Research has shown that a greater adherence to the Mediterranean diet is associated with a reduced risk of major chronic disease. However, the existing literature leads to debate for different issues, such as the measurement of the adherence to the Mediterranean diet, the use of a wide variety of dietary indices with various food components and the large heterogeneity across the studies. In order to summarise the evidence and evaluate the validity of the association between the adherence to the Mediterranean diet and multiple health outcomes, an umbrella review of the evidence across meta-analyses of observational studies and randomised clinical trials (RCTs) was performed. Thirteen meta-analyses of observational studies and 16 meta-analyses of RCTs investigating the association between the adherence to the Mediterranean diet and 37 different health outcomes, for a total population of over than 12 800 000 subjects, were identified. A robust evidence, supported by a P-value<0.001, a large simple size, and not a considerable heterogeneity between studies, for a greater adherence to the Mediterranean diet and a reduced the risk of overall mortality, cardiovascular diseases, coronary heart disease, myocardial infarction, overall cancer incidence, neurodegenerative diseases and diabetes was found. For most of the site-specific cancers, as well as for inflammatory and metabolic parameters, the evidence was only suggestive or weak and further studies are needed to draw firmer conclusions. No evidence, on the other hand, was reported for bladder, endometrial and ovarian cancers, as well as for LDL (low density lipoprotein)-cholesterol levels.
The Mediterranean diet is a model of eating based on the traditional foods and drinks of the countries surrounding the Mediterranean Sea. Over the last few decades, it has been promoted worldwide as one of the healthiest dietary pattern and has been reported to be consistently beneficial with respect to chronic diseases and longevity.1 There are individual foods and components within the Mediterranean Diet which are particularly beneficial to health (for example, extra virgin olive oil), but overall it is the combination of foods which is linked to improved health.2 This awareness shifted the research’s focus from the analysis of individual nutrients to the evaluation of the whole diet and the interaction between its components.
Numerous different dietary indices estimating the adherence to the Mediterranean diet have been operationalised so far, finding that an increasing adherence to this diet is associated with a healthier life. In particular, recent meta-analyses demonstrated that a 2-point increase in adherence score determines a significant reduction of overall mortality, and a reduced risk of cardiovascular disease, cancer and neurodegenerative diseases.3
However, the existing literature is somewhat confusing and leads to debate for different issues such as the measurement of the adherence to the Mediterranean diet, the use of a wide variety of dietary indices with various food components and the large heterogeneity across the studies. In particular, 22 indexes quantifying the compliance to the Mediterranean diet have been described.4 Several different methods such as the comparison between the highest and the lowest tertile, quintile or adherence category of the Mediterranean score used in each study, or continuous variables (1-point or 2-point increase in adherence score) have been used to pool data obtained with different indexes in meta-analyses.
Umbrella reviews or overviews of existing systematic reviews and/or meta-analyses are relatively new study designs that help providing a comprehensive and systematic examination of the scientific literature available for a specific research topic.5 To the best of our knowledge, no attempts of reviewing the existing literature through an umbrella review in this issue has been conducted. Umbrella review offers the possibility to understand the strength of evidence and extent of potential biases in the association between the adherence to the Mediterranean diet and different health outcomes.
Hence, aim of the present study was to summarise the available evidence on the existing meta-analyses on the Mediterranean diet and different health outcomes, and to provide an overview of the validity of the studied associations, by evaluating also possible hints of biases.
According to the Joanna Briggs Institute Umbrella Review Methodology http://joannabriggs.org/assets/docs/sumari/ReviewersManual-Methodology-JBI_Umbrella%20Reviews-2014.pdf two reviewers (MD, FS) conducted a systematic literature search in Medline (1950 through February 2017), Embase (1980 through February 2017), Scopus (through February 2017), Cochrane database of systematic reviews, and Google Scholar (up to February 2017). Additional studies were searched by checking references of the identified articles. The following key words, used in combination as MeSH terms and text words were used: ‘diet’, ‘Mediterranean’ and their variants, which were used in combination with words relating to health status ‘plasma lipids’, ‘cholesterol’, ‘triglycerides’, ‘glycaemia’, ‘hematic parameters’, ‘cancer’, ‘circulatory diseases’, ‘cardiovascular disease’, ‘ischaemic heart disease’, ‘cerebrovascular disease’, ‘mortality’, ‘health effects’, ‘health status’ and their variants. The most updated or complete publication was used when more than an article was present for a single study. In addition, separate meta-analyses on multiple outcomes presented in a single article were assessed separately. Missing data or additional information were requested from the corresponding authors of the articles.
