Yearly attained adherence to Mediterranean diet and incidence of diabetes in a large randomized trial

Abstract

Background: Several large observational prospective studies have reported a protection by the traditional Mediterranean diet against type 2 diabetes, but none of them used yearly repeated measures of dietary intake. Repeated measurements of dietary intake are able to improve subject classification and to increase the quality of the assessed relationships in nutritional epidemiology. Beyond observational studies, randomized trials provide stronger causal evidence. In the context of a randomized trial of primary cardiovascular prevention, we assessed type 2 diabetes incidence according to yearly repeated measures of compliance with a nutritional intervention based on the traditional Mediterranean diet.

Methods: PREDIMED (''PREvención con DIeta MEDiterránea'') was a Spanish trial including 7447 men and women at high cardiovascular risk. We assessed 3541 participants initially free of diabetes and originally randomized to 1 of 3 diets: low-fat diet (n = 1147, control group), Mediterranean diet supplemented with extra virgin olive (n = 1154) or Mediterranean diet supplemented with mixed nuts (n = 1240). As exposure we used actual adherence to Mediterranean diet (cumulative average), yearly assessed with the Mediterranean Diet Adherence Screener (scoring 0 to 14 points), and repeated up to 8 times (baseline and 7 consecutive follow-up years). This score was categorized into four groups: < 8, 8-< 10, 10- < 12, and 12-14 points. The outcome was new-onset type 2 diabetes.

Results: Multivariable-adjusted hazard ratios from time-varying Cox models were 0.80 (95% confidence interval, 0.70-0.92) per + 2 points in Mediterranean Diet Adherence Screener (linear trend p = .001), and 0.46 (0.25-0.83) for the highest (12-14 points) versus the lowest (< 8) adherence. This inverse association was maintained after additionally adjusting for the randomized arm. Age- and sex-adjusted analysis of a validated plasma metabolomic signature of the Mediterranean Diet Adherence Screener (constituted of 67 metabolites) in a subset of 889 participants also supported these results.

Conclusions: Dietary intervention trials should quantify actual dietary adherence throughout the trial period to enhance the benefits and to assist results interpretation. A rapid dietary assessment tool, yearly repeated as a screener, was able to capture a strong inverse linear relationship between Mediterranean diet and type 2 diabetes. Trial registration ISRCTN35739639.

Keywords: Diabetes; Dietary assessment tools; Feeding trial; Monounsaturated fats; Nutritional epidemiology; Olive oil.

Fig. 1

Hazard Ratios (HR) of developing type 2 diabetes according to cumulative adherence to the Mediterranean diet (0 to 14 score of the MEDiterranean diet Adherence Screener or MEDAS). Multivariable adjusted hazard ratios for successive categories of the MEDAS (cumulative averages using yearly repeated measures) according to estimates adjusted only for age and sex (grey bars) or according to estimates with multivariable adjustment for age, sex, baseline smoking status (never, current, or former smoker), fasting glucose level, prevalence of dyslipidemia (yes/no) and hypertension (yes/no), total energy intake level (kcal/d), physical activity level (metabolic equivalent of min/d), education level (primary education, secondary education, academic/graduate), propensity scores for group allocation and randomized group allocated (adjacent bars in sand color). All estimates were stratified by recruitment center, and robust SEs were used. MEDAS Mediterranean Diet Adherence Screener, CI confidence interval.

Fig. 2

Cumulative incidence (Nelson-Aalen curves) of diabetes according to average MEDAS scores during years 1 to 7 and randomized intervention group. MEDAS scores are dichotomized, where ≥ 10 points are regarded as high adherence. Both intervention groups with Mediterranean Diet have been merged and they were compared against the control group. Adjusted for age, sex, baseline smoking status (never, current, or former smoker), prevalence of dyslipidemia (yes/no) and hypertension (yes/no), family history of CVD, total energy intake level (kcal/d), physical activity level (metabolic equivalent of min/d), education level (primary education, secondary education, and academic/graduate) and recruitment center using inverse probability weighting