The worldwide incidence of people with type 2 diabetes mellitus (T2DM) is projected to rise from 171 million in 2000 to 380 million in 2025, causing enormous costs for health-care systems( Reference Hossain, Kawar and El Nahas 1 ). The typical Mediterranean diet (MD), which was first postulated by Ancel Keys in the 1960s( Reference Keys, Menotti and Karvonen 2 ), is characterized by high intakes of MUFA, vegetables and fruits, plant proteins, whole grains, fish and low-fat dairy products, moderate alcohol (red wine) intake, and low red meat consumption( Reference Willett, Sacks and Trichopoulou 3 ). In the Seven Countries Study, Keys reported the beneficial effects of an MD on human health( Reference Keys, Menotti and Karvonen 2 ). Since then, dietary pattern analysis has been widely used to detect the cumulative effects of diet on health. T2DM is a major chronic disease originating from different genetic, behavioural and environmental risks( Reference Neel 4 ), with impaired glucose tolerance being an indicator for the development of a diabetic metabolic state( Reference Group 5 ). Lifestyle intervention studies in randomized trial settings have demonstrated a reduction in the relative risk to develop T2DM of 30–67 %( Reference Yamaoka and Tango 6 ) with sustainable long-term reduction of incidence( Reference Lindstrom, Ilanne-Parikka and Peltonen 7 , Reference Li, Zhang and Wang 8 ), although a need to develop standardized lifestyle interventions may be necessary to enhance their efficacy( Reference Yoon, Kwok and Magkidis 9 ). MD is a homogeneous and straightforward construct having its roots in southern European eating patterns. Adherence to an MD is associated with significant improvements in health status ranging from reduced risk of mortality to neurodegenerative diseases( Reference Sofi, Abbate and Gensini 10 , Reference Schwingshackl and Hoffmann 11 ). However, to date no systematic review and meta-analysis has summarized the effects of MD adherence on the risk of T2DM. Therefore, the aim of the present study was to meta-analyse randomized controlled trials and cohort studies that have assessed the effect of adherence to an MD on the risk of T2DM.
Literature searches were performed using the electronic databases MEDLINE, SCOPUS, EMBASE and the Cochrane Trial Register until 2 April 2014 using the following keywords: ‘Mediterranean diet AND diabetes’. Moreover, the reference lists from retrieved articles were checked to search for further relevant studies, and systematic reviews and meta-analysis were searched by two authors. The literature search as well as article abstraction was conducted independently by two authors (L.S. and B.M.). Relevance of studies was assessed with a hierarchical approach (title, abstract and full manuscript). After an initial screening of titles and abstracts, studies included by both reviewers were compared and disagreement was resolved by consensus.
Studies were included in the meta-analysis if they met all of the following criteria: (i) randomized controlled trial or cohort study design; (ii) reported adjusted relative risks (RR) for T2DM with corresponding 95 % confidence intervals; (iii) evaluated the association of an a priori score used for assessing adherence to an MD; and (iv) when a study seemed to have been published in duplicate, the version containing the most comprehensive information was selected. Studies including pregnant women were excluded, as were case–control studies, cross-sectional studies and studies that were not published as original reports (such as conference abstracts, letters to the editor without original data, commentaries and reviews).
Full copies of studies were independently assessed for methodological quality by two authors (L.S. and B.M.) using the Jadad score for randomized controlled trials and the Newcastle Ottawa Scale for cohort studies( Reference Jadad, Moore and Carroll 12 , Reference Wells, Shea and O’Connell 13 ) (Table 1).
MD, Mediterranean diet; RCT, randomized controlled trial; FPG, fasting plasma glucose; OGGT, oral glucose tolerance test; NDDG, National Diabetes Data Group; ADA, American Diabetes Association; MI, myocardial infarction; GDM, gestational diabetes mellitus; n.d., not determined; M, men, W, women; PA, physical activity; WC, waist circumference; NYHA, New York Heart Association; CCS, Canadian Cardiovascular Society; ACE, angiotensin-converting enzyme; WHR, waist:hip ratio.
ADA criteria: symptoms of diabetes plus casual plasma glucose concentration ≥11·1 mmol/l or FPG ≥7·0 mmol/l, or 2 h post-load glucose ≥11·1 mmol/l during an OGGT; in the absence of unequivocal hyperglycaemia.
* Jadad score.