Studies were included if they met the following criteria, established by using the PICOS strategy (Supplementary Table 1):
Study design: systematic reviews including meta-analyses (quantitative analysis) of observational studies (prospective cohort studies, cross-sectional studies and case-control studies) or meta-analyses of randomised clinical trials (RCTs).
Study population: >18 years.
Outcomes: health outcomes (for example, overall mortality, cardiovascular diseases, cancer, cognitive disorders, metabolic disorders), modifications of metabolic risk parameters (for example, anthropometric measurements, blood pressure, flow-mediated dilation, lipid profile and glycaemic profile) or modifications of inflammatory parameters (for example, pro-inflammatory cytokines).
We excluded meta-analyses that did not present study specific data (effect size and 95% confidence intervals (CI)). The decision to include studies was made on the basis of the study title, study abstract and full-text screening.
Literature search and data extraction were conducted independently by two authors (MD, FS), with disagreements resolved by consensus with a third reviewer (GP). As the mixture of studies with different study design may increase heterogeneity, wherever possible we considered summary results separately in meta-analyses of prospective cohort studies, cross-sectional studies, case-control studies and RCTs. The following data were extracted from each eligible meta-analysis and organised using a standard form: first author, year of publication, outcome examined, number of included studies, study design, type of comparison, number of events and population (in meta-analyses of cross-sectional and prospective cohort studies), number of cases and controls (in meta-analyses of case-control studies), number of subjects assigned to the intervention and the control groups (in meta-analyses of RCTs), maximally adjusted effect size measurements (that is, relative risk/hazard ratio, odds ratio, mean difference) along with the corresponding 95% CI and quality of the included studies in each meta-analysis (when a qualitative assessment was performed). When the data were provided in mg/dL, they were transformed into mmol/l for consistency of results. Outcomes were categorised into six categories: overall mortality, cardiovascular outcomes, cancer outcomes, cognitive disorders, metabolic disorders and inflammatory parameters.
The summary effect size and its CIs by 95% were estimated using both fixed effects and random effects models for each meta-analysis, by using Review Manager (RevMan, version 5.3 for Macintosh; The Cochrane Collaboration, Copenhagen, Denmark). For the summary random effects, we estimated the 95% prediction interval (PI), which further accounts the degree of between-study heterogeneity and gives a range for which we are 95% confident that the effect in a new study examining the same association lies within.6 Statistical heterogeneity between studies was evaluated using the I2 statistic.7 Where I2 exceeded 50% or 75%, the heterogeneity was considered substantial or considerable, respectively.
In order to detect any evidence for small study effects, we performed the Egger’s regression asymmetry test8 and we calculated the standard error of the effect size (under random effects) for the largest study of each meta-analysis. The largest study was defined on the basis of the smallest standard error. If the P-value for Egger’s test was <0.10 and the largest study had smaller effect size compared to the summary effect size (more conservative) both criteria for existence of small study effects were fulfilled.9
Finally, according to previous umbrella reviews,10, 11 we categorised the observed associations as convincing or not, by using the following criteria: significance at P0.05 and P0.001, which is considered to be a more appropriate threshold of statistical significance to reduce the number of false-positive findings;12 inclusion of over than 500 or 1000 cases for binary outcomes (more than 2500 or 5000 total participants if the metric was continuous); absence of considerable heterogeneity (I2<75%); 95% PI excluding the null value and absence of small study effects. Specifically, we identified five categories:
Convincing evidence: significance threshold reached at P0.001 for both random- and fixed effects calculation; >1000 cases (or >5,000 total participants if the metric was continuous); not large heterogeneity between studies (I 2<50%); 95% PI excluding the null value; no evidence of small study effects (if it could be tested).
