sample Demographics, Anthropometry, and Serum Metabolic Profile by Major Depressive Disorder Status

	Total sample				MDD				No MDD
	(n = 525)				(n = 179)				(n = 346)
	n	%			n	%			n	%			χ2a	p
Sex													6.5	0.01*
Male	213	40.6			59	33.0			154	44.5
Female	312	59.4			120	67.0			192	55.5
Race													1.2	0.54
White	451	85.9			150	83.8			301	87.0
Black	63	12.0			24	13.4			39	11.3
Other	11	2.1			5	2.8			6	1.7
Smoking status, current													10.7	0.001*
Yes	128	24.4			59	33.0			69	19.9
No	393	74.9			119	66.5			274	79.2
BMI status													3.8	0.29
Underweight	2	0.4			0	0.0			2	0.6
Normal weight	145	27.6			50	27.9			95	27.5
Overweight	163	31.0			48	26.8			115	33.2
Obese	210	40.0			79	44.1			131	37.9
Medications, current
BP	47	9.0			13	7.3			34	9.8			1.2	0.27
Cholesterol	35	6.7			9	5.0			26	7.5			1.4	0.24
Diabetes	30	5.7			15	8.4			15	4.3			3.2	0.07
Antidepressants	64	12.2			31	17.3			33	9.5			6.1	0.01*
SSRIs	45	8.6			21	11.7			24	6.9			3.1	0.08
Hypertension, current (systolic BP ≥140 and/or diastolic BP ≥90)													0.7	0.40
Yes	55	10.5			16	8.9			30	8.7
No	465	88.6			162	90.5			303	87.6
Hypertension, current (systolic BP ≥140 or diastolic BP ≥90 or currently taking medication for hypertension)													1.7	0.19
Yes	92	17.5			26	14.5			66	19.1
No	432	82.3			153	85.5			279	80.6
Metabolic syndrome													4.6	0.03*
Yes	152	29.0			41	22.9			111	32.1
No	370	70.5			136	76.0			234	67.6
	n	Mean	Median	SD	n	Mean	Median	SD	n	Mean	Median	SD	t statistic1t statistic1	p
Age at adult follow-up	525	44.38	44.00	2.85	179	44.24	44.00	2.71	346	44.45	45.00	2.92	−0.8	0.42
Anthropometry
BMI (kg/m2)a BMI (kg/m2)a	520	29.69	28.32	7.32	177	30.69	28.75	8.44	343	29.18	28.13	6.63	1.9	0.06
Waist circumference (cm)	519	96.45	95.48	18.23	174	96.59	95.34	18.55	345	96.37	95.50	18.10	0.1	0.90
BP, systolic (mmHg)	520	116.96	115.80	15.02	178	116.99	115.80	15.45	342	116.94	115.75	14.82	0.0	0.97
BP, diastolic (mmHg)	520	74.17	72.90	10.18	178	73.49	72.68	10.20	342	74.53	73.45	10.16	−1.1	0.27
Serum metabolic profile
Total cholesterol (mg/dL)	519	196.28	191.00	40.64	177	196.54	192.00	42.41	342	196.15	191.00	39.75	0.1	0.92
HLD cholesterol (mg/dL)a HLD cholesterol (mg/dL)a	519	51.59	49.00	16.33	177	53.26	50.00	17.60	342	50.72	48.00	15.59	1.5	0.13
Triglycerides (mg/dL)a triglycerides (mg/dL)a	519	130.78	102.00	92.93	177	126.19	96.00	99.68	342	133.16	105.00	89.30	−1.3	0.19
LDL cholesterol (mg/dL)	501	118.13	115.00	36.18	173	117.78	115.40	37.99	328	118.32	115.00	35.25	−0.2	0.87
Glucose (mg/dL)a,b Glucose (mg/dL)a,b	366	98.02	91.00	33.38	129	93.63	89.00	20.10	237	100.41	93.00	38.58	Z = −2.6	0.01*
HbA1c (%)a,b HbA1c (%)a,b	384	5.56	5.46	0.92	152	5.69	5.60	0.89	232	5.48	5.35	0.94	Z = 2.9	<0.001*
CRP (mg/L)a	447	8.62	2.36	17.24	171	10.57	3.78	18.85	276	7.41	2.05	16.07	2.9	<0.001*

^a Diffferences by MDD status compared using using χ² test (categorical variables) or t-test (continuous variables). BMI, HDL-C, triglycerides, and CRP ln transformed to normalize skewed |0.8| distributions before analysis. Glucose and HbA1c highly skewed and compared using nonparametric (Wilcoxon) methods. Significant results (p < 0.05) bolded and marked with an asterisk (^*).

