Introduction

Dementia is a multifactorial disorder characterised by new onset and usually progressive deterioration of cognitive functions including memory, language and executive function [1]. Current figures estimate that around 57 million people worldwide live with dementia, and this figure is predicted to grow to 152 million in 2050 [2]. Targeting dietary factors may have great preventative potential for dementia. At present, dietary interventions are involved in the prevention of many conditions which increase the risk of dementia, including diabetes and cardiovascular diseases [1]. Additionally, considerable evidence suggests that those who have a healthier diet have a lower dementia risk [3, 4].

Literature on diet and dementia has expanded from examining single nutrients and their relation to dementia risk, to studying whole diet adherence. In the context of dementia prevention, the Mediterranean diet, a traditional dietary pattern followed by populations in Italy, Spain, Greece and other communities bordering the Mediterranean, has received the most attention. It is characterised by high consumption of fruit, vegetables, unrefined cereals and olive oil, high-to-moderate intake of fish, low-to-moderate intake of dairy products (mainly cheese and yogurt), low intake of meat and poultry and moderate intake of red wine, usually with meals [5]. Adherence to the Mediterranean diet has been associated with better global cognition in older adults [6], slower cognitive decline [7,8,9], lower risk of dementia and Alzheimer’s disease [9] and lower mortality [10]. However, methodological differences between studies such as differences in how the diet is operationalised and evaluated have resulted in inconsistent results, and other studies have not found such effects [3, 11, 12]. Examining Mediterranean diet components instead is a more flexible approach that may help in standardising research methodology and minimising heterogeneity in findings. Additionally, public health recommendations can be communicated in a more meaningful way by describing food groups to include or avoid in one’s diet [13].

The benefits of studying Mediterranean diet components are threefold:

First, breaking the diet into its core components would allow leverage for small deviations from its adherence but still capture its multifactorial aspect.

Second, it would help quantify intake of specific food groups to ensure maximum benefits, as currently there is little evidence to guide the amount and frequency of consumptions for main food components like fish, wholegrains and vegetables.

Third, the relative contribution of each food component to the diet’s beneficial effects is currently unknown.

This study aims to examine the association between Mediterranean diet components and all-cause dementia risk.

Methods

Study population

The UK Biobank is a multi-centre, population-based cohort of 502,538 participants aged 42–69 years at recruitment, who were recruited between 2006 and 2010. Participants attended one of the 22 centres across England, Scotland and Wales for baseline assessments for phenotyping, biological sample collection, self-reported questionnaires and nurse interviews. All participants provided informed consent using a signature-capture device. Participants also consented to the linkage of electronic health records from primary care, hospital attendances and death certification to their study data. UK Biobank received ethical approval from the National Information Governance board for Health and Social Care and the National Health Service North West Multicentre Research Ethics Committee in January 2006. We applied to and were given permission to use the UK Biobank data under study number 40055. We restricted our analysis to those individuals with data available across all variables of interest and to those 55 or older at baseline, as a dementia diagnosis in younger participants is likely to be relatively rare and due to factors unrelated to diet. Individuals who were diagnosed with dementia within 3 years of baseline assessments were also excluded.

Exposure

Dietary assessment

Dietary measures were ascertained at baseline via a self-reported electronic Food Frequency Questionnaire with 29 food groups covering average consumption frequency over the past year. Diet-related items included in the current study are processed meats (such as bacon, ham, sausage, meat pies, kebabs, burgers, chicken nuggets), unprocessed poultry, unprocessed beef, unprocessed pork, unprocessed lamb/mutton, oily fish, non-oily fish, dried fruit, fresh fruit, cooked vegetables, raw vegetables, cereal, bread, cheese. Consumption of meats and fish was assessed with the question: “How often do you eat …?”. For each of the meat, oily and non-oily fish consumption questions, possible answers consisted of 6 categories of weekly intake: “Never”, “Less than once a week”, “Once a week”, “2–4 times a week”, “5–6 times a week”, “Once or more daily”. To assess intake of dried and fresh fruit, participants were asked how many pieces of dried or fresh fruit they would eat a day, and intake was recorded as an integer number of pieces of daily intake. Intake of cooked and raw vegetables was ascertained by asking for average numbers of heaped tablespoons of raw/cooked vegetables consumed per day. For both food groups, participants could select “Less than one” if they ate less than one piece of fruit or heaped spoon of vegetables.

To assess cereal and bread type consumption, participants were asked about which type they mainly ate. There were 5 answer categories for cereal: “Bran cereal (e.g. All Bran, Bran flakes)”, “Biscuit cereal (e.g. Weetabix)”, “Oat cereal (e.g. Ready Brek, porridge)”, “Muesli”, “Other (e.g. Cornflakes, Frosties)”. For bread consumption, possible answers consisted of 4 categories: “White”, “Brown”, “Wholemeal or wholegrain”, “Other type of bread”.