Highly suggestive evidence: significance threshold reached at P0.001 for both random- and fixed effects calculation; >1000 cases (or >5000 total participants if the metric was continuous); not considerable heterogeneity between studies (I 2=50–75%).
Suggestive evidence: significance threshold reached at P0.001 for random-effect calculation; 500–1000 cases (or 2500–5000 total participants if the metric was continuous).
Weak evidence: significance threshold reached at P0.05 for random effects calculation.
No evidence: significance threshold not reached (P>0.05).
In addition, two reviewers (MD, FS) independently evaluated methodologic quality of the included meta-analyses using the modified version of the Assessment of Multiple Systematic Reviews (AMSTAR) questionnaire, developed to specifically address quality of meta-analyses on the Mediterranean diet—the AMSTARMedSD.13 The AMSTARMedSD contains a total of 14 questions, with a maximum score of 21. The scale’s individual items describing individual methodologic aspects related to quality and rater’s agreement, are described by Huedo-Medina.13
The concordance between the direction, the magnitude (overlapping CIs) and the statistical significance was examined when multiple meta-analyses of observational studies were present for the same outcome, while the concordance regarding the direction and the statistical significance was examined for multiple meta-analysis of RCTs. Statistical analysis was performed using the statistical package PASW 20.0 for Macintosh (SPSS Inc., Chicago, IL, US).
Overall, 1231 articles were identified, out of which 1135 were excluded after a screening of titles and abstracts (including duplicates). Of the 96 remaining articles, 71 were excluded after full-text screening for different reasons (Figure 1). The 25 remnant articles reported data from 140 different meta-analyses.
Meta-analyses of observational studies
Thirteen meta-analyses of observational studies3, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25 reported the possible association between the adherence to the Mediterranean diet and 35 different outcomes, for a total of 12 625 301 subjects. Study characteristics are summarised in Table 1. Overall, cohort prospective studies were analysed in 12 different meta-analyses, case-control studies in 4, and cross-sectional studies in 5. Eleven meta-analyses15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25 compared high versus low adherence to the Mediterranean diet, 2 meta-analyses3, 14 evaluated health effects associated with a 2-point increase in the adherence score to the Mediterranean diet, whereas 1 meta-analysis21 evaluated health effects associated with a 1-point increase in the adherence score.
By grouping the outcomes into the proposed categories, the included meta-analyses of observational studies resulted as follows: overall mortality (n=1), cardiovascular outcomes (n=12), cancer outcomes (n=26), cognitive disorders (n=14), and metabolic disorders (n=15). The median number of included studies in each meta-analysis was 4 (range: 1–16), while the median number of cases was 1,752 (range: 111–82,198), and the median number of population/controls was 15 595 (range: 296–2 720 221). For some specific outcomes (gastric cancer, liver cancer and oesophageal cancer), the selected meta-analyses reported only 1 cohort study and only 1 case-control study.
Over than one meta-analysis was available for 21 different outcomes. As regards the concordance between the magnitude, statistical significance of the effect and direction, agreement was present for CVD incidence/mortality, overall cancer, colorectal cancer, neurodegenerative diseases and diabetes. Meta-analyses examining stroke, head/neck cancer, esophageal cancer, gastric cancer, breast cancer, liver cancer, endometrial cancer, cognitive impairment, depression and metabolic syndrome reported disagreement in terms of the statistical significance of the effect and/or in the magnitude, but not in the direction of the effect, being all positively related to the protection for a greater adherence to the Mediterranean diet. Finally, opposite results were found for dementia and prostate cancer, and non-comparable results were reported for waist circumference, triglycerides, HDL-cholesterol and glucose.
The forest plot with the summary effect for each health outcome evaluated in meta-analyses of observational studies is depicted in Figure 2. Whether for an outcome overlapping meta-analyses with the same study design existed, we retained the meta-analysis with the largest number of studies. Furthermore, whether the overlapping meta-analyses had the same number of studies, we reported the most recent.