^b No glucose data from EDHEALTH subjects, n = 156; no HbA1c data collected as part of LEAP, n = 110.

BMI, body mass index; BP, blood pressure; CRP, C-reactive protein; HbA1c, hemoglobin A1c; HDL-C, high-density lipoprotein cholesterol; MDD, major depressive disorder; SD, standard deviation.

As expected, subjects with MDD were significantly more likely than those without to be female (67% vs. 55.5%, p = 0.01), smoke (33.0% vs. 19.9%, p < 0.001), and use antidepressant medications (17.3% vs. 9.5%, p = 0.01). Those without MDD were significantly more likely to meet criteria for metabolic syndrome (32.1% vs. 22.9%, p = 0.03). However, they were not significantly more likely than those with MDD to have any one of the five individual metabolic syndrome criterion (data not shown).

Among the serum metabolic biomarkers, those with MDD had significantly higher levels of CRP than those with no MDD (medians: 3.78 mg/L vs. 2.05 mg/L, p < 0.05) and HbA1c (medians: 5.60% vs. 5.35%, p < 0.05) (Table 1), as well as more likely to have levels above clinically relevant thresholds (CRP: 51.4% vs. 29.5%, p < 0.05; HbA1c: 36.9% vs. 19.7%, p < 0.05) (Table 2). Median levels of CRP among those with MDD, but not those without, were well above clinically relevant threshold for this biomarker. Seemingly inconsistent with the HbA1c results, the levels of fasting glucose serum were greater among those without MDD (93.0 U vs. 89.0 U, p = 0.01) (Table 1). However, the frequency of those with fasting glucose levels above clinically relevant thresholds was similar by MDD status (16.2% vs. 18.5%, p = 0.34) (Table 2).

Table 2.

Rates of Abnormal Serum Biomarker Levels Based on Clinical Thresholds by Major Depressive Disorder Status

	Total sample		MDD		No MDD
	(n = 525)		(n = 179)		(n = 346)
	n	%	n	%	n	%	χ2a	p
High glucose serum, fasting glucose ≥100 mg/dL							0.9	0.34
Yes	93	17.7	29	16.2	64	18.5
No	273	52.0	100	55.9	173	50.0
Missing	159	30.3	50	27.9	109	31.5
High CRP, ≥3 mg/L							12.2	<0.01*
Yes	194	37.0	92	51.4	102	29.5
No	253	48.2	79	44.1	174	50.3
Missing	78	14.9	8	4.5	70	20.2
High HbA1c, ≥5.7%							8.0	<0.01*
Yes	134	25.5	66	36.9	68	19.7
No	250	47.6	86	48.0	164	47.4
Missing	141	26.9	27	15.1	114	32.9

^a Differences by MDD status compared using χ² test. Significant results (p < 0.05) bolded and marked with an asterisk (^*).

Among women, the interaction between BMI (normal weight vs. obese) and MDD status was not significant in adjusted and unadjusted models for HbA1c and CRP (HbA1c_Adjusted: β = 0.05, p = 0.13, effect size = 0.50; CRP_Adjusted: β = 0.89, p = 0.10) (Table 3). The interaction for HbA1c met statistical significance when using more stringent diagnostic criteria for MDD (i.e., MDD-R; HbA1c_Adjusted: β = 0.08, p = 0.04). As revealed in Figure 1, women with MDD-R who were obese (n = 20), exhibited significantly higher HbA1c levels than those with MDD-R who had normal-weight BMIs (n = 21; effect size = 1.06), while only minimal differences were observed between BMI groups for women without MDD (obese: n = 37; normal-weight: n = 56; effect size = 0.51). In contrast, among men, the interaction between BMI and MDD status was nonsignificant in both adjusted and unadjusted models for HbA1c and CRP.