Two investigators (L.S. and B.M.) independently collected the relevant reports using a standardized form. The following data were extracted from each study: (i) first author’s last name; (ii) year of publication; (iii) country of origin; (iv) sample size; (v) study length; (vi) age at entry; (vii) sex; (viii) T2DM diagnosis criteria; (ix) study design; (x) components of the MD score; (xi) adjustment factors; (xii) quality score; (xiii) dietary assessment method; and (xiv) risk estimates with their corresponding 95 % confidence intervals. When a study provided several risk estimates, the maximally adjusted models were chosen.
The meta-analysis was performed with the adjusted RR of the highest compared with the lowest MD adherence score based on random-effects modelling according to DerSimonian and Laird incorporating both within- and between-study variability. To evaluate the weighting of each study, the se for log(RR) of each study was calculated and regarded as the estimated variance of log(RR)( Reference DerSimonian and Laird 14 ). Random-effects modelling with restricted maximum likelihood estimate, which is more appropriate if the number of included studies is small, was also used to confirm the final risk estimates. Heterogeneity was estimated by the Cochrane Q test together with the I 2 statistic. An I 2 value >50 % indicates substantial heterogeneity across studies( Reference Higgins, Thompson and Deeks 15 ). Funnel plots were used to assess potential publication bias. To determine the presence of publication bias, the symmetry of the funnel plots in which mean differences were plotted against their corresponding se was assessed. In addition, the Egger test was performed to test for potential publication bias( Reference Egger, Davey Smith and Schneider 16 ).
Nine papers extracted from 1408 articles met the eligibility criteria and were included in the quantitative analysis( Reference Abiemo, Alonso and Nettleton 17 – Reference Salas-Salvado, Bullo and Estruch 25 ). One cohort study was excluded since the primary outcome was the risk of gestational diabetes( Reference Tobias, Zhang and Chavarro 26 ). The detailed steps of the meta-analysis article selection process are described as a flow diagram in Fig. 1.
One randomized controlled trial and eight prospective cohort studies (sample size: 122 810 subjects) published between 2007 and 2014 were included for meta-analysis. Study duration plus follow-up ranged between 3·2 and 20 years. All cohort studies used an FFQ for dietary assessment. General study characteristics are summarized in Table 1.
To assess the quality of the included studies, a 9-point scoring system was used according to the Newcastle Ottawa Scale. A score of ≥7 points was defined as high study quality study (see online supplementary material, Table S1); all cohort studies were of high quality. The cut-offs for the high v. low MD score differed between the studies (three cohorts used quintiles( Reference Abiemo, Alonso and Nettleton 17 , Reference de Koning, Chiuve and Fung 18 , Reference Mozaffarian, Marfisi and Levantesi 20 ), two studies used quartiles( Reference Rossi, Turati and Lagiou 23 , Reference Tobias, Zhang and Chavarro 26 ) and tertiles( Reference Martinez-Gonzalez, de la Fuente-Arrillaga and Nunez-Cordoba 19 , Reference Romaguera and Guevara 22 ), and one study used cluster analysis( Reference Brunner, Mosdol and Witte 21 )).
Mediterranean diet and risk of diabetes
The pooled RR for the highest v. lowest adherence to the MD score was 0·81 (95 % CI 0·73, 0·90, P<0·0001, I 2=55 %; Fig. 2). Stratified RR are summarized in Table S2 (see online supplementary material). For studies with a long-term follow-up (≥10 years), an RR of 0·75 (95 % CI 0·68, 0·83, P<0·00001, I 2=0 %) was found, and for large sample size (≥10 000), an RR of 0·82 (95 % CI 0·72, 0·94, P=0·005, I 2=67 %) could be observed. There was a significant association in the European analysis (RR=0·81; 95 % CI 0·71, 0·93, P=0·002) but not in the US analysis (RR=0·82; 95 % CI 0·68, 1·00, P=0·05). Exclusion of studies by Tobias et al.( Reference Tobias, Hu and Chavarro 24 ) (history of gestational diabetes) and Mozaffarian et al.( Reference Mozaffarian, Marfisi and Levantesi 20 ) (secondary prevention of CVD) confirmed the results of the main analysis (RR=0·84; 95 % CI 0·75, 0·94, P=0·003, I 2=57 %). Since cut-off and MD scores varied between studies, we performed a sensitivity analysis investigating the association between a 2-point increase of adherence to the MD score (defined by Trichopoulou( Reference Trichopoulou, Costacou and Bamia 27 )) and the risk of T2DM. Six of nine( Reference Abiemo, Alonso and Nettleton 17 – Reference Martinez-Gonzalez, de la Fuente-Arrillaga and Nunez-Cordoba 19 , Reference Romaguera and Guevara 22 , Reference Rossi, Turati and Lagiou 23 , Reference Tobias, Zhang and Chavarro 26 ) studies were included and showed a significant risk reduction (RR=0·93; 95 % CI 0·89, 0·98, P=0·007, I 2=69 %; see online supplementary material, Fig. S1).
