We focused on prospective studies investigating the association between adherence to a Mediterranean diet and health outcomes. We searched PubMed, Embase, Web of Science, and the Cochrane Central Register of Controlled Trials databases up to 30 June 2008, using a search strategy that included both truncated free text and exploded MeSH terms. MeSH headings included “Mediterranean”, “diet”, “dietary pattern”, “disease”, “health”, “cardiovascular disease”, “cerebrovascular disease”, “coronary heart disease”, “degenerative diseases”, “cancer”, “neoplasm”, “prospective”, “follow-up”, or “cohort”, and their variants. The search strategy had no language restrictions. We also consulted references from the extracted articles and reviews to complete the data bank. When multiple articles for a single study were present, we used the latest publication and supplemented it, if necessary, with data from the most complete or updated publication. We assessed the relevance of studies by using a hierarchical approach based on title, abstract, and the full manuscript.
We identified studies that prospectively evaluated the association of an a priori score used for assessing adherence to a Mediterranean diet and adverse clinical outcomes. We excluded the studies if they had a cross sectional or case-control design, if they analysed adherence to a non-specific dietary pattern or to a recommended dietary guideline and not to a Mediterranean diet, if they evaluated a cohort of patients with a previous clinical event (that is, secondary prevention), if they did not adjust for potential confounders, and if they did not report an adequate statistical analysis.
Figure 1 shows the process of study selection. Our initial search yielded 62 reports, of which we excluded 20 on the basis of the title or abstract. Of the remaining 42 articles, we excluded 26 for the following reasons: a non-specific dietary pattern, instead of a Mediterranean diet, was evaluated (n=3); cross sectional or case-control design was used (n=18); and the study population was in secondary prevention (n=5). We excluded four additional articles because they represented duplicate studies, so we included only the latest or the more complete paper in the final analysis. Finally, 12 articles fulfilled our inclusion criteria.w1-w12
Fig 1Process of study selection
We extracted the following baseline characteristics from the original reports by using a standardised data extraction form and included them in the meta-analysis: lead author, year of publication, cohort name, country of origin of the cohort, sample size of the cohort and number of outcomes, duration of follow-up, age at entry, sex, outcome, components of the score for adherence to a Mediterranean diet, and variables that entered into the multivariable model as potential confounders (table 1). Two investigators (FS and FC) collected the data, and disagreements were solved by consensus and by the opinion of a third author (AC), if necessary. Outcomes of interest were overall mortality, mortality from cardiovascular diseases, incidence of or mortality from cancer, as well as occurrence of Parkinson’s disease and Alzheimer’s disease.
We assessed the quality of the studies according to the number of participants, the duration of follow-up, and adjustment for potential confounders. We considered studies with a high number of participants; long duration of follow-up; and adjustment for confounders including demographic, anthropometric, and traditional risk factors to be of high quality.
Definition of adherence to Mediterranean diet
Adherence to a Mediterranean diet was defined through scores that estimated the conformity of the dietary pattern of the studied population with the traditional Mediterranean dietary pattern. Values of zero or one were assigned to each dietary component by using as cut offs the overall sex specific medians among the study participants. Specifically, people whose consumption of components considered to be part of a Mediterranean diet (vegetables, fruits, legumes, cereals, fish, and a moderate intake of red wine during meals) was above the median consumption of the population were assigned a value of one, whereas a value of zero was given to those with consumptions below the median. By contrast, people whose consumption of components presumed not to form part of a Mediterranean diet (red and processed meats, dairy products) was above the median consumption of the population had a value of zero assigned, and the others had a value of one. However, some differences among the studies existed, especially in relation to the food category of vegetables (grouped with potatoes in one studyw5), meat and meat products (grouped with poultry in some studiesw4 w6), and nuts and seeds (grouped with fruits in some studies,w4 w6 w7 w12 with legumes in one study,w5 and considered a group by themselves in some othersw8 w10 w11), as well as milk and dairy products (not present in some studiesw8 w10 w11) and fish (present only in more recent studiesw4-w12). Thus, the total adherence scores (estimated as the sum of the above indicated scores of zero and one) varied from a minimum of 0 points indicating low adherence to a maximum of 7-9 points reflecting high adherence to a Mediterranean diet.
We used RevMan, version 4.2 for Windows by the Cochrane Collaboration to analyse data. We used the results of the original studies from multivariable models with the most complete adjustment for potential confounders; table 1 shows the confounding variables included in this analysis. We used a random effects model that accounts for interstudy variation and provides a more conservative effect than a fixed model. We calculated random summary relative risks with 95% confidence intervals by using an inverse variance method.
We grouped the studies according to the different clinical outcomes (mortality from all causes, mortality from cardiovascular diseases, incidence of or mortality from cancer, and incidence of Parkinson’s disease and Alzheimer’s disease). We assessed the potential sources of heterogeneity by using the standard χ2 test. In addition, we used the I2 statistic to investigate heterogeneity by examining the extent of inconsistency across the study results. To examine the potential source of heterogeneity across studies evaluating overall mortality, we did sensitivity analyses according to some characteristics of the studies—sex (male, female), country of origin (European countries, United States, other countries), follow-up time (below or above the median follow-up time of the studies: 8 years), and the quality of the studies (low, high). To assess the presence of publication bias, we computed the “failsafe N” for each of the main outcomes; this value is an estimate of the number of studies with null results that would need to be added to the meta-analysis to reduce the overall observed significant result to non-significance.