Adherence to Mediterranean Diet and Depressive Symptomatology Among Boston Area Puerto Ricans

Abstract

Background

Higher Mediterranean diet (MeD) adherence has been linked with lower depressive symptomatology, but research examining this association is limited, especially among Latinos, including mainland U.S. Puerto Ricans. Hence, we examined cross-sectional and longitudinal associations between MeD adherence and self-reported depressive symptomatology in Boston area Puerto Rican adults.

Methods

The Center for Epidemiologic Studies Depression Scale (CES-D) was used to evaluate depressive symptoms. Adherence to MeD was assessed at all 3 visits. We used multivariable linear regression for baseline cross-sectional analysis, and linear mixed effects modeling over 3 waves of follow-up for longitudinal analysis. We also assessed whether baseline MeD adherence affected 5y CES-D trajectory. We conducted sensitivity analyses among participants without diabetes, and among participants with complete MeD and CES-D measures at all visits.

Results

MeD adherence was significantly associated with CES-D score at baseline (β = −2.0, 95% confidence interval [CI] −4.0, −0.04 for highest vs lowest tertile, p trend = .04) and across 3 waves (β = −1.9, 95% CI = −3.0, −0.8 for highest vs lowest tertile, p trend = .0005). Results were similar in analyses restricted to participants without diabetes, as well as among participants with complete CES-D and MeD scores at all visits.

Conclusions

While CES-D score was consistently lower in those with higher MeD adherence over 5 years of follow up, no relationship between baseline MeD adherence and 5y CES-D trajectory was observed.

Healthful diet is one of the most important modifiable (1) factors in maintaining a healthy lifestyle and has been linked with reduced risk of several chronic diseases such as diabetes, cardiovascular diseases, many types of cancer, as well as all-cause mortality (2,3). Growing evidence suggests that adhering to a quality diet, rich in vitamins, minerals, and antioxidants, may protect the brain from oxidative stress, improve neurovascular function, and support optimal mental health (4–6). Consuming foods with higher nutritional value, including healthy fats, lean protein, fruits, and vegetables may protect against a wide range of neurological and neuropsychiatric disorders (7–9).

Overall dietary quality and pattern are thought to have more substantial impact on physical and mental health than consumption of any specific individual nutrient. The Mediterranean diet (MeD), characterized by higher intake of vegetables, fruits, pulses/legumes, nuts, cereals, fish, monounsaturated fats from olive oil, and lower intake of dairy products, meat, poultry, and saturated fats than the typical U.S. diet, along with moderate intake of wine with meals (3), has shown protective results for many aspects of health, including depression (10–12).

Inflammation is thought to play a key role in etiology of depression. Higher proinflammatory and lower antiinflammatory cytokines, as well as reduced neuronal plasticity, have been associated with depression (13). The MeD has antiinflammatory characteristics and is rich in antioxidants, polyphenols, monounsaturated, and polyunsaturated fats (13). It has, thus, been suggested that a diet with Mediterranean characteristics may influence neurological health and reduce risk of depressive disorders, potentially by reducing systemic inflammation (14). Nutrients including omega-3 polyunsaturated fatty acids, antioxidants, B-vitamins, and dietary polyphenols may encourage adult neurogenesis, influence neurotransmitter concentrations, protect against neuronal death, and improve endothelial cell function (5,15). Several components of MeD, such as fruit, vegetables, seafood, and olive oil, have also been shown to have beneficial effects against depression (16–18).

A systematic review on a link between polyphenols in MeD and depressive symptoms supported the association and suggested that the diet rich in polyphenols, such as MeD, may effectively alleviate depressive symptoms (19). In agreement with these findings, another cross-sectional study (20), suggested that MeD adherence may promote healthy aging and exert protective effects on neuropsychiatric health among participants with existing multicomorbidities, older 65 years of age. MeD adherence was significantly associated with reduced severity of depressive symptoms (r = −0.206, p = .014) as well as medical multicomorbidities (r = −0.247, p = .003) in this study. A recently published systematic review and meta-analysis, evaluating association between MeD consumption and physical performance as well as cognitive health in the age group 60 years and older, also reported that higher MeD adherence was associated with better cognitive health and memory in cross-sectional studies. Though this study observed protective effect of MeD adherence against cognitive decline among nondemented participants in longitudinal studies (RR = 0.26, 95% confidence interval [CI]: 0.23–0.29, I2 = 100%, p < .00001), no such association was found for dementia (21).