Table 3.

Effect Modification by Body Mass Index Status (Normal Weight vs. Obese) on the Association Between Major Depressive Disorder Status and Hemoglobin A1c, by Sex

	n		HbA1ca
			β	SE	p
Total sample	262
MDD	111
		Unadjustedb	0.02	0.04	0.64
		Adjustedc	0.02	0.03	0.50
MDD-R	53
		Unadjusted	0.09	0.04	0.03*
		Adjusted	0.07	0.05	0.15
Men	92
MDD	34
		Unadjusted	−0.06	0.06	0.31
		Adjusted	−0.03	0.06	0.58
MDD-R	12
		Unadjusted	0.08	0.07	0.29
		Adjusted	0.05	0.12	0.66
Women	170
MDD	77
		Unadjusted	0.06	0.05	0.15
		Adjusted	0.05	0.03	0.13
MDD-R	41
		Unadjusted	0.09	0.06	0.11
		Adjusted	0.08	0.04	0.04*

^a Median regression models using quantile regression and ln transformed values. Significant results (p < 0.05) bolded and marked with an asterisk (^*).

^b Unadjusted models include main effects for MDD and BMI status and a term for their interaction. Models are adjusted for intrafamilial clustering.

^c Adjusted models include both fetal exposure status (low birth weight or preeclampsia) and metabolic syndrome. Sample sizes for adjusted models are slightly smaller (total sample overall: n = 259; total sample MDD: n = 109; and total sample MDD-R: n = 51) than for unadjusted models.

SE, standard error.

FIG. 1.

Median levels of HbA1c by MDD status, BMI status (normal weight vs. obese), and sex median levels of HbA1c (%) are plotted with respect to MDD status, BMI (normal weight vs. obese), and sex. (a) Median HbA1c levels by MDD status and BMI; (b) median HbA1c by MDD status, sex, and BMI; (c) median HbA1c levels by MDD-R status, sex, and BMI; and in (c), the interaction between MDD and BMI status for HbA1c met statistical significance for individuals with MDD-R (HbA1c_Adjusted: β = 0.08, p = 0.04). Solid lines indicate MDD group [MDD in (a) and (b); MDD-R in (c)]; dashed lines indicate non-MDD group. Gray lines in (b) and (c) indicate data from women. BMI, body mass index; MDD, major depressive disorder. HbA1c, hemoglobin A1c.

There was no effect modification by BMI status (normal weight vs. obese) on the association between serum glucose levels and MDD status overall or by sex in either unadjusted or adjusted models.

Sensitivity analyses

The interaction between BMI (normal weight vs. overweight) and MDD/MDD-R status was not significant in adjusted and unadjusted models for HbA1c in the overall sample or by sex. However, the interaction for HbA1c with MDD and MDD-R met statistical significance when combining the overweight and obese categories (vs. normal weight) and was specific to women (MDD: HbA1c_Adjusted: β = 0.06, p = 0.03; effect size = 0.54 and MDD-R; HbA1c_Adjusted: β = 0.09, p = 0.01; effect size = 0.77) (Supplementary Table S1). We did not find a significant interaction between central adiposity and MDD/MDD-R on HbA1c levels (Supplementary Table S2).

The association between comorbid MDD-R and OB and HbA1c levels in women was similar (effect size = 0.95) when excluding subjects on diabetes medications (n = 30), but the statistical significance was attenuated, likely due to more limited power in the untreated subset (Supplementary Table S3).

Discussion

Several epidemiologic studies support a bidirectional relationship between OB and depression,³⁵ yet few clinical studies have examined the impact of their comorbidity on biomarkers related to cardiometabolic outcomes. In this population-level study, we hypothesized that specific metabolic disturbances would be associated with both OB and major depression and their co-occurrence and that these associations would vary by sex.