To assess intake of cheese, participants were asked “How often do you eat cheese? (Include cheese in pizzas, quiches, cheese sauce etc.)”. Possible answers consisted of 6 categories of weekly intake: “Never”, “Less than once a week”, “Once a week”, “2–4 times a week”, “5–6 times a week”, “Once or more daily”.

For each of the above food questions, participants were instructed to provide an estimate of average intake or select “Do not know” if they were unsure. Additionally, for each question there was an answer category of “Prefer not to answer”.

Quantifying food consumption

The responses on unprocessed beef, pork, poultry, lamb/mutton and processed meat were converted into weekly-based consumption frequencies of 0, 0.5, 1, 3, 5.5 and 7 times per week, using the median for each category. Unprocessed beef, pork and lamb/mutton were summed into one group of “unprocessed red meat”, and all meat groups were summed into “total meat”. To rank participants by weekly meat consumption according to the distribution of data, intake frequencies for each meat type was categorised into five groups: processed meat (0, 0.1–0.9, once, 2.0–4.9, ≥ 5.0 times/week), unprocessed red meat (0, 0.1–1.0, 1.1–1.9, 2.0–2.9, ≥ 3.0 times/week), unprocessed poultry (0, 0.1–0.9, once, 2.0–4.9, ≥ 5.0 times/week), total meat (0, 0.1–3.0, 3.1–4.9, 5.0–6.9, ≥ 7.0 times/week).

For both oily and non-oily fish, the responses were converted into weekly based consumption frequencies of 0, 0.5, 1, 3, 5.5, 7 times per week, and consumption was summed to provide total fish intake. Intake frequency for fish was categorised into 5 groups (0–0.9, once, 1.1–1.9, 2.0–3.9, ≥ 4.0 times/week).

For consumption of dried and fresh fruit, responses were converted into daily intake—two pieces of dried fruit and one piece of fresh fruit counted as one serving [14].

The items were summed to provide total fruit consumption, and daily servings of fruit were grouped into 5 categories (0–0.9, 1.0–1.9, 2.0–2.9, 3.0–4.0, ≥ 4.5 servings a day).

For consumption of cooked and raw vegetables, the same pattern was followed as for fruit. Two heaped tablespoons of cooked vegetables or raw vegetables were counted as one serving.

Weekly intake frequency of cheese was converted into groups of 6 categories of weekly intake (0, 0.1–0.9, once, 2.0–4.0, 5.0–6.0, daily or more).

Categories for each food group were determined based on data distribution to provide similar-sized groups.

Responses from questions on bread type and cereal type were combined to create an index of grain consumption. Participants’ answers on consumption of cereal type were collapsed into two categories of refined grains (muesli, cornflakes, Frosties) and wholegrains (bran cereal, biscuit cereal (Weetabix), oat cereal). Answers on consumption of bread type were also collapsed into two categories of refined grains (white bread) and wholegrains (wholegrain/wholemeal, brown bread). Combining both cereal and bread type answers, a score was generated to reflect the type of grains most commonly consumed with those scoring higher consuming wholegrains more commonly.

Outcome

Incident all-cause dementia cases was the primary outcome, as ascertained through self-report and data linkage to hospital inpatient admissions, primary care records and death registries. Date of diagnosis was set as the earliest date of dementia codes recorded regardless of source used. Diagnoses were recorded using the International Classification of Diseases (ICD) coding system 9 and 10 [15, 16]. Across UK Biobank and all three linked databases, the positive predictive value for all-cause dementia has been found to be 82.5% [17].

Covariates

Demographic characteristics were collected at baseline through self-reported electronic questionnaires and physical measurements. Covariates in our analyses were chosen a priori from the literature on the basis of their potential to confound the relationship between diet and dementia. They were grouped into socio-demographic (age, sex, Townsend deprivation index [18], household income, age left education), lifestyle (physical activity, smoking status, alcohol intake), mental health (loneliness and depression) and physical health factors (BMI, total cholesterol, diabetes, hypertension, cardiovascular events, major dietary changes in the last 5 years) (Table 1).

Table 1 List of covariate groups and their measure methods
Full size table

Statistical analysis

All analyses were carried out using Stata SE, version 15.1, and all main analyses were pre-specified. Baseline characteristics of the sample were summarised for those with and without incident dementia as mean (SD) or median (IQR) as appropriate for continuous variables and frequency and percentage for categorical variables. Follow-up time was determined as time from baseline to the earliest of dementia diagnosis, loss to follow-up or death. We used Cox proportional hazard regression models for time to dementia diagnosis.

We first conducted univariable Cox regression for each of the six food groups separately (meat, fish, vegetables, fruit, cheese, wholegrains). We then adjusted each separate food group model sequentially for sociodemographic, lifestyle, mental health and physical health covariates. Our final model included all dietary components and was fully adjusted for all covariates.

