Meat consumption and risk of incident dementia: cohort study of 493,888 UK Biobank participants

ABSTRACT Background Worldwide, the prevalence of dementia is increasing and diet as a modifiable factor could play a role. Meat consumption has been cross-sectionally associated with dementia risk, but specific amounts and types related to risk of incident dementia remain poorly understood. Objective We aimed to investigate associations between meat consumption and risk of incident dementia in the UK Biobank cohort. Methods Meat consumption was estimated using a short dietary questionnaire at recruitment and repeated 24-h dietary assessments. Incident all-cause dementia comprising Alzheimer disease (AD) and vascular dementia (VD) was identified by electronic linkages to hospital and mortality records. HRs for each meat type in relation to each dementia outcome were estimated in Cox proportional hazard models. Interactions between meat consumption and the apolipoprotein E (APOE) ε4 allele were additionally explored. Results Among 493,888 participants included, 2896 incident cases of all-cause dementia, 1006 cases of AD, and 490 cases of VD were identified, with mean ± SD follow-up of 8 ± 1.1 y. Each additional 25 g/day intake of processed meat was associated with increased risks of incident all-cause dementia (HR: 1.44; 95% CI: 1.24, 1.67; P-trend < 0.001) and AD (HR: 1.52; 95% CI: 1.18, 1.96; P-trend = 0.001). In contrast, a 50-g/d increment in unprocessed red meat intake was associated with reduced risks of all-cause dementia (HR: 0.81; 95% CI: 0.69, 0.95; P-trend = 0.011) and AD (HR: 0.70; 95% CI: 0.53, 0.92; P-trend = 0.009). The linear trend was not significant for unprocessed poultry and total meat. Regarding incident VD, there were no statistically significant linear trends identified, although for processed meat, higher consumption categories were associated with increased risks. The APOE ε4 allele increased dementia risk by 3 to 6 times but did not modify the associations with diet significantly. Conclusion These findings highlight processed-meat consumption as a potential risk factor for incident dementia, independent of the APOE ε4 allele.


Introduction
Dementia is a major public health concern with around 50 million cases globally and an incidence of nearly 10 million new cases per annum (1,2). It comprises Alzheimer disease (AD), which contributes to 50-70% of dementia cases, vascular dementia (VD), which contributes to ∼25%, and other forms of dementia (2,3). Dementia development and progression are associated with both genetic and environmental factors, including diet and lifestyle (4,5). Lifestyle-related and dietary factors associated with dementia are potentially modifiable and thus represent targets for primary prevention (6).
Meat consumption has gained increasing interest in relation to health, since high consumption of processed meat and probably red meat were found to be consistently associated with an increased risk of colorectal cancer (7). In recent decades meat consumption has doubled or even tripled globally, especially in developing countries (8). This dietary transition has been associated with increasing AD prevalence in Japan, Peru, Cuba and other low-and middle-income countries in both ecological and cross-sectional studies (9,10). A study of cognitively healthy individuals in Sweden showed that low consumption of meat and meat products was associated with better cognitive performance in clinical dementia screening tests and greater total brain volume after a 5-y follow-up period (11). Our previous review on meat consumption and cognitive disorders including dementia showed that most meat-related studies were embedded in complex dietary patterns with considerable heterogeneity, and the evidence of associations between risk of dementia and specific types or amounts of meat consumption was limited (12).
A consistent association has been established between carriage of the apolipoprotein E (APOE) ε4 allele and elevated risk of dementia or AD (13). Previous stratified analyses by APOE ε4 status showed that unfavourable lifestyle factors (e.g., less healthy dietary pattern, less physical activity, smoking, and social isolation) were associated with higher risk of dementia in APOE ε4 noncarriers but not in carriers (14). The discrepancy between carriers and noncarriers indicates that APOE genotype may modify associations between lifestyle factors and dementia risks, and might be explained by a potential masking of weak associations from lifestyle factors by the strongly associated APOE ε4 allele. However, at present whether APOE ε4 allele carriage interacts with lifestyle factors, such as diet, influencing risk of dementia remains unclear.
In the present study we examined the hypothesis that high consumption of meat increases the incidence of dementia in the general population, which may be more pronounced among APOE ε4 noncarriers.