Despite of the growing research on effects of nutrition on neurological and psychiatric health, longitudinal evidence linking dietary patterns with depressive symptomatology is limited, especially among at-risk minority populations like mainland U.S. Puerto Ricans. Puerto Ricans represent the second largest Hispanic subgroup in the U.S., after Mexican Americans (22,23). United States mainland Puerto Ricans have distinct food consumption patterns compared to the general U.S. population (24). In a recent study, mainland Puerto Ricans had 2.4 times higher odds of having worse dietary habits, compared to those in Puerto Rico, when evaluated with the HEI-2010 dietary instrument (odds ratio [OR] = 3.4, p = .02) (25). Puerto Rican adults have been shown to have substantially higher prevalence of depression and several other chronic diseases and disorders than non-Hispanic Whites (NHW) (26,27). Puerto Rican adults residing in the United States have a higher risk of depression and high depressive symptomatology, among other psychological and psychiatric conditions due to poverty, lower or lack of education, poor psychological acculturation as well as higher prevalence of chronic comorbidities (27). Despite this, we are aware of no studies to date that have evaluated a link between MeD and depression among Puerto Ricans living in the United States. Identifying relatively underexplored and modifiable risk factors for depression such as diet could be an important step towards lowering depression burden as well as setting standards for future dietary recommendations. Thus, the aim of this study was to evaluate cross-sectionally, as well as longitudinally, the association between adherence to MeD and self-reported depressive symptomatology in a cohort of urban Puerto Rican residents of Massachusetts.

Method

Study Design and Data Collection

Study design and data collection methods for the Boston Puerto Rican Health Study (BPRHS) have been described previously (23). Briefly, the BPRHS is a longitudinal cohort of 1 500 self-identified Boston-area Puerto Rican adults, aged 45–75 years at baseline (2004–2009), who subsequently underwent 2y and 5y follow-up assessments. At baseline, 2y and 5y study visits, information on participant demographics, health conditions, diseases, dietary habits, and patterns was collected at participants’ homes by trained bilingual interviewers in either English or Spanish, along with anthropometric and blood-pressure measurements. A certified phlebotomist drew blood samples for analysis. This study was approved by Institutional Review Boards at Tufts University and the University of Massachusetts Lowell, and has clinicaltrials.gov registration NCT01231958. In our analyses, 1 404 participants had complete data on Center for Epidemiologic Studies Depression Scale (CES-D) and MeD score at baseline, 1 194 at 2y and 803 at 5y follow-up visits (Figure 1).

Figure 1.

Flow diagram of participants included in the analysis.

¹Identified as Puerto Rican; ²Excluded due to serious illness/being homeless or if the participant moved away from study area; ³Declined participation due to being busy, lack of interest, or refused to have blood drawn; ⁴Excluded due to low Mini-Mental State Examination score (MMSE <10); ⁵Baseline cohort available for analyses; ⁶Excluded due to implausible dietary intake; ⁷Particpants with complete CES-D and MeD measures at baseline; ⁸Participants with complete CES-D and MeD measures at 2y; ⁹Participants with complete CES-D and MeD measures at 5y.

CES-D = Center for Epidemiologic Studies Depression Scale; MeD = Mediterranean diet, MMSE = Mini-Mental State Exam.

Dietary Assessment and Development of MeD Adherence Score

Dietary intake over the past year was assessed, as part of interviewer administered home interview, using a detailed Food Frequency Questionnaire (FFQ) designed for and validated with Puerto Rican adults. The traditional Puerto Rican diet differs from the typical U.S. diet. Our group therefore, adapted a FFQ based on the (NCI)/Block FFQ with several modifications from the Hispanic Health and Nutrition Examination Survey dietary recalls for Puerto Rican adults (24) including addition of Hispanic-specific foods, such as plantains, avocado, mango, cassava, empanadas, and custard, as well as open-ended portion sizes. The FFQ has been validated in prior studies focused on Hispanic adults (28,29). The Nutrient Data System for Research software was used to compute nutrient intake information from the FFQs. We excluded participants with implausible dietary intake who left more than 10 questions unanswered on the FFQ, as well as those with unreliable energy intake (over 600 or <4 800 kcal/day). We excluded 90 participants at baseline, 63 participants at 2y, and 149 participants at 5y, due to implausible dietary intake.