Meta-analyses of randomised controlled studies
Sixteen different meta-analyses of randomised controlled studies (RCTs)14, 15, 17, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37 analysed the effects of a dietary intervention with the Mediterranean diet on 26 outcomes (Table 2), grouped as follows: overall mortality (n=1), cardiovascular outcomes (n=26), metabolic disorders (n=41), and inflammatory parameters (n=3). The total number of subjects in the included RCTs was 202 148. The median number of the included studies in each meta-analysis was 6 (range: 1–39), and the median number of participants was 1571 (101–25 370). Two meta-analyses of only 1 RCT included heart failure and diabetes, respectively, as clinical outcome.
Over than 1 meta-analysis was present for 15 different outcomes. For body weight, waist circumference, body mass index, total cholesterol, glucose and C-reactive protein agreement existed across the meta-analyses for the statistical significance of the effect and the direction. On the other hand, meta-analyses examining CVD mortality, stroke, systolic blood pressure, diastolic blood pressure, triglycerides, HDL-cholesterol, insulin, and HbAc1 reported disagreements in terms of the significance of the effect, but not for the direction. As regarding LDL-cholesterol levels, no evidences were obtained in the 3 different meta-analyses.
The forest plot with the summary effects for each health outcome explored in RCTs is depicted in Figure 3.
Evaluation of bias, heterogeneity and quality
The evaluation of the level of significance for both random- and fixed-effect calculations, the sample size, the heterogeneity, the 95% PI, and the presence of small study effects is reported in Supplementary Tables 2 and 3, for meta-analyses of observational studies and RCTs respectively. With regard to the bias assessment performed in order to detect evidences for small study effects, results are presented in Supplementary Tables 4 and 5.
On the basis of the AMSTARMedSD assessment, meta-analyses achieved a medium-to-high quality score (mean±s.d.: 16.36±2.36). As reported in Supplementary Table 6 meta-analyses totalled >80% of the highest achievable score (>16 points out of 21), and the remaining 14 meta-analyses totalled between 50% and 80% of the highest score (between 11 and 16 points). All the meta-analyses (100%) provided an ‘a priori’ design and performed a comprehensive literature search (22% at least searched electronic databases and 88% supplemented those with other sources such as reference list from other articles), and most of meta-analyses (84%) reported that there was duplicate study selection. Almost all the meta-analyses (96%) made it possible to replicate the literature search, but only 40% permitted the inclusion of grey literature. The list of included studies was provided in 92% of meta-analyses, 96% included characteristics of included studies, and 72% assessed and documented scientific quality of the included studies. In addition, the Newcastle Ottawa Scale (NOS) was used in 9 meta-analyses of observational studies (Supplementary Table 7). Five meta-analyses3, 17, 24, 27, 30 performed a quality assessment using criteria set by the authors, 6(refs. 19, 26, 28, 32, 36, 37) assessed the potential existence of bias using the Cochrane risk of bias tool, and 134 used the Jadad score. Despite this, only 10% of meta-analyses reported how results might depend on study quality. All the meta-analyses used appropriate methods to combine study findings, whereas the likelihood of publication bias was assessed in 80% of meta-analyses.
Strength of evidence
Figure 4 shows the strength of the evidence estimated on the basis of the following criteria: level of significance for both random- and fixed-effect calculations, sample size, heterogeneity, 95% PI, and the presence of small study effects.
A positive effect of a greater adherence to the Mediterranean diet was found to be present in the convincing/highly suggestive categories for 12 different health outcomes (overall mortality, cardiovascular disease, coronary heart disease, myocardial infarction, overall cancer incidence/mortality, overall cancer incidence, breast cancer, neurodegenerative disease, cognitive impairment, Alzheimer’s diseases, dementia and diabetes). Among these, overall mortality, cardiovascular disease, coronary heart disease, myocardial infarction and diabetes were examined by both meta-analyses of observational studies and RCTs, the latter showing no evidence (except for diabetes). For all the other outcomes, including overall cancer mortality, colorectal, gastric, pancreatic, liver and respiratory cancers, depression, as well as anthropometrical, metabolic and inflammatory risk parameters, the grade of evidence resulted suggestive or weak. As regarding stroke, systolic, diastolic blood pressure, and metabolic parameters, we observed mixed results, depending on the design of the included studies.
