As expected, greater MeDi adherence was characterized by higher intake of vegetables, fruits, legumes, cereals and fish and lower intake of meat and dairy products (Table 1).

Greater MeDi adherence was associated with male sex and being married but not education, income, or physical activity (Table 2). Individuals in the middle or high MeDi categories had a lower mean BMI and higher mean energy intake than those in the lowest category. Mean baseline MMSE, FCSRT and BVRT scores did not differ across MeDi categories. However, mean annual decline on the MMSE and FCSRT significantly decreased with increasing MeDi adherence with a difference of 0.7 point of MMSE score and 1.1 point of FCSRT score between the lowest and highest MeDi categories, over 5 years. Incidence of dementia was not significantly associated with the MeDi score.

The associations between cognitive decline and the MeDi score as a continuous variable are shown in Table 3 and associations with the MeDi score as a categorical variable are shown in Table 4.. In longitudinal analyses, each additional unit of MeDi score was associated with fewer MMSE errors at follow-up (Table 3, model 2). This association was attenuated when stroke was added to the model (Model 3). Conversely, MeDi adherence analyzed as a categorical variable was not associated with MMSE performance over time (Table 4), and MeDi as a continuous or a categorical variable was not associated with FCSRT, IST or BVRT performances over time (Tables 3 and ​and44).

Those who developed incident dementia (N=99) or AD (N=66) over 5 years were older (79.4y and 79.1y respectively vs 75.7y, P<.0001) at wave 1 than those who remained free from dementia over time. MeDi adherence, as a continuous or categorical variable, was not associated with the risk for incident dementia or AD in multivariate analyses (Table 5).

In sensitivity analyses, we assessed the association between cognitive decline and MeDi adherence as a continuous variable (Table 6) or as a categorical variable (Table 7), excluding individuals with incident dementia (n=99). In longitudinal analyses, each additional unit of MeDi score was associated with fewer MMSE errors at follow-up only when cardiovascular risk factors were taken into account but not when stroke was added to the model (Table 6). A similar pattern for the FCSRT was also observed when MeDi adherence increased. Results were similar when MeDi was considered as a categorical variable (Table 7). For example, a women who remained free from dementia over 5 years, with low MeDi adherence (score 0–3) would lose 0.9 point on the MMSE score and 2.2 points on the FCSRT score over 5 years, whereas she would have lost respectively 0.5 point and 1.1 points with a high MeDi adherence (score 6–9).

Indeed, multi-vitamins and multi-supplements use seems to be a major difference between US and French dietary behaviours. Indeed, dietary supplement use is increasingly common in the US where 52% of adults reported taking a dietary supplement in the past month and 35% reported regular use of a multi-vitamin-multimineral product (Radimer et al. Am J Epidemiol, 2004;160:339–349). Moreover, the US older persons were more likely to be multi-vitamin users (63%) (Rock, Am J Clin Nutr, 2007;85:227S–279S). In France, the multi-vitamin use seems largely lower than in the USA. Less than 10% of the participants of 3C study (Bordeaux sample) reported a regular use of multi-vitamin (Féart et al. Int J Vitam Nutr Res, in press). In the French E3N-EPIC cohort, 29% of women, aged 45 to 60y, were regular dietary supplements users (at least 3 per week) (Touvier et al. Eur J Clin Nutr, 2009;63:39–47).

Moreover, adherence to the MeDi may reflect specific health concerns and behaviours that may differ between countries regarding in particular use of supplements or other food groups that are not considered in the MeDi score computation. This may explain the inconsistent results since the MeDi score was computed according to sex-specific medians of consumption of only 9 food groups of each study sample.

Other explanations for such discrepancies are related to cognitive decline in the prodromal phase of dementia. We used 4 tests evaluating different cognitive domains that could be affected in a specific sequence during the time course preceding the clinically defined dementia syndrome.²⁵ Interestingly, the two tests related to diet in our study, i.e. the MMSE and the FCSRT, assess respectively global cognitive abilities and episodic memory, which are considered as the hallmark of pathological cognitive aging.^{25, 26} Conversely, the BVRT involves working memory, a cognitive domain particularly sensitive to normal aging. Therefore, our results suggest that the MeDi may delay decline in cognitive functions specifically involved in pathological brain aging but only at least five years before the clinical diagnosis of dementia. The MeDi would be unable to delay dementia onset or slow down cognitive decline in the five years preceding the clinical diagnosis.