The MeD pattern measured adherence to 9 different dietary components: whole grains, vegetables, fruits, legumes/pulses/nuts, fish, monosaturated:saturated fat ratio, meat/poultry, dairy, and alcohol (3). The MyPyramid Equivalents Database, 2.0 for USDA Survey Foods, 2003–2004 was used to compute ounce/cup equivalents of each food item/day. These calculated food equivalents were then summed within each food group. MeD score was derived following Trichopoulou et al. (3), but with the modification of replacing total-grains with whole grains, due to high intake of refined grains in this population (30,31). The residual method was used to adjust the food equivalents of each food group for total energy intake (32). For each of the positive food components of MeD (whole grains, vegetables, fruits, legumes, fish, and monosaturated:saturated fat ratio), 1 point was assigned when intake was greater than is equal to recommendations by sex-specific median cutoff, versus 0 otherwise. Conversely, for the meat and dairy product groups, 1 point was assigned for intakes below the recommended sex-specific median cutoff. For alcohol, moderate intake (men ≤2 drinks/day; women ≤1 drink/day) was assigned 1 point, with 0 for either more than this or none (Supplementary Table 4). A MeD score for each participant was then computed by summing the points from these 9 different food components; ranging from 0 to 9 (30), with higher score indicating better adherence.

Outcome: Self-reported Depressive Symptomatology

The CES-D scale contains 20 questions about depressive symptoms experienced in the week preceding the interview, including poor appetite, feelings of fear, loneliness, sadness, hopelessness, helplessness, and worthlessness, overall sleep quality and crying spells. Four-point Likert scales (0 = rarely/never, 1 = some or few times, 2 = occasionally or moderate amount, 3 = most or all the time) were used for responses. The final CES-D score was the sum of these scores, ranging from 0 to 60, with high scores representing more severe depressive symptomatology (33). This instrument has been demonstrated to be reliable in the Puerto Rican adult population (27,34) and correlates with Diagnostic and Statistical Manual of Mental Disorders IV (DSM-IV) categories of depression among Hispanics (35). Previous studies have reported Cronbach α = 0.90 for a CES-D scale component in the BPRHS (27). A score of 16 or higher generally indicates depressive symptomatology (33). As expected for this disadvantaged population, a very high burden of depression has been observed in the BPRHS cohort (27), where almost 60% of the participants reported CES-D score ≥16 at baseline and hence, the CES-D score was used as a continuous instead of a dichotomous outcome variable at 3 time-points: baseline, 2y, and 5y in our analysis.

Assessment of Covariates

Covariates were selected based on previous literature as well as their respective associations with MeD pattern and depressive disorders. For our analyses, we included several sociodemographic factors, such as age, sex, body mass index (BMI), energy intake education status, marital, and food security status, income to poverty ratio, psychological acculturation as well as social support (tangible and emotional).

We also included information on participants medical history, health, and lifestyle factors, such as smoking, physical activity, plasma pyridoxal-5′-phospate (PLP) concentration, chronic medical conditions as well as antidepressant medication use variables in our analyses. Information on all variables except the food security status was collected at all 3 time-points. Food security status (secure vs insecure) was collected at baseline and 5y visit; hence participant’s baseline food security status was carried forward to their 2y visit.

