Using data from over 60,000 participants, we demonstrated that higher adherence to the MedDiet is associated with lower risk of incident all-cause dementia. Specifically, participants with the highest MedDiet adherence had 23% lower risk of developing dementia in comparison with those with the lowest level of adherence (highest vs. lowest MEDAS continuous tertiles), which was equivalent to an absolute risk difference (reduction) of 0.55%. We found no significant interaction between MedDiet adherence, defined by both the MEDAS continuous and PYRAMID scores, and polygenic risk for dementia. In addition, we found that a continuous MEDAS score was a more sensitive predictor of dementia risk when compared with a binary MEDAS or PYRAMID scores.
Previous studies exploring associations between MedDiet adherence and dementia risk have produced inconsistent findings. Indeed, a systematic review by Limongi and colleagues [
9] reported lower risk of Alzheimer’s disease and all-cause dementia in four out of seven and zero out of five studies (with the other studies reporting null findings), respectively. A more recent cohort study analysis found lower risk of all-cause and non-Alzheimer’s, but not Alzheimer’s, dementia among those with higher MedDiet adherence [
16]. Previous investigations have used different approaches for collecting dietary intake data (e.g., food frequency questionnaires and 24-h recall methods), and have employed various MedDiet scoring systems, each of which define adherence to this dietary pattern in distinctly different ways. Such heterogeneity could hinder efforts to interpret and compare results from different studies [
9]. Indeed, although we observed broadly consistent findings across the different MedDiet scores in this study, the strength of association with dementia risk differed. Whilst diet may be an important tractable risk factor for dementia, it is not emphasised in all dementia prevention guidelines (e.g., [
2]), which may reflect the lack of consistent evidence about the dietary patterns that are associated with lower dementia risk. A better understanding of the best ways to operationalize a healthy dietary pattern (including the MedDiet) will be valuable for future research studies and for the formulation of dietary guidelines to minimise dementia risk.
There is limited and inconclusive evidence about the interaction between diet (defined by MedDiet adherence or another dietary index) and genetic risk on dementia incidence [
13,
18‐
20]. For example, higher MedDiet adherence was associated with lower dementia risk in
APOE ε4 carriers but not non-carriers in one study [
13]. In contrast, other studies have reported that higher adherence to both the MIND diet (a hybrid between the MedDiet and Dietary Approach to Stop Hypertension) [
18] and a ‘healthy’ diet [
19] are more protective against dementia in
APOE ε4 non-carriers. In the present study, we found no significant interaction between polygenic risk for dementia and MedDiet adherence defined by the MEDAS continuous or PYRAMID scores in our primary analyses. Likewise, when we explored the interaction between MedDiet adherence and genetic risk defined by
APOE genotype, there was a similar pattern of response for both
APOE ε4 carriers/non-carriers. Thus, our findings suggest similar associations between MedDiet adherence and dementia risk irrespective of genetic risk for this condition. Nevertheless, we acknowledge a degree of uncertainty in this conclusion, given that findings were not consistent across all sensitivity analyses. Further research into the interaction between diet and genetics on dementia risk is therefore warranted.
This study has several strengths. The majority of previous studies exploring associations between MedDiet adherence and dementia risk have involved relatively small numbers of participants (
n = 1000–6000) with limited dementia cases (
n = 20–400) and may have lacked statistical power [
9]. In contrast, our study involved a much larger cohort (
n = ~ 60,000) with more dementia cases (
n = 882) than most previous investigations. We defined genetic risk for dementia using a comprehensive polygenic risk score whereas most previous studies have explored gene-diet interactions for individual genetic variants (e.g.,
APOE genotype) [
13,
18‐
20]. A further strength of this study is that we carried out a wide range of sensitivity analyses which demonstrate the robustness of our findings. Several limitations should also be considered. Firstly, the observational design of this study precludes drawing causal inferences. Nevertheless, our findings are supported by the results from randomised controlled trials. This includes data from the Navarra [
37] and Barcelona [
38] cohorts of the PREDIMED trial, which demonstrated clinically meaningful benefits of a MedDiet intervention on cognitive function. A further limitation is the potential risk of reverse causality, given lower MedDiet adherence could be a consequence rather than a cause of dementia [
39]. Although we did not find any evidence of reverse causality in sensitivity analyses where we excluded participants who developed dementia in the first two or five years of follow up, this does not eliminate the possibility that diet quality declined earlier in individuals who developed dementia, given the long pre-clinical phase of this condition [
40,
41]. The measurement of dietary intake is a major challenge in research, and there are specific limitations related to the assessment of dietary intake in this study which should be considered when interpreting our results. We used dietary data from the Oxford WebQ, a self-administered 24-h recall method which provides results broadly comparable to those achieved via interviewer-administered 24-h recalls [
25], to derive our MedDiet scores. Since multiple 24-h recalls are required to provide a ‘true’ representation of habitual diet [
42], and many participants in UK Biobank completed only one or two recalls, it is possible that calculated MedDiet scores are not fully representative of the participants usual dietary intake. However, 20,348 participants repeated the touchscreen questionnaire at median 4.4 years after the initial dietary assessment. Analyses of the resulting data showed that there was moderate to substantial agreement between the responses to the dietary touchscreen questions at baseline and at the repeat visit [
43]. Based on this evidence, we conclude that the estimates of dietary intake available in UK Biobank represent habitual intake and that this limits the likelihood of participant misclassification. In addition, we were unable to determine the amount of olive oil consumed from the available dietary data, which increases the risk of misclassification, as individuals who consume large amounts of olive oil may have been awarded lower scores than if we had been able to accurately quantify their intake of this MedDiet component. Nevertheless, findings from our previous research suggest that few individuals in a UK setting consume the requisite amount of olive oil to be awarded a full point for this MedDiet component [
17], suggesting that this is likely to have had a limited impact on our MedDiet scores overall. Similarly, it was difficult to quantify accurately intake of sofrito, a sauce containing typically tomatoes, onions, and garlic cooked with olive oil which is popular in Mediterranean cuisine and is one of the components of the MEDAS/MEDAS continuous scores. We used self-reported intake of tomato-based sauces as the closest proxy for sofrito intake, which may have resulted in some misclassification depending upon the ingredients and preparation method. A further limitation of our work is that dementia cases were ascertained via linkage to hospital inpatient records and death registry only, which may miss some cases [
44,
45]. However, previous studies have suggested good agreement with dementia ascertainment through primary care records [
45]. Finally, UK Biobank participants are generally healthier and of higher socioeconomic status than the wider UK population [
46] but this is unlikely to jeopardise valid assessment of exposure-disease relationships that are widely generalizable [
46]. Nevertheless, since we restricted our sample to individuals of European ancestry aged ≥ 60 years at recruitment, our findings require substantiation in other populations (e.g., different ethnicities).