Taken together, these results suggest that the MeDi may have a long-term effect. Beneficial effects would exist during the long prodromal phase of dementia²⁵ rather than in the very last years preceding dementia. There might be a window of opportunity where greater MeDi adherence could provide beneficial effects on cognitive decline. After that time, the physiopathological processes conducting to dementia could probably not be reversed by diet. During the prodromal phase of dementia, successive emergence of cognitive deficits on IST and BVRT seems to appear more than 10 years before the diagnosis of dementia²⁵. This result could partly explain the significant difference between baseline IST scores across categories of MeDi adherence. Over time, slopes of cognitive decline relative to IST appeared to be similar across categories of MeDi which may explain in part the non-significant effect.

The clinical implications of these results may seem modest but would be sizeable with longer follow-up and as expected with respect to the literature on cognitive decline in older persons. The differences of mean annual decline of cognitive performances across categories of MeDi adherence indicated a clinically significant effect on the MMSE and FCSRT which was confirmed by multivariate analyses in subjects free from dementia over time.

A solid biological foundation for the health benefits of the MeDi has already been provided. The potential mechanisms involve a decrease in oxidative stress, inflammation and vascular disease, which also participate in the pathophysiology of neurodegenerative disease.^{27, 28} Individuals with higher MeDi adherence have been shown to have higher plasma concentrations of some presumed beneficial biomarkers.^29–31 The strong evidence relating the MeDi to lower risk of vascular disease is supported in part by our results, which indicate that the association between cognitive decline and MeDi was attenuated when adjusting for stroke, but the opposite was true for cardiovascular risk factors in general. More research is needed to better understand the biological mechanisms involved in the relation between the MeDi and cognitive decline.

Our results should be interpreted with caution because of some potential methodological limitations. The MeDi score is based on a traditional Mediterranean reference pattern defined a priori which does not consider the overall correlation between foods.³³ Dietary patterns derived by a priori³⁴ or a posteriori methods¹⁵ were already associated with a significantly reduced risk for AD and fewer cognitive symptoms in the 3C cohort. The use of sex-specific cut-off points to develop the MeDi score does not really measure adherence to a universal traditional MeDi pattern but rather to a specific pattern.³⁵ The hallmark of the traditional MeDi is a very high consumption of olive oil leading to a high MUFA-to-SFA ratio.³⁶ As already reported in a non-Mediterranean population,^{11, 12} we found a relatively low MUFA-to-SFA ratio, but we verified that consumption of olive oil was positively correlated with the MeDi score and with the MUFA-to-SFA ratio (data available on request). A relatively short follow-up could also introduce a bias due to sub-clinical dementia inducing changes of dietary habits in the phase preceding dementia. In sensitivity analyses excluding incident dementia cases over 5 years, the significant association between greater MeDi adherence and slower cognitive decline tends to confirm such a potential bias. Another interpretation of these results might be the lack of protective effect of the MeDi when the neurodegenerative process of dementia is too advanced to be reversed by diet. Moreover, a selection bias cannot be dismissed. Participants with missing follow-up (N=114) were older, had a lower mean BMI, higher mean CES-D score, higher mean consumption of drugs/d and lower cognitive performance on each test than those with available follow-up. Individuals with missing follow-up also had a slightly lower mean MeDi score (4.09 vs 4.36, P=.09) than the others. Another limitation concerning cognitive tests was their potential ceiling or floor effects. Moreover, our study lacked the power to definitively be able to detect an association with incident dementia; our cohort size provided an a posteriori power of less than 10% to detect a HR of 0.9 for incident dementia as observed in our study. Finally, we cannot rule out the possibility of residual confounding by unknown risk factors such as a general healthier lifestyle of MeDi adherents. Despite these limitations, the strengths of the present study are its size, the population-based design, and control for several potential confounders. In particular, we controlled for depressive symptomatology since links between cognitive impairment and depression are well documented.^{25, 37}