We used a 3-category variable for sex, incorporating women’s estrogenic status, and categorized as male, female taking estrogen and/or premenopausal, and female not taking any estrogen and/or postmenopausal. During the interview participant’s body weight and height was measured by a trained field worker and the BMI (kg/m²) was calculated. We used BMI as a continuous variable in our analysis. Physical activity was measured as a weighted score of daily activities (sedentary, light, moderate, or vigorous) plus hours spent sleeping per 24 hours. It was measured with self-reported modified Paffenbarger questionnaire of the Harvard Alumni Activity Survey, which has been previously validated for adult Puerto Rican population (36–38). Education was assessed as no schooling/less than 9th grade, 9th to 12th grade/GED or at least some college education. Smoking was evaluated as never, past but not current, or current. A chronic medical conditions score was derived by summing self-reported chronic medical conditions including diabetes, hypertension, arthritis, heart attack, heart disease (other than heart attack), stroke, respiratory disease, liver or gallbladder disease, kidney disease, Parkinson disease, cancer except skin, Tuberculosis, Hepatitis (Type A, B, or C), and AIDS/HIV. The mean imputation technique was used if response to 3 or fewer medical conditions was missing; the score was coded missing otherwise. Vitamin B6 was measured as plasma PLP in nmol/L. Marital status was categorized as married/living with spouse or partner in the household versus not married/spouse not in the household, divorced or widowed. We evaluated social support using the Norbeck Social Support Questionnaire. Tangible support included questions about actual physical or financial help such, and emotional support included questions about emotional nurturance/feeling loved/cared for or respected by participants’ social network such as, how much each person in participant’s network makes them feel liked, loved, respected, or admired, supports their actions and thoughts etc. Participants were asked to report the size of their social network (up to 16 individuals) and for each person in their network, to answer questions about tangible and emotional support on a Likert 0–4 scale. These scores were used on a continuous scale, with maximum attainable score of 128 for tangible support and 256 for emotional support.

Statistical Analyses

Descriptive statistics, by MeD tertiles, were assessed at baseline using t test or ANOVA for continuous variables, and chi-square for categorical variables.

Cross-sectional Analyses

We used multivariable linear regression to examine cross-sectional associations between MeD score tertiles and CES-D score at baseline among 1 404 participants with complete baseline CES-D and MeD measures. We fit a series of nested models to assess this association: in Model 1, we adjusted for age (y), sex/estrogenic status, BMI (kg/m²), energy intake (kcal/day), and education status. Model 2 additionally included physical activity score and smoking status. In Model 3, we further adjusted for chronic medical conditions score, plasma PLP concentration (vitamin B6) with natural log transformation, marital status, food security status, and tangible support.

Longitudinal Analyses

We used linear mixed-effects models to evaluate the longitudinal association between MeD adherence and CES-D across 3 waves of the study (baseline; 2y; 5y). These models used time-varying MeD score as the exposure, with adjustment for time-varying covariates: age, sex/estrogenic status, BMI, energy intake, and education status in Model 1; additionally, adjusting for physical activity score and smoking status in Model 2; adding chronic medical conditions score, log-transformed plasma PLP, marital status, food security status, and tangible support in Model 3. While both emotional and tangible support were assessed in our cohort, due to high collinearity between tangible and emotional support (Pearson correlation = 0.86), we adjusted only for tangible support in our main analyses along with other covariates. We conducted sensitivity analyses adjusting for emotional support in place of tangible support. We also conducted additional sensitivity analyses adjusting for income to poverty ratio, psychological acculturation, and antidepressant medication use (yes/no).

The lme package in R was used to perform linear mixed-effect modeling. We used the auto-regressive covariance structure for fitting the random intercept-random slope repeated measures mixed-effect model. Model fit was assessed with the Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), and log-likelihood ratio (LL) statistics. We selected the model that provided lower AIC, BIC, and higher log likelihood values for best model fit (39). We assessed the linear trend using a linear variable representing medians of the MeD tertiles. To examine whether baseline MeD adherence was associated with 5y CES-D trajectory, we fit a fully adjusted LMM (eg, Model 3 covariates as outlined previously) with baseline exposure and predictors, adding multiplicative terms between time and each of the MeD tertiles.

To assess the impact of loss to follow-up in our cohort, we repeated analyses among participants with complete data on MeD adherence and CES-D scores at all 3 time-points. We also compared participants with and without complete data on MeD score and CES-D at all 3 time-points. Because diabetes may confound the relationship between MeD diet and depressive symptomatology, and due to the high prevalence of diabetes in our cohort, we further conducted sensitivity analyses restricted to participants without diabetes. All analyses were conducted using R statistical software, version 3.5.3.