The assumption of proportional hazards was verified using Schoenfeld residuals and checking the actual versus expected Kaplan Meier plots for each diet component. We measured proportion of missing data in all variables, and analyses for all models were limited to only those who had complete data for all of the variables.

Results

At baseline, 502,490 participants were assessed. After excluding participants younger than 55 years at baseline (n = 194,192), and those with reported dementia in the first 3 years of enrolment (n = 516) and with missing data (n = 58,271), 249,511 participants were included in the analysis (Fig. 1).

Fig. 1
figure 1

STROBE diagram showing participant selection

Full size image

Baseline characteristics of the participants (overall and by dementia status) are shown in Table 2.

Table 2 Baseline characteristics of participants overall and by dementia status
Full size table

Over 2,868,824 person-years follow-up time at risk (median 11.4, range 0.01–14 years), 4282 (1.7%) incidences of all-cause dementia occurred in the 249,511 participants, equivalent to 1.49 cases per 1000 person-years. The earliest dementia case recorded in the sample was 3 years and 3 days after baseline assessment.

Mean age of participants was 62 (SD 4) at baseline. The sample consisted of 116,569 (46.7%) men. Generally, people who developed dementia were older, lived in more deprived neighbourhoods, had lower household income, were less physically active, had higher blood pressure, were more likely to report diagnoses of diabetes and cardiovascular events and were more likely to report smoking, and feelings of loneliness and depression. More men than women were diagnosed with dementia in the study population. Those who developed dementia were also more likely to report major dietary changes in the past 5 years due to illness.

Individual food components and dementia risk

In the unadjusted model, compared to no consumption of processed meat, consumption of processed meat of more than 5 times a week was associated with increased risk of dementia (Table 3) (hazard ratio (HR) 1.24; 95%CI 1.05–1.46). However, when the model was adjusted for sociodemographic factors, the effect was no longer present. In the fully adjusted model, compared to no consumption, consumption of processed meat of less than once or once a week was found to be protective (0.1–0.9 times/week HR = 0.80; CI 0.70–0.92; once/week HR 0.83; 95%CI 0.72–0.95).

Table 3 Association between Mediterranean diet components and all-cause dementia in five adjustment models
Full size table

Compared to no consumption, moderate to high consumption of unprocessed red meat was associated with decreased risk of dementia, with the effect present throughout the subsequently adjusted models (fully adjusted model 1.1–1.9 times/week HR 0.77; 95%CI 0.66–0.90; 2.0–2.9 times/week HR 0.77; 95%CI 0.66–0.90; ≥ 3 times/week HR 0.80; 95%CI 0.70–0.97).

No associations between consumption of unprocessed poultry and dementia risk were found. In the unadjusted model, compared to no consumption, consumption of total meat of more than 7 times a week was found to increase dementia risk (HR 1.21; 95%CI 1.00–1.46); however, the association was no longer present once covariates were added. No associations between total meat consumption and dementia were found in the fully adjusted models.

Compared to consumption of less than once a week, intake of fish of between once a week and up to 2.0–3.9 times a week was associated with decreased risk of dementia—an effect present throughout the subsequent adjusted models (fully adjusted model 1.1–1.9 times/week HR 0.81, 95%CI 0.69–0.94; 2.0–3.9 times/week HR 0.80, 95%CI 0.69–0.93).

One serving a day of fruit was associated with decreased dementia risk in the fully adjusted model (HR 0.83; 95%CI 0.72–0.96) compared to consumption of less than one serving.

Consumption of one serving a day of vegetables was associated with decreased dementia risk (HR 0.84; 95%CI 0.72–0.99) compared to consumption of less than one serving.

In the fully adjusted model, consumption of cheese of once a week was found to decrease dementia risk (HR 0.81; 95%CI 0.67–0.98) compared to no consumption.

No association between wholegrain score and dementia risk was present.

Whole model (final model)

We included all food groups and all covariates in one model to examine the effects on dementia risk. The results followed the same pattern as in the single model analyses (Table 4). Specifically, compared to consumption of less than once a week, moderate consumption of fish was found to reduce the risk of dementia (2.0–3.9 times/week HR 0.84, 95%CI 0.71–0.98), and consumption of fruit between 1 and 1.9 servings a day was also protective compared to eating fruit less than once a day (HR 0.85, 95%CI 0.74–0.99). We found that those consuming high intake of processed meat (more than five servings a week) were at an increased risk of dementia compared to those not consuming processed meat; however, the associations were not statistically significant (HR 1.23, 95%CI 0.99–1.53, p = 0.059).