In our overall sample (n = 525), we showed that subjects with MDD had significantly higher HbA1c levels, as well higher prevalence of clinically relevant threshold levels for HbA1c (i.e., prediabetes) than those without MDD. Among women alone, this association with HbA1c levels was significantly modified by BMI status—specifically, obese women with recurrent MDD had higher levels of HbA1c than obese women without MDD. No modification by BMI status was observed among men. Our findings suggest that obese women with more severe, persistent forms of MDD (i.e., recurrent vs. single episode) might be the most susceptible to metabolic disturbances.

HbA1c is the gold standard measurement to assess glycemic control in the clinical setting and serves as an important biomarker for prediabetes (insulin resistance) and type 2 diabetes. Self-reported depressive symptoms are highly prevalent in patients with both type 1 and type 2 diabetes³⁶ and have been associated with elevated HbA1c levels.³⁷ In our cohort, we found that subjects with MDD had not only higher HbA1c levels compared with those without MDD, but also were significantly more likely to have their HbA1c levels above the clinical threshold for prediabetes. Although the prevalence of diabetes was low in our population (<10%), the average HbA1c levels in our subjects with OB were within the prediabetes range (HbA1c 5.7%–6.4%).³⁸

Among individuals with OB in our cohort, those with coexisting MDD had worse glycemic control compared with those without MDD, suggesting that comorbid MDD may enhance the severity of glucose dysregulation associated with OB. Our findings are consistent with observations that depressive symptoms can occur in the early prediabetes stage³⁹ and that metabolically unhealthy individuals with OB are more likely to develop depression compared to their metabolically healthy counterparts with OB.⁴⁰ Previous studies focused on patients with type 2 diabetes suggest a robust relationship between baseline HbA1c levels and subsequent development of depressive symptoms in elderly individuals.⁴¹ Prenatal stress models have demonstrated shared fetal programming of both cardiometabolic disruption and mood and/or anxiety-related symptomatology.⁴²

Furthermore, among women, we found a significant interaction between recurrent MDD and OB on HbA1c levels, which was not present among the men. Among the few studies that have evalutated sex differences in the comorbidity of OB and depression and impact of HbA1c, ours is a novel finding. Our findings are consistent with prior studies conducted in middle-aged women showing that recurrent depression was more strongly associated with diabetes and CVD risk than single-episode MDD,⁴³ although these studies did not examine the impact of comorbid OB or sex differences in these relationships. Additional data from a large Australian cohort study of individuals with type 2 diabetes analyzed using latent class growth modeling suggest a strong relationship between sex, BMI, and recurrent depression, with women with higher BMIs demonstrating a more persistent depressive course than their male counterparts.⁴⁴

In addition, among women, we also found significant effect modification by BMI status in the association between HbA1c and recurrent MDD when combining the overweight and obese participants in our analysis, although the effect size was somewhat attenuated in comparison to the analysis contrasting the obese and normal-weight groups. Although our main analyses were focused on the relationship between comorbidity of OB and MDD/MDD-R and HbA1c levels, our results suggest that overweight women with MDD on this comorbid spectrum are susceptible to metabolic disturbances. Taken together, our findings suggest that there is a significant metabolic burden to glucose homeostasis among individuals with both overweight/OB and depression that may be sex dependent, and stronger in those with a recurrent history of depression.

Several pathophysiological mechanisms have been proposed linking glucose dysregulation with depression, including inflammation, decreased serotonergic neurotransmission, hyperactivity of the hypothalamic-pituitary-adrenal (HPA) axis, circadian rhythm disturbances, and early life factors, such as low birth weight and childhood OB predisposing individuals to both OB/diabetes and depression.^35,42 Consistent with prior studies,¹⁹ preliminary findings in our subjects with MDD showed significantly higher levels of inflammation measured by CRP than those without MDD.