Results

At baseline, the mean age of participants was 57.2 ± 7.6 years (y; mean ± SD). Mean BMI, physical activity score and vitamin B6 (PLP) concentration were 31.9 kg/m², 31.4, and 56.4 nmol/L, respectively. Participants reported a mean of 2.8 chronic medical conditions. Baseline CES-D score ranged from 0 to 60 (mean ± SD = 20.2 ± 13.2), 2y range = 0–54 (mean ± SD = 17.9 ± 12.4), and 5y range = 0–52 (mean ± SD = 16.0 ± 11.1).

Participants in the highest tertile of MeD adherence had higher physical activity score, more chronic medical conditions, higher plasma PLP concentration and somewhat higher tangible support, compared to those in the lowest MeD tertile. They were also less likely to be heavy alcohol users, less likely to be nondiabetic and more likely to be living with spouse/partner than those in the lowest MeD tertile. Mean age, BMI, income to poverty ratio, psychological acculturation, emotional support score, and proportions for sex, education, smoking status, food-security status, and antidepressant use categories did not differ significantly across tertiles of MeD score (Table 1 and Supplementary Table 5).

Table 1.

Baseline Characteristics by Tertile of Mediterranean Diet (MeD) Adherence Score (n = 1 404)

	Tertile 1* n = 525	Tertile 2* n = 671	Tertile 3* n = 208
				p †
MeD score‡	3.4 ± 0.8	5.4 ± 0.5	7.3 ± 0.5	―
Age (y)	57.3 ± 7.7	57.1 ± 7.4	57.0 ± 7.7	.87
BMI (kg/m2)	31.7 ± 6.7	32.0 ± 6.5	32.4 ± 6.7	.43
Physical activity score§	31.0 ± 4.6	32.0 ± 4.8	31.3 ± 4.1	.012*
Chronic medical conditions score‖	2.8 ± 1.7	2.7 ± 1.6	3.0 ± 1.6	.049*
Plasma PLP (nmol/L)	51.6 ± 42.8	58.3 ± 62.5	62.5 ± 75.3	.039*
Tangible support¶	29.6 ± 20.5	31.5 ± 19.6	33.3 ± 21.1	.061
Sex				.66
Male	150 (28.6)	193 (28.8)	53 (25.5)
Female (estrogen/premenopausal)	79 (15.0)	85 (12.7)	31 (14.9)
Female (no estrogen/postmenopausal)	296 (56.4)	393 (58.5)	124 (59.6)
Education				.28
<9th grade (ref)	250 (47.7)	299 (44.6)	106 (51.0)
9th–12th grade/GED	202 (38.6)	256 (38.1)	75 (36.0)
At least some college	72 (13.7)	116 (17.3)	27 (13.0)
Smoking				.16
Never	237 (45.1)	306 (45.7)	96 (46.2)
Past smoker but not current	145 (27.6)	204 (30.4)	72 (34.6)
Current smoker	143 (27.2)	160 (23.9)	40 (19.2)
Marital status				.007**
Married/ living with spouse/partner in household	130 (24.9)	211 (31.5)	73 (35.1)
Spouse divorced/widowed/ never married/ not in household	393 (75.1)	458 (68.5)	135 (64.9)

Note: BMI = body mass index; SD = standard deviation; PLP = pyridoxal-5′-phospate.

*Values are mean ± SD for continuous and frequency (percent) for categorical variables.

^† p ANOVA for continuous and p Chi-square for categorical variables.

^‡Adherence to 9 components; whole grains, vegetables, fruit, legumes/pulses/nuts, fish, monounsaturated/saturated fat intake ratio, meat/poultry, dairy, and alcohol; range = 0–9.

^§Weighted score of hour/day sleeping and lying down, performing sedentary, light, moderate, or vigorous activities; range = 0–120, observed baseline range = 24.3–62.6.

^‖Computed using mean imputation technique (conditions include diabetes, hypertension, arthritis, heart attack, heart disease; other than heart attack, stroke, respiratory disease, liver or gallbladder disease, kidney disease, Parkinson disease, cancer except skin, tuberculosis, hepatitis, and HIV/AIDS), observed baseline range = 0–10.2.