The funnel plots for diabetes incidence indicated low asymmetry (with the exception of one outlier), suggesting that the risk of publication bias is low (see online supplementary material, Fig. S2). The Egger test for small study effects yielded non-significant results (P=0·254).
The present meta-analysis including >100 000 subjects revealed a significant association between adherence to dietary patterns exhibiting specific MD characteristics and decreased risk of T2DM. The overall quality of evidence rated according to the GRADE guidelines for the main outcome was moderate( Reference Guyatt, Oxman and Akl 28 ). To the best of our knowledge, the present study represents the first meta-analysis showing a possible association between adherence to an MD and risk of T2DM.
With respect to potential mechanisms of action, there seems to be a causal link between oxidative stress, inflammation, endothelial dysfunction and diabetes( Reference Marin, Yubero-Serrano and Lopez-Miranda 29 ). Interconnections between MD and markers of inflammation/endothelial function have been suggested by a recent meta-analysis of intervention trials( Reference Schwingshackl and Hoffmann 30 ). Different adherence scores measuring compliance to an MD were used in epidemiological studies since Keys’ first findings, and one of the most relevant scores was established by Trichopoulou et al.( Reference Trichopoulou, Costacou and Bamia 27 ). Several studies of the present meta-analysis used MD scores based on Trichopoulou et al. with only slight variations. The combination of components and the adherence to the whole concept of an MD seems pivotal: increasing fruit and vegetable intake as a sole option only showed weak overall effects on risk for developing T2DM( Reference Cooper, Forouhi and Ye 31 ). Making use of synergistic effects of different nutrient compounds especially in an MD has shown to be more distinct and effective. Thus, compared with targeting single nutrients on their health impact, dietary pattern analysis is especially useful in a public health context providing clear-cut messages for stakeholders.
In our meta-analysis, the relative risk for developing T2DM when adhering to an MD was significantly different when comparing European and US studies. Risk for T2DM development was significantly more decreased in European studies as compared with their US counterparts. This finding may be due to confounders that were not assessed in the various studies. Subgroup analysis including studies of great population (sample size ≥10 000) and long-term follow-up (≥10 years) resulted in a more pronounced effect of MD adherence on T2DM (RR=0·82, P=0·005, I 2=67 % and RR=0·75, P<0·00001, I 2=0 %) when compared with studies with a smaller sample size (<10 000) or short-term studies (<10 years). Especially long-term follow-up showed a prominent and significant decrease of T2DM risk of 25 %, suggesting a beneficial long-term effect of MD on T2DM risk.
There are some limitations to our study that should be considered. First, although most studies used a stringent MD score computation based on Trichopoulous et al.( Reference Trichopoulou, Costacou and Bamia 27 ), a more homogeneous MD score computation would be desirable to make studies even more comparable. Second, two studies with a slightly different study and population design (Mozaffarian et al., secondary prevention of CVD( Reference Mozaffarian, Marfisi and Levantesi 20 ); Tobias et al., volunteers with history of gestational diabetes( Reference Tobias, Hu and Chavarro 24 )) were included in the present meta-analysis. Thus, all subjects in these studies had a history of a disease, which may be associated with higher diabetes risk than for healthy subjects. However, results of the primary analysis could be confirmed following exclusion of the respective studies. High inter-study heterogeneity most likely results in a reduction in reliability of estimates. Since considerable heterogeneity could be observed for the primary analysis in the present systematic review, additional sensitivity analyses were performed to evaluate whether this could have been due to a single study. However, no single study seemed to have a considerable effect on heterogeneity. Furthermore, meta-analyses are limited by the included study types and cohort studies are at risk of various potential sources of bias. For example, response bias as well as errors in the assessment of diets is a persistent problem when analysing associations between diet and disease.
In conclusion, the current systematic review and meta-analysis shows that adherence to an MD is associated with a decreased risk of becoming diabetic at a reasonable magnitude (19 %). Overall, the present study supports the relevancy and effectiveness of a long-term adherence to an MD as an important measure in the primary prevention of T2DM.
Acknowledgements: The authors are grateful to Francesco Sofi (MD, PhD) for statistical advice. Financial support: This research received no specific grant from any funding agency in the public, commercial or not-for-profit sectors. Conflict of interest: None. Authorship: L.S. and B.M. conducted the data analysis; L.S., B.M., J.K. and G.H. interpreted the results, drafted the manuscript and finalized the manuscript. All authors read and approved the final manuscript. Ethics of human subject participation: Ethical approval was not required.