Finally, no association between the adherence to the Mediterranean diet and bladder, endometrial, and ovarian cancers in meta-analyses of observational studies was found. Similarly, there was no evidence for heart failure, major cardiovascular events and LDL-cholesterol in meta-analyses of RCTs.
The present is the first umbrella review estimating the association between the adherence to the Mediterranean diet and 37 different health outcomes including overall mortality, cardiovascular outcomes, cancer outcomes, cognitive disorders, metabolic disorders, as well as inflammatory parameters. The overall analysis comprised 13 meta-analyses of observational studies and 16 meta-analyses of RCTs, for a total population of over than 12 800 000 subjects. Most summary estimates supported the notion that a greater adherence to the Mediterranean diet determines a reduction of the risk of chronic diseases and overall mortality. Furthermore, meta-analyses of RCTs demonstrated that subjects allocated to a Mediterranean diet had, as compared with subjects following a control diet, better anthropometrical, metabolic and inflammatory risk parameters.
The largest proportion of the included meta-analyses examined metabolic disorders and cardiovascular outcomes. We observed a robust evidence, supported by a P-value 0.001, large simple size, and not a considerable heterogeneity between studies for cardiovascular disease, coronary heart disease, and myocardial infarction, evaluated through meta-analyses of both observational studies and RCTs. The beneficial effects of the Mediterranean diet against these outcomes has usually been attributed to its influence on traditional atherosclerotic risk factors.38, 39, 40 Accordingly, our analysis found suggestive evidence supporting the greater effectiveness of the Mediterranean diet in reducing weight, BMI and waist circumference, lowering total cholesterol levels, and increasing HDL-cholesterol levels, when compared to control diets. Conversely, no association was reported for LDL-cholesterol levels. Among metabolic disorders, our paper indicates that the association between the Mediterranean diet and reduced risk of diabetes was the most robust, whereas evidence for a protective effect against the metabolic syndrome was weaker. In addition, meta-analyses of RCTs provided suggestive evidence for a better glycaemic control, associated with reduced insulin resistance, in subjects following a Mediterranean diet in comparison to a control diet.
Actually, some authors have suggested that anti-inflammatory effects in the vascular wall may be another important mechanism that helps explaining the link between the Mediterranean diet and cardiovascular disease.41 In addition, recent findings from the PREDIMED Study suggested that the reduction in serum nitric oxide and endothelin-1 as well as endothelin-1 receptors gene expression explain, at least partially, the effect of a Mediterranean diet high in olive oil or nuts on lowering blood pressure.42 Few meta-analyses evaluated inflammatory parameters,27, 34 but all confirm an inverse relationship between a higher score of diet, CRP and IL-6, even if the grade of evidence is weak, partly due to a low statistical power. As regarding stroke, systolic and diastolic blood pressure, all meta-analyses of cross-sectional studies presented no evidences, in contrast to meta-analyses of prospective cohort studies. This finding might be explained by intrinsic limitations of cross-sectional studies, that failed to assess causal relationship between measures.43
As regarding cancer outcomes, selected meta-analyses suggested that adopting the Mediterranean diet may help prevent cancer, although the current evidence is highly suggestive only for overall cancer, and in particular for overall cancer incidence. A possible explanation of these results could be that incidence and mortality are two different outcomes, with cancer mortality being mainly affected by the treatment approaches. The analysis of specific localisations of cancer provided a small number of studies, showing a suggestive or weak association for colorectal, liver, and pancreatic cancer. Conversely, meta-analyses evaluating bladder, endometrial and ovarian cancer, provided null results. As far as the other cancer sites such as breast, gastric, prostate, esophageal, respiratory and head/neck cancer are considered, results resulted to be controversial. The low number of studies that analysed this issue, and the low sample size, as well as the limited cases ( 500) identified in some meta-analyses, may have limited statistical power to detect an association. Moreover, the few existing studies have different study design, so the comparison between studies is difficult. For example, as