^¶Physical or financial support from NSSQ; range = 0–128.

*p < .05; ** p < .01; *** p < .001.

At baseline, after adjusting for covariates, greater MeD adherence score was significantly associated with lower CES-D score (β = −2.0, 95% CI = −4.0, −0.04 for highest vs lowest tertile, p trend = .04; Table 2, Model 3§). Similarly, in the fully adjusted longitudinal model, greater MeD score was significantly associated with less depressive symptomatology (CES-D) over 5 years (β = −1.9, 95% CI = −3.0, −0.8 for highest vs lowest tertile, p trend = .0005; Table 3, Model 3§).

Table 2.

Cross-sectional Association Between Mediterranean Diet (MeD) Score and Self-reported Depressive Symptoms in Puerto Rican Adults at Baseline (n = 1 404)

	MeD Tertiles
	Tertile 1 (0–4)	Tertile 2 (5–6)	Tertile 3 (7–9)	p Trend*
		β (95% CI)	β (95% CI)
Model 1†	Ref	−2.3 (−3.7, −0.8)	−2.9 (−4.9, −0.8)	.004**
Model 2‡	Ref	−1.9 (−3.3, −0.4)	−2.5 (−4.5, −0.4)	.01*
Model 3§	Ref	−1.3 (−2.7, 0.08)	−2.0 (−4.0, −0.04)	.04*

Note: CI = confidence interval; BMI = body mass index; PLP = pyridoxal-5′-phospate.

*Using a linear variable representing tertile medians.

^†Adjusted for age, sex, BMI, energy intake, and education level (n = 1 392).

^‡Adjusted for († + physical activity score and smoking; n = 1 391).

^§Adjusted for (^‡ + chronic medical conditions score, log-transformed plasma PLP, marital status, food security status, and tangible support; n = 1 301).

Table 3.

Mediterranean Diet in Relation to Depressive Symptomatology (CES-D Score) Over 5 Years in Puerto Rican Adults

	MeD Tertiles
	Tertile 1 (0–4)	Tertile 2 (5–6)	Tertile 3 (7–9)	p Trend*
		β (95% CI)	β (95% CI)
Model 1†	Ref	−1.01 (−1.8, −0.2)	−2.0 (−3.1, −0.9)	.0003***
Model 2‡	Ref	−1.0 (−1.7, −0.2)	−1.8 (−3.0, −0.7)	.001**
Model 3§	Ref	−0.6 (−1.4, 0.1)	−1.9 (−3.0, −0.8)	.0005***

Note: CI = confidence interval; BMI = body mass index; MeD = Mediterranean diet; PLP = pyridoxal-5′-phospate.

*Using a linear variable representing tertile medians.

^†Linear mixed effects model adjusted for age, sex, BMI, energy intake, education level, and time (visit number; 3 301 observations; 1 454 groups).

^‡Adjusted for † + physical activity score, smoking (3 279 observations; 1 453 groups).

^§Adjusted for ^‡ + chronic medical conditions score, log-transformed plasma PLP, marital status, food security status, and tangible support (3 053 observations; 1 407 groups).

Baseline MeD score did not modify the 5y CES-D trajectory (p = .49 for MeD_Tertile2 × time and p = .92 for MeD_tertile3 × time; Table 4, Model 3§). While overall CES-D score in the highest MeD tertile was consistently lower than the lower MeD tertiles, we observed parallel slopes, suggesting that depressive symptoms in this cohort improved over 5 years of follow-up, but not differentially according to extent of MeD adherence.

Table 4.

Longitudinal Associations Between Baseline Tertiles of Mediterranean Diet (MeD) Adherence and Depressive Symptomatology (CES-D) Trajectory Over 5 Years of Follow-up in the Boston Puerto Rican Health Study

	Model 1*			Model 2†			Model 3‡
	β	(SE)	p Value	β	(SE)	p Value	β	(SE)	p Value
MeD_Tertile 1	Ref			Ref			Ref
MeD_Tertile 2	−1.82	0.73	.01	−1.49	0.71	.04	−1.02	0.69	.14
MeD_Tertile 3	−2.60	1.02	.01	−2.26	1.00	.02	−1.79	0.96	.06
Time	−0.87	0.13	<.0001	−0.86	0.13	<.0001	−0.79	0.13	<.0001
MeD_Tertile2 × Time	−0.04	0.17	.79	−0.05	0.17	.75	−0.12	0.17	.49
MeD_Tertile3 × Time	−0.03	0.23	.89	−0.03	0.23	.89	−0.02	0.24	.92

Note: BMI = body mass index; PLP = pyridoxal-5′-phospate; SE = standard error.

