Adherence to a Mediterranean-style eating pattern and risk of diabetes in a U.S. prospective cohort study

The Mediterranean-style eating pattern is recommended by major U.S. health organizations to reduce chronic disease risk. It is generally plant-based, relatively high in olive oil and seafood, and low in dairy, red meat, and refined grains. Prior studies have linked higher adherence to this pattern with improved cardiovascular risk factors, fewer cardiovascular events, and lower cardiovascular mortality. The American Diabetes Association recommends it to prevent cardiovascular complications in diabetes. However, whether adherence to a Mediterranean-style diet reduces diabetes risk in U.S. populations remains unclear, with inconsistent findings compared to consistent inverse associations observed in Mediterranean/European cohorts. Potential differences in racial/ethnic heterogeneity, BMI, physical activity, social networks, and lifestyle factors between U.S. and Mediterranean/European populations may influence translatability. Given that minority race and higher BMI are key diabetes risk factors in the U.S., and dietary habits vary across these groups, the study aimed to assess associations between Mediterranean pattern adherence (aMed score) and incident diabetes in a U.S. community-based cohort and to examine consistency across racial and BMI subgroups.

Evidence from Mediterranean and European settings shows consistent inverse associations between Mediterranean diet adherence and diabetes risk. In U.S. cohorts, findings have been mixed: some cohorts (e.g., Health Professionals Follow-up Study) reported up to 25% lower diabetes risk with higher Mediterranean diet scores, while others (e.g., Multi-Ethnic Study of Atherosclerosis) found no association overall or within racial subgroups. Additionally, prior research indicates that benefits of healthy dietary patterns may differ by race and BMI, with some evidence of stronger associations in certain subgroups. The broader context suggests that Mediterranean lifestyle factors (physical activity, social networks, rest) may confound or modify diet–disease associations, and U.S. dietary patterns may not align with traditional high-monounsaturated fat Mediterranean diets, complicating comparisons.

Design and population: Prospective cohort analysis within the Atherosclerosis Risk in Communities (ARIC) study. Baseline (visit 1: 1987–1989) included 15,792 adults aged 45–65 years from four U.S. communities. Exclusions: prevalent/history of CVD, diabetes, or cancer (n=3318); implausible energy intake or ≥10 missing FFQ items (n=270); Asian (n=28) or Indian (n=14) due to small numbers; black participants from Maryland (n=25) and Minnesota (n=19) due to race–center confounding; missing baseline covariates (n=20); missing follow-up time; missing food items for aMed score (n=63). Final analytic sample: n=11,991. Diet assessment: A trained interviewer administered a 66-item semi-quantitative food frequency questionnaire (FFQ) at visits 1 and 3, adapted to capture fish, cooking fats, and baked goods. Nutrient intakes were derived using USDA Food Composition Databases via Harvard’s Channing Laboratory. Dietary and alcohol intakes from visits 1 and 3 were averaged; if diabetes developed or censoring occurred before visit 3, visit 1 intake alone was used. Exposure: Alternate Mediterranean Diet (aMed) score (0–9 points) tailored for U.S. populations. Components (1 point each): above sex-specific median intake of vegetables, fruits, whole grains, nuts, legumes, fish, MUFA:SFA ratio; below median intake of red and processed meat; moderate alcohol intake (5–15 g/day for women, 10–25 g/day for men). Outcome: Incident diabetes ascertained via any of: self-reported physician diagnosis; self-reported diabetes medication use in previous 2 weeks; fasting glucose ≥126 mg/dL; non-fasting glucose ≥200 mg/dL. Ascertainment occurred at study visits and annual follow-up calls; participants were followed from visit 1 through December 31, 2016; censoring at death, loss to follow-up, or end of follow-up. Covariates: Demographics (age, sex, race-center), education, smoking status, physical activity (Baecke questionnaire score 1–5), BMI (kg/m²; categories: normal, overweight, obese), blood pressure/hypertension (per JNC7 criteria or antihypertensive use), LDL cholesterol, fasting glucose, and family history of diabetes. Height/weight and BP were measured using standardized protocols. Statistical analysis: Cox proportional hazards models with follow-up time (days) as the timescale estimated hazard ratios (HRs) and 95% CIs for aMed quintiles and per 1-point increase. Trend tests treated quintiles as ordinal. Linear spline modeling (knot at aMed=2) assessed dose–response shape. Proportional hazards assumption evaluated via log–log plots and Schoenfeld residuals. Models: Model 1 adjusted for total energy intake, age, sex, race-center, education; Model 2 additionally adjusted for smoking and physical activity; Model 3 additionally adjusted for fasting glucose, hypertension status, LDL cholesterol, BMI category, and family history of diabetes (to estimate direct association). Effect modification by race and baseline BMI was prespecified; sex and education were tested post hoc. Interactions assessed via likelihood ratio tests. Component analyses examined individual aMed components and incident diabetes. Analyses conducted in Stata 15.