Study design
The UK Biobank is a largescale population-based cohort study of half a million participants aged 40-69 y recruited from across the United Kingdom between 2006 and 2010 (15). The Biobank recruited participants using National Health Service patient registers and conducted the baseline assessments across 22 assessment centers in England, Scotland, and Wales which included a touchscreen questionnaire, verbal interview, physical measures, and biosample collection. At recruitment, participants electronically signed consent forms and completed various touchscreen questionnaires and measurements. All available resources are listed on the UK Biobank website (http://www. ukbiobank.ac.uk/resources/). Ethical approval was granted for the UK Biobank by the North West-Haydock Research Ethics Committee (REC reference: 16/NW/0274). The UK Biobank dataset for this project included 502,493 participants.
As an enhancement to the baseline touchscreen brief FFQ, the Oxford WebQ dietary questionnaire (17), which assesses a more detailed dietary intake over the previous 24 h was added to the assessment centers from April 2009 to September 2010. After that the WebQ questionnaire was administered online once every 3-4 mo and repeated for a total of 4 rounds over a 16-mo period from February 2011 to June 2012 for 24-h dietary assessments. The Oxford WebQ asked participants to select the number of portions for each item they consumed over the previous 24-h period with instructions specifying 1 standard portion size such as 1 sausage, 1 rasher of bacon, or 1 serving of beef. The daily intakes in grams were calculated by multiplying reported numbers of portions by standard portion sizes (16). Similar foods were then combined together into distinct meat types to match the baseline touchscreen questionnaire. A subgroup of participants (n = 126,844) who completed at least two 24-h dietary assessments were included in this study (18) (see comparisons between participants without or with 1+, 2+, and 3+ completions of the Oxford WebQ in Supplemental Table 1); values from multiple assessments were averaged for each participant with 2+ completions. We then calculated the mean intakes from the 24-h dietary assessments within each category of meat types from the touchscreen brief FFQ. The corresponding mean daily intakes in each category were used in combination with frequency from the touchscreen questionnaire as continuous variables to examine the effect sizes per specific increment of meat intakes (25 g/d for processed meat and unprocessed poultry; 50 g/d for unprocessed red meat and total meat). These increments correspond to usual average portion sizes for regular eaters of these products, especially in men in the UK Biobank (19), and are consistent with other study presentations of results (20). The mean daily intakes in each meat category were also used to test the P-trend across 5 categories of each meat type, as well as to correct for the potential regression dilution bias in the touchscreen brief FFQ reported in previous studies (16,21) (more details seen in Supplemental Methods 1).

Ascertainment of dementia
Prevalent and incident dementia cases within the UK Biobank were ascertained through data linkage to hospital inpatient admissions and death registries. Self-reported dementia cases at recruitment were additionally classified as prevalent cases. The electronic linkage to hospital inpatient data and death registry records includes primary or secondary events across healthcare systems in England, Scotland, and Wales. Date of diagnosis was set as the earliest date of dementia codes recorded regardless of source used. According to the International Classification of Diseases (ICD), AD was defined as code 331.0 in edition 9 and codes F00 and G30 in edition 10; VD was defined as codes 290.4 in edition 9 and codes F01 and I67.3 in edition 10; allcause dementia was defined as all of the above codes plus ICD-9 codes 290, 291.

APOE genotyping
Genotypes of nearly one-half million participants in the UK Biobank were assayed using 2 very similar genotyping arrays manufactured by Affymetrix: the BiLEVE Axiom array for ∼50,000 participants and the UK Biobank Axiom array for the remaining ∼450,000 participants; genotyping quality control was performed by UK Biobank centrally (22). Data from UK Biobank participants with unusually high heterozygosity and missingness (>5%) and disagreement between reported sex and genetic sex were excluded in genotype-related analyses (23). In addition, we used genetic kinship to other participants (Biobank field ID 22,021) as a covariate to limit confounding from population relatedness (24). The APOE haplotypes (ε2/ε3/ε4) were directly genotyped and determined by 2 genetic variants, rs429358 and rs7412. Participants with 1 or 2 ε4 alleles were defined as APOE ε4 carriers and otherwise as APOE ε4 noncarriers. After quality control procedures, APOE genotypes were available on 405,126 UK Biobank participants and were included in APOE genotype related analyses.