*Linear mixed effects model adjusted for age, sex, BMI, energy intake, education level, and time (visit number; 3 456 observations, 1 392 groups).

^†Adjusted for * + physical activity score, smoking (3 438 observations, 1 386 groups).

^‡Adjusted for ^† + chronic medical conditions score, log-transformed plasma PLP, marital status, food security status, and tangible support (3 229 observations, 1 296 groups).

Sensitivity analyses among 713 participants with complete data on MeD and CES-D score yielded similar results as the main analyses (Supplementary Table 2, Model 3c). MeD adherence was associated with lower CES-D score over 5 years of follow-up (β = −1.8, 95% CI = −3.2, −0.5 for highest vs lowest tertile, p trend = .006). Participants with complete data on MeD score and CES-D at all visits were younger (p Anova = .036), had higher BMI (p Anova = .016), had somewhat higher tangible support (p Anova = .066), and were more likely to be married/living with partner in household as well as nondiabetic than those missing at least 1 measure of MeD score or CES-D (p chi square = .006). No other covariate differed significantly between these 2 groups (Supplementary Table 3).

Results of analyses restricted to 321 participants without diabetes at any visit (Supplementary Table 1, Model 3c) were also similar to our main findings. As in other analyses, MeD adherence was associated with lower CES-D score among those without diabetes (β = −1.7, 95% CI = −3.2, −0.2 for highest vs lowest tertile, p trend = .03). Adjusting for emotional support instead of tangible support did not alter the study conclusions (β = −1.9, 95% CI = −3.1, −0.8 for highest vs lowest tertile, p trend = .0005).

Discussion

In this study of Boston Area Puerto Rican adults, participants with better MeD adherence reported lower depressive symptoms at baseline and over 5 years follow-up, compared to those with the lower MeD adherence. However, baseline MeD adherence did not alter the trajectory in CES-D scores over the 5y follow-up: depressive symptomatology improved over time for all participants, irrespective of their MeD adherence. Results were consistent in sensitivity analyses among participants with complete data on MeD adherence and CES-D over the 3 follow-up waves, as well as in analyses restricted to those without diabetes.

Our study results suggest that participants with better MeD diet adherence experienced less depressive symptomatology, which is in agreement with prior, mostly cross-sectional studies (10,12). We observed a consistent inverse association between MeD adherence and depression over 3 time-points in this study that persisted after controlling for several covariates. However, baseline MeD adherence did not alter the trajectory of depressive symptomatology. These results are consistent with the study that has shown no association between MeD and incident depression (9). Similar to our study, a randomized control trial among participants at high-risk for cardiovascular disease showed no significant effect of a MeD intervention supplemented with nuts on incident depression, compared to the group following a low-fat diet overall, although an effect of MeD diet was observed among a subgroup of participants with type 2 diabetes (9). A prospective study among 166 participants with major depressive disorder, reported that MeD diet adherence was inversely associated with depressive symptoms at baseline, but did not predict any significant improvement in these symptoms over the period of 12 months (40).

In contrast, some studies have reported a protective effect of MeD adherence on incident depression. In one longitudinal study, lower annual incidence of CES-D-based depressive symptomatology was observed among those in the highest versus lowest, tertile of adherence to MeD pattern (41). In a cohort of initially healthy adults, reduced risk of self-reported clinical depression was associated with better MeD adherence, with a dose–response relationship (HR = 0.58, 95% CI: 0.44,0.77, p trend < .001) (42). Another recently published longitudinal study highlighted protective effects of DASH (Dietary Approaches to Stop Hypertension; β = −0.10, CI: −0.20, −0.0064) and MIND (Mediterranean-DASH Intervention for Neurodegenerative Delay; β = −0.12, CI: −0.23, −0.0092) diet on depression over 6 years of follow-up (43). A meta-analysis by Psaltopoulou et al. showed a consistent protective effect of higher adherence to MeD against depression, reporting 32% reduced risk of depression for the highest MeD score (6–9) compared to the lowest (0–3). (RR = 0.68, 95% CI = 0.54–0.86) (44). Still several other studies reported no association between MeD Diet and depression, even cross-sectionally, or association only with certain MeD components (45–47). For example, one cross-sectional study among older adults observed that higher fish intake, but not overall MeD adherence, was associated with lower severity of depression (45).

Prior work in the BPRHS cohort has suggested that adherence to a diet with MeD characteristics may be beneficial for brain health. In cross-sectional analysis, the highest MeD intake score (6–9) was associated with 13% lower odds of cognitive impairment (30). Similarly, significantly better cognitive health was seen in individuals with diabetes, over 2y of follow-up, with less decline in global cognitive function and memory associated with higher MeD adherence (6). Our findings on depressive symptomatology add to the body of literature regarding the protective role of MeD in a wide range of neurological and neuropsychiatric disorders.

Strengths and Limitations

Our study has several strengths, including its longitudinal design, repeated measures of depressive symptomatology at 3 time-points, dietary assessment through a detailed FFQ specifically developed and validated for this population, and inclusion of several key covariates. We used the CES-D, a reliable measure for assessing depression (27,34), which has been shown to correlate with DSM-IV categories of depression among Hispanics (35). The MeD adherence score used fits well with the traditional Puerto Rican diet that has several similarities, stemming from an island environment with Spanish culture (48). Therefore, adherence to the MeD pattern may be more culturally relatable and accessible for this population than other diet indices.

Our study has several limitations. Our study population was subject to substantial low to follow-up with 35% of the cohort lost to follow-up over the 5y study period. When comparing the characteristics of participants with complete measures of MeD score and CES-D at all 3 time-points to those who were missing data on either MeD or CES-D at any time point, we observed that, while some participant characteristics, such as marriage status, diabetes, age and tangible support, differed between the retained and lost groups of participants, most characteristics, including, importantly, the exposure (MeD) and outcome (CES-D) scores were similar, suggesting that differential loss to follow-up is unlikely to be a major driver of our findings. Furthermore, analyses restricted to participants who had complete data on MeD adherence and depressive symptomatology (CES-D score) at all three visits led to results similar to our main analyses, further strengthening the validity of our findings.

Our FFQ-derived dietary data was based on self-report, which is subject to recall and measurement error. However, we used a detailed FFQ specifically developed and validated for the study population, which has been validated in prior studies (28,29). While we adjusted for a number of covariates in our models, residual confounding is still a possibility. Lastly, this study was conducted among an adult Puerto Rican population residing in the United States, with a particular dietary and cultural background, thus our findings might not be fully generalizable to other populations.

In summary, participants with higher MeD diet adherence had lower depression symptomatology over the 5 years of follow-up, but baseline MeD diet adherence did not alter 5-year CES-D trajectory in this study. More large-scale prospective research studies and randomized control trials are needed to assess the role of MeD adherence in the pathogenesis of development, progression, and treatment of depression.

Funding

This work was supported by the National Institutes of Health (P50 HL105185, P01 AG023394, and R01 AG055948).

Conflict of Interest

None declared.

Author Contributions

N.S. conceived the manuscript idea, performed the statistical analysis, wrote the first manuscript draft, and revised subsequent drafts; J.S.L. contributed statistical expertise and reviewed drafts of the manuscript; X.Z. contributed statistical expertise and reviewed drafts of the manuscript; T.S. provided expertise in assessment of depressive symptomatology and reviewed drafts of the manuscript; L.P. provided epidemiological expertise and reviewed drafts of the manuscript; K.L.T. is the PI of the BPHRS study, provided expertise in nutritional epidemiology and reviewed drafts of the manuscript; N.P. provided expertise in development of the manuscript idea and study designing, guided the data analysis, manuscript writing and reviewed drafts of the manuscript.