Our findings are consistent with those from other studies reporting elevated inflammatory biomarkers associated with the development of both diabetes and depression,^20,21 particularly among obese individuals.²⁰ In an open-label study, patients with MDD and either abdominal OB or metabolic syndrome treated with an insulin-sensitizing agent, pioglitazone, demonstrated improvements in depressive symptoms, in addition to reductions in insulin resistance and inflammatory markers.⁴⁵ These findings suggest shared common metabolic and inflammation-related pathways may be implicated in the comorbidity of OB and major depression and its treatment. Although we did not find significant sex differences with interaction of BMI status and MDD on CRP levels, our data suggest that the effect may be specific to women. Future research is needed to examine whether these trends extend to more specific proinflammatory and anti-inflammatory markers (i.e., IL-1β, IL-6, and adiponectin), which have been more consistently found to differ according to BMI status.⁴⁶

Our study has notable strengths and some limitations. Strengths of this study include a well-characterized cohort of individuals involved in a population-level study and use of rigorous psychiatric interviewing instruments to define MDD diagnostically. Furthermore, most studies to date have explored either BMI status or MDD status alone, with few studies incorporating both factors within the context of the cardiometabolic profile. In addition, we examined the sex differences underlying this comorbidity, which have not been sufficiently addressed in previous studies.

We acknowledge that the cross-sectional design precluded inference regarding causality and change over time in primary cardiometabolic outcomes. The sample size of men with MDD was relatively low, restricting the power to detect differences between women and men. Furthermore, we had inadequate power to test for three-way interaction between sex, MDD, and BMI status. We acknowledge that the age range of our cohort may not be representative of higher CVD risk population of older men and women, but emphasize that it may inform preventative efforts on at-risk populations. Yet, we were able to detect more severe deficits in glycemic control levels in participants with MDD versus no MDD, particularly in women with OB.

Given limited sample size and power, we did not exclude subjects on diabetes medications in our main analysis. However, results were similar with and without inclusion of treated subjects, suggesting that it was appropriate to include them in the analyses. These findings suggest that premenopausal women with major depression may be at risk for developing early metabolic disease that is further exacerbated by OB. Compared to diabetic patients without MDD, those with MDD tended to be younger and female, reflecting the earlier onset of the disorder.⁴⁷ In a similar age population to our cohort, one cross-sectional study showed that premenopausal women (ages 20–49) with MDD had higher prevalence of metabolic syndrome than those without MDD,⁴⁸ an association that was not present in the older age group (ages 50–82).

In addition, due to ascertainment strategies, there was a higher frequency rate of history of MDD in our sample (34%) compared to national lifetime prevalence rates (20%). This oversampling on MDD was purposeful in keeping with the aims of this study to examine comorbidity between MDD and OB, although this may have impacted the generalizability of our findings to the US population. Finally, we were limited by missing data on HbA1c from one of our substudies, which decreased our power to detect effects for this metabolic marker. However, missing data primarily resulted from lack of collection from the entire sample, and given that all three substudies drew from the same population-level study, it is unlikely that these missing data systematically impacted our results.

In summary, we found elevated HbA1c in individuals with both OB and a history of MDD, compared to BMI- and MDD-matched comparison groups, specifically among women and those with recurrent MDD, suggesting a disease burden related to glucose homeostasis. These data emphasize the importance of early screening and prevention of type 2 diabetes in women with OB and MDD for primary health care professionals and psychiatrists to prevent future cardiometabolic complications in this high-risk population. Future studies would benefit from examining these measures over time to elucidate the directionality of pathophysiology associated with comorbid MDD and OB.

Author Disclosure Statement

J.M.G. is on the scientific advisory board for and has an equity interest in Cala Health (a neuromodulation company). There is no conflict of interest with the study reported here. In addition, J.M.G.'s interests are reviewed and managed by the Massachusetts General Hospital and Mass General Brigham Health care in accordance with their institutional policies. No competing financial interests exist for any of the authors.

Funding Information

This study was supported by the State of Arizona Arizona Biomedical Research Commission (ABRC) ADHS14-00003606 (Handa & Goldstein, multi-PIs) and the National Institutes of Health (R01MH074679 (Goldstein, PI), RC2AG036666 EL, SB, MPIs, R01AG023397 (Goldstein, PI), and K08HL132122, Huang, PI).

Supplementary Material

Supplementary Table S1

Supplementary Table S2

Supplementary Table S3