regarding breast cancer, the meta-analysis of case-control studies,18 including approximately 15 000 subjects, showed a reduction in risk, whereas the 2 meta-analyses of cohort studies,18, 19 including almost 1 500 000 subjects, reported no evidences. The possible interpretation of this discrepancy could rely on the inherent limitations of self-reported dietary data and report bias. Indeed, data originating from case-control studies result less robust because of the likelihood of dietary report bias in subjects asked to report their dietary habits retrospectively after a diagnosis of disease. Conversely, cohort studies generally assess dietary patterns before the onset of diseases, being more likely to provide insight concerning the relationship between dietary exposures and disease outcomes.44 It is nevertheless important to know the role of dietary factors within the disease process, in order to determine the relevant time-period of dietary assessment in cohort studies of chronic diseases.45
The adherence to the Mediterranean diet was also correlated to cognitive functions. To date, there are meta-analyses3, 20, 22 providing convincing evidence in favour of a positive relationship with neurodegenerative diseases, in particular with Alzheimer’s disease and dementia, although, again, the meta-analysis of cross-sectional studies provided no evidence.16 As previously discussed,46 standardisation of tools used to assess cognitive function, is needed for evaluating the effectiveness of the Mediterranean diet for the prevention of cognitive impairment.
The methodologic quality of the meta-analyses and systematic reviews on the Mediterranean diet has been previously evaluated by Huedo-Medina and colleagues.13 They applied the AMSTARMedSD quality scale on 24 studies examining the association between the Mediterranean diet and cardiovascular disease outcomes, by identifying the studies that did not fully comply with contemporary methodologic quality standards. Our application of the AMSTARMedSD tool obtained better results since all the investigated meta-analyses achieved a medium-to-high quality score, so suggesting that current meta-analyses evaluating the effects of the Mediterranean diet on health status partially or almost fully comply with methodologic quality standards. This difference could be explained by the fact that we included only meta-analyses, with respect to the reviews included by Huedo-Medina et al., that tended to achieve higher scores.13
The present umbrella review has several limitations that should be considered. First of all, very few studies exist for several outcomes like specific localisations of cancer or inflammatory parameters, making difficult to reach a definitive conclusion. Further and better designed studies are needed to confirm the present findings. Second, several primary studies achieved a low quality score when authors performed a quality assessment. Since the quality of meta-analyses is related to the quality of included studies, the results regarding the poorly covered outcomes need to be interpreted with caution. Third, a large part of evidence from RCTs is weak or suggestive, pointing out the need of further and better designed trials. Forth, the interpretation of tests for statistical bias offers suggestion of bias, but not definitive proofs.
In conclusion, this umbrella review provides a comprehensive resume of the published meta-analyses in relation to the Mediterranean diet and health outcomes, and maps the status of evidence. To date, the scientific literature has identified robust evidence for overall mortality, cardiovascular diseases, overall cancer incidence, neurodegenerative diseases and diabetes. The relationship between the Mediterranean diet and other outcomes could be genuine, but there is still limited evidence for them. The large heterogeneity of dietary assessment methods and inadequacies relating to the study design, necessitate recommendations for future interventions to be sufficiently powered to detect clinical outcomes. Further studies adopting more uniform methodology and analyses, and detailed reporting of population, intervention, comparison and outcome data, would allow the quantification of the association of the Mediterranean diet adherence with health outcomes and quality of life.
FS had the original idea for the study and designed the study. MD, GP and FS did the analysis. MD and FS wrote the first draft of the manuscript. All authors contributed to writing and reviewing the manuscript. AC and FS had final responsibility for the decision to submit the paper for publication.
About this article
Supplementary Information accompanies this paper on European Journal of Clinical Nutrition website (http://www.nature.com/ejcn)