Table 4 Dementia risk for Mediterranean diet food groups in fully adjusted model*
Full size table

Compared to no consumption, moderate consumption of unprocessed red meat was found to reduce risk of dementia; however, the associations were not statistically significant (1.1–1.9 times/week HR 0.82, 95% CI 0.66–1.00, p = 0.058; 2.0–2.9 times/week HR = 0.82, 95% CI 0.66–1.01, p = 0.067).

Unprocessed poultry, vegetables, cheese and wholegrain consumption was not found to be associated with dementia risk.

Discussion

In this cohort of people aged 55 and over with long follow-up, we found that once all covariates and other food components were taken into account, individual components of the Mediterranean diet are not associated with dementia risk except for fish consumption and low consumption of fruit.

Initial univariable models showed an increased dementia risk from higher consumption of processed meat, reduced risk from higher consumption of unprocessed meat, reduced risk with increasing consumption of fruit and vegetables and cheese. Most of these effects were no longer statistically significant once sociodemographic factors were adjusted for, and none of these associations except for moderate unprocessed meat consumption, fish consumption and low fruit consumption were statistically significant once all covariates were adjusted for. In the fully adjusted model, once all covariates and other food groups were taken into account, only moderate fish consumption and low fruit consumption were associated with dementia risk.

Fish shows the strongest association in the single models as well as the whole model when food groups are mutually adjusted for. Our measure of fish consumption included both oily (sardines, salmon, anchovies, mackerel, herring) and non-oily fish (cod, tinned tuna, haddock), irrespective of type (canned, fresh, frozen) and cooking method (fried, boiled, roasted). Our findings support the general consensus on fish consumption and brain health benefits. Fish is one of the food groups most consistently associated with lower cognitive decline [19] and decreased risk of dementia [20]. Cohort studies conducted in France, the Netherlands, Scandinavia, Italy and the USA all show associations between regular fish consumption and decreased risk of incident dementia, Alzheimer’s Disease and cognitive decline [14], but mechanisms underlying this effect have not been elucidated.

We found low consumption of fruit to be associated with decreased dementia risk. Most cohort studies have examined the association between fruits in combination with vegetable consumption, and generally results show decreased risk of dementia [14]. Those who have examined fruit consumption on its own have generally failed to find any significant associations with dementia or cognitive decline [13, 14]. It is possible that previous studies have failed to detect any significant associations due to inadequate adjustment in their models or lack of statistical power. We could not examine potential reasons for this finding. It may be that those consuming lower levels of fruit are snacking less and eating diets that are generally healthier. Lower consumption of fruit may provide some benefits in terms of fibre intake but without additional sugar and calories from higher fruit consumption. It may also be that this is a spurious result, and replication in other cohorts would be needed before firm conclusions can be drawn.

Our results on meat consumption do not replicate previous findings from cohort studies [21]. We did not find any significant associations between processed meat consumption and increased dementia risk, although a trend was present which may relate to the stricter selection of covariates and the adjustment for other dietary components in our study.

In the context of the Mediterranean diet, it is possible for fish consumption to drive the beneficial effects. Alternatively, it is possible for the diet to influence dementia risk through secondary pathways such as cardiovascular risk factors and diabetes.

There are a number of strengths of our study. We had detailed information available on food components and robust adjustment for a wide range of covariates including sociodemographic, lifestyle, mental health and physical health factors. We included both a measure of socioeconomic deprivation and a measure of income as consumption of certain foods is likely to be associated with personal wealth. Additionally, deprivation scores relate to the area a person lives in and are not necessarily a reflection of the individual’s level of deprivation. We had a large sample size and reasonably long follow-up. Additionally, we excluded people who had dementia at baseline or within 3 years of the baseline assessment to exclude the possibility of reverse causality as much as possible. Limitations were that food consumption was only measured once in all individuals so we could not see how changes in diet may have affected dementia risk. Additionally, diet as well as many covariates were based on self-report which may be prone to bias. The UK Biobank sample is recruited from healthy volunteers so we cannot be sure the associations are generalisable.

Although we have adjusted for many covariates, we cannot rule out the possibility that the association is spurious and cannot be sure the effect is causal. This work can be further expanded to examine the influence of altered caloric intake or meal timing together with the effects of healthy diets such as the Mediterranean on brain health. This study has not collected information on calorie restriction. Emerging work in animal models has shown that prolonged reduction in daily caloric intake and periodic fasting cycles may lead to a delay of the onset and progression of disease [22]. In humans, intermittent fasting has been shown to improve cognitive dysfunction and inhibit hippocampal neuronal damage against oxidative stress [23]. Thus, future cohort studies should consider the integration of a balanced nutritious diet with periods of fasting or controlled calorie intake to further examine its effects on brain health.

Overall, our study provides evidence that consumption of fish is associated with lower risk of dementia but that other components of the Mediterranean diet are not associated with dementia risk when other covariates and other dietary components are taken into account. We cannot draw conclusions about the reasons behind this association, and this deserves further study.