Statistical analysis
Participants with prevalent dementia, and those with incomplete data on meat-related variables were excluded before analyses. Given the possibility that underlying dementia may cause changes in dietary behaviors in advance of diagnosis, we excluded incident dementia cases that occurred in the first-y period from baseline dietary data collection to dementia diagnosis to limit the possibility of reverse causality (25). A more stringent 3-y cut-off was also applied as a sensitivity analysis (see the flowchart in Supplemental Figure 1).
Baseline sociodemographic, lifestyle, and main dietary characteristics were summarized and stratified by dementia status (incident dementia and no dementia). Among incident cases, all-cause dementia, AD, and VD were treated as separate outcomes. The associations between incident dementia and reported consumption of processed meat, unprocessed poultry, unprocessed red meat, and total meat were fitted in Cox proportional hazards regressions with the duration of follow-up in years as the timescale and the second lowest category of meat intakes as the reference; HRs with 95% CIs were reported for all analyses.
Three models were applied in our analyses: unadjusted models, minimally adjusted models, and fully adjusted models. The minimally adjusted model was adjusted for age at baseline, gender, self-reported ethnicity (White, Asian, Black, mixed, other/unknown), socioeconomic status (low, moderate, or high deprivation), educational level (with university/college degree or not), determined by a directed acyclic graph (26) (Supplemental Methods 2). The fully adjusted model was additionally adjusted for region (England, Wales, Scotland), BMI (in kg/m 2 ; <25, 25-29.9, and ≥30), physical activity level (low, moderate, and high), smoking status (never, past, and current), typical sleep duration (<7, 7-8, >8 h/d), stroke history, family history of dementia, and dietary factors including total consumption of vegetables and fruits, total fish, tea and coffee, and alcohol. Processed meat, unprocessed poultry, and unprocessed red meat were also mutually adjusted for in the models. More details on covariates can be seen in Supplemental Methods 3. For covariates where participants answered "do not know" or "prefer not to answer," these responses were classified as missing. An "unknown" category was created to replace missing values for each covariate; the effect of replacement of missing values was assessed by a sensitivity analysis conducted in participants with complete data on all covariates.
To investigate potential modifying effects of the APOE ε4 allele on risk of dementia from meat consumption, stratified analyses by APOE ε4 carrying status were conducted and additionally P-interaction between each meat type and APOE ε4 status was tested. As a sensitivity analysis, the main analyses were repeated among participants aged ≥60 y at baseline since individuals aged >60 y have a higher risk of incident dementia (27). Statistical analyses were conducted using Stata/IC, version 16.1 (Stata Corp LP).

Results
During a mean follow-up of 8 ± 1.1 y, excluding cases arising in the first year of follow-up (n = 77), 2896 incident cases of allcause dementia occurred, of which 1006 were AD and 490 were VD. Baseline characteristics stratified by dementia status are provided in Table 1. Dementia cases were generally older, more economically deprived, less educated, more likely to smoke, less physically active, more likely to have stroke history and family dementia history, and more likely to be APOE ε4 carriers. More men than women were diagnosed with dementia in the study population. Participant characteristics across 5 categories of reported consumption of processed meat, unprocessed poultry, unprocessed red meat, and total meat are shown in Supplemental Tables 2, 3, 4, and 5 respectively. Generally, compared with those in the lowest category, participants in higher categories of reported consumption of processed meat and total meat were more likely to be men, less educated, smokers, and overweight or obese, and had lower intakes of vegetables and fruits and higher intakes of energy, protein, and fat (including saturated fat).
The associations between each meat type and each dementia outcome were analyzed in three adjustment models. For the incident all-cause dementia (Figure 1), there was a significant  In terms of incident AD (Figure 2), a similar picture to allcause dementia was seen. Higher consumption of processed meat was associated with increased risk of AD

FIGURE 1
HRs (95% CIs) for the associations between incident all-cause dementia and meat consumption in UK Biobank (n = 493,888). The black squares and horizontal lines represent HRs and 95% CIs respectively in Cox proportional-hazards regressions. The distribution of ticks on the x axis is exponential. Participants were categorized based on the data distribution of baseline meat intakes. Mean daily intakes in each category were calculated from the multiple 24-h dietary assessments which were used to test the linear trend per increment. Minimally adjusted models adjusted for age, gender, ethnicity, education, socioeconomic status. Fully adjusted models additionally adjusted for region, smoking status, physical activity, BMI, sleep duration, stroke history, and family history of dementia, and dietary covariates including vegetables and fruits, total fish, tea and coffee, alcohol drinking, processed meat, unprocessed poultry, and unprocessed red meat were also mutually adjusted for.

FIGURE 2
HRs (95% CIs) for the associations between incident Alzheimer disease and meat consumption in UK Biobank (n = 493,888). The black squares and horizontal lines represent HRs and 95% CI, respectively, in Cox proportional-hazards regressions. The distribution of ticks on the x axis is exponential. Participants were categorized based on the data distribution of baseline meat intakes. Mean daily intakes in each category is calculated from the multiple 24-h dietary assessments which were used to test the linear trend per increment. Minimally adjusted models adjusted for age, gender, ethnicity, education, and socioeconomic status. Fully adjusted models additionally adjusted for region, smoking status, physical activity, BMI, sleep duration, stroke history, family history of dementia, and dietary covariates including vegetables and fruits, total fish, tea and coffee, alcohol drinking; processed meat, unprocessed poultry, and unprocessed red meat were also mutually adjusted for. per additional 25 g/d; 95% CI: 1.18, 1.96; P-trend = 0.001). Higher consumption of unprocessed red meat was associated with reduced risk of AD (HR: 0.70 per additional 50 g/d; 95% CI: 0.53, 0.92; P-trend = 0.009). Regarding the risk of incident VD (Figure 3), there were no statistically significant linear trends identified, although for processed meat, the highest consumption categories were associated with increased risk. For all dementia outcomes, 0 times/wk consumption of each meat type appeared to be different from other higher frequencies (Figure 1, 2, and 3); however, most HRs in this category were not significant in the fully adjusted models.
The stratified analyses by APOE ε4 carrying status and P values for interaction between APOE ε4 carriage and meat consumption are shown in Table 2 and Supplemental Table 6. Compared with APOE ε4 noncarriers, carriers had increased risks of developing all-cause dementia by ∼3 times, AD by ∼6 times, and VD by ∼5 times, independent of any type of meat consumption. However, there were no statistically significant interactions between APOE ε4 carriage and meat consumption in the fully adjusted models. Increased risks of incident all-cause dementia were observed per 25 g/d increments of processed meat in both APOE ε4 carriers and noncarriers. However, APOE ε4 carriers but not noncarriers had reduced risks of incident all-cause dementia and incident AD per 50 g/d increment of unprocessed red meat.
When we additionally excluded dementia cases occurring within the first 3-y follow-up (n = 329) for more rigorous controls of potential reverse causality, the HRs were of similar magnitude (Supplemental Figure 2, 3, and 4, Supplemental Table 7). When we conducted a sensitivity analysis in participants with complete data on all covariates (n = 381,809), the HRs were very similar to the main results (Supplemental Figure 5, 6, and 7, Supplemental Table 8). Exclusion of participants aged <60 y at baseline also did not significantly change these associations (Supplemental Figure 8, 9, and 10, Supplemental Table 9).

Discussion
In this population-based, nationwide UK Biobank cohort study our results showed that consumption of processed meat was associated with increased risks of incident all-cause dementia and AD while unprocessed red meat was associated with lower risks. Related cohort studies remain few and inconsistent, and detailed knowledge of which type and amount of meat consumption would be the most influential is not clear. The Three-City Once a week 2.0-4.9 times/wk ≥ . 5 0 times/wk 25 g per day 0 times/wk 0.1-0.9 times/wk Once a week 2.0-4.9 times/wk ≥ . 5 0 times/wk 25 g per day 0 times/wk 0.1-1.0 times/wk 1.1-1.9 times/wk 2.0-2.9 times/wk ≥ .
3 0 times/wk 50 g per day 0 times/wk 0.1-3.0 times/wk 3.1-4.9 times/wk 5.0-6.9 times/wk ≥ . 7 0 times/wk 50 g per day Total meat FIGURE 3 HRs (95% CIs) for the associations between incident vascular dementia and meat consumption in UK Biobank (n = 493,888). The black squares and horizontal lines represent HRs and 95% CI respectively in Cox proportional-hazards regressions. The distribution of ticks on the x axis is exponential. Participants were categorized based on the data distribution of baseline meat intakes. Mean daily intakes in each category is calculated from the multiple 24-h dietary assessments which were used to test the linear trend per increment. Minimally adjusted models adjusted for age, gender, ethnicity, education, socioeconomic status. Fully adjusted models additionally adjusted for region, smoking status, physical activity, BMI, sleep duration, stroke history, and family history of dementia, and dietary covariates including vegetables and fruits, total fish, tea and coffee, alcohol drinking; processed meat, unprocessed poultry, and unprocessed red meat were also mutually adjusted for.
(3C) cohort study took meat consumption of high frequency (≥4 times/wk) as the reference and found that low frequency (≤1 times/wk) was related to an increased risk of incident dementia and AD over 10 y of follow-up (28), which is inconsistent with our findings; however, the methods of collapsing data and reference selection are different. In addition, excessive category combination may have attenuated the study power and specific meat types were not explored in that study. A cohort study conducted in French citizens aged 68 and over showed that compared with daily meat consumers, weekly or less consumers had a higher incidence rate of all-cause dementia and AD after 7 y of follow-up; however, those associations were not significant probably because of small sample sizes (170 incident dementia including 135 AD among 1674 participants) (29). Longitudinal analysis among 2622 elderly German participants suggested no significant association between risk of incident AD and consumption frequency of meat and sausage after 4 y of followup (30); however, this study only investigated single meat items.
Our results also showed that presence of the APOE ε4 allele increased the risk of incident dementia, especially AD; however, there were only minor differences in associations between meat consumption and dementia risk among APOE ε4 noncarriers and carriers, and all P values for interaction were nonsignificant. Currently, evidence on the interaction between APOE genotype and dietary factors with dementia has mostly focused on dietary patterns and dietary fat intake; those studies found older individuals (aged ≥60 y) who had a diet high in fatty fish or higher polyunsaturated fat intake were