- Sample and follow-up: n=11,991; median follow-up 22 years; 4,024 incident diabetes cases; overall incidence rate 1.7 per 100 person-years. - Overall association (aMed quintiles): Model 1 HR Q5 vs Q1 = 0.83 (95% CI 0.73–0.94); p-trend < 0.001. Model 2 HR = 0.88 (0.77–0.99); Model 3 HR = 0.94 (0.82–1.07); trends remained significant in Models 2 (p=0.005) and 3 (p=0.03). - Per-point association: Model 1 HR per 1-point higher aMed = 0.96 (0.95–0.98); Model 2 = 0.97 (0.96–0.99); Model 3 = 0.98 (0.96–0.99). - Dose–response: Linear inverse relationship for aMed ≥2 points (spline; p<0.001). - Effect modification: • Race: Stronger inverse associations among Black participants than White participants (interaction p<0.001). Example Model 1 estimates shown in figure: Blacks—per 1-point HR 0.95 (0.91–0.98); Q5 vs Q1 HR 0.74 (0.59–0.94). Whites—per 1-point HR 0.97 (0.95–0.99); Q5 vs Q1 HR 0.87 (0.75–1.01). • BMI: Strongest among normal BMI, attenuated in overweight, not significant in obese (interaction p<0.01). Example Model 1: Normal BMI—per 1-point HR 0.94 (0.91–0.98), Q5 vs Q1 HR 0.80 (0.62–1.03); Overweight—per 1-point HR 0.97 (0.94–0.98), Q5 vs Q1 HR 0.88 (0.72–1.06); Obese—per 1-point HR 0.98 (0.95–1.01), Q5 vs Q1 HR 0.88 (0.71–1.08). No interaction by sex or education. - Components (Model 1): Higher nuts HR 0.92 (0.86–0.98); legumes HR 0.92 (0.86–0.98); moderate alcohol HR 0.83 (0.75–0.91); lower red/processed meat HR 0.91 (0.84–0.97) associated with lower diabetes risk. Whole grains, vegetables, fruits, MUFA:SFA, and fish components were not independently associated. In Model 3, only alcohol remained significant [HR 0.81 (0.74–0.90)].

The study addressed whether adherence to a Mediterranean-style eating pattern is linked to lower diabetes risk in a diverse U.S. cohort. Higher aMed scores were associated with reduced incident diabetes overall, with attenuation after adjustment for behavioral and clinical factors, suggesting part of the association may be mediated through these factors. The inverse association was more pronounced among Black participants than White participants and strongest among individuals with normal BMI, attenuating in overweight and disappearing in obese participants. This pattern implies that excess adiposity may overwhelm potential dietary benefits for diabetes prevention absent weight loss. Component analyses suggest nuts, legumes, moderate alcohol, and lower red/processed meat intake primarily drove the association. Findings help reconcile inconsistent U.S. evidence by demonstrating significant associations in a large cohort with long follow-up and by revealing subgroup differences by race and BMI. Comparisons with European/Mediterranean cohorts point to possible contextual and lifestyle differences (e.g., physical activity, social networks, rest) that were not fully captured and may contribute to stronger associations observed abroad. The FFQ-based aMed in this U.S. setting reflected a pattern higher in plant foods but not a traditional high-olive-oil, high-monounsaturated-fat diet, limiting direct inference to classic Mediterranean diets. Overall, results support promoting Mediterranean-style dietary patterns as part of comprehensive lifestyle strategies for diabetes prevention, particularly in Black and normal-weight populations, while emphasizing the importance of achieving and maintaining healthy body weight.

A Mediterranean-style eating pattern characterized by higher intakes of fruits, vegetables, whole grains, legumes, nuts, and fish, alongside moderate alcohol consumption, was associated with a lower risk of incident diabetes in a large, community-based U.S. cohort. Benefits were stronger among Black and normal-weight individuals and absent among those who were overweight or obese, underscoring the critical role of body weight in diabetes prevention. Public health efforts to promote healthy eating patterns, particularly in predominantly Black communities, may help reduce diabetes burden. Future research should test whether a calorically restrictive Mediterranean-style diet that induces clinically meaningful weight loss can reduce diabetes risk in overweight and obese populations and should further disentangle dietary versus broader lifestyle components of the Mediterranean pattern.

- Dietary assessment limitations: The ARIC FFQ was not designed specifically for Mediterranean diet assessment and did not capture olive oil intake, a key component; limited variation and item coverage for fruits, vegetables, and whole grains may have obscured component-specific associations. - Pattern misalignment: The aMed score in this U.S. cohort did not reflect a traditional high total and monounsaturated fat (olive oil–rich) Mediterranean diet; higher aMed quintiles had lower total/saturated fat with little variation in mono- and polyunsaturated fats, limiting applicability to classic Mediterranean diets. - Self-report and measurement error: Dietary intake and alcohol consumption were self-reported; energy intake estimates from FFQs are imprecise, and no objective energy validation (e.g., doubly labeled water) was available. Although averaging two time points likely reduced error, residual misclassification remains. - Outcome classification: Incident diabetes included self-report and glucose measures; type 1 vs type 2 diabetes could not be distinguished, though most cases are presumed type 2 given age and exclusions. - Residual confounding: Despite multivariable adjustment (including energy intake, behaviors, and clinical factors), unmeasured/lifestyle factors (e.g., social networks, sleep/rest patterns) could confound associations. - Effect modification by adiposity: Lack of association in overweight/obese groups suggests potential confounding/mediation by adiposity and limits generalizability of benefits to those with excess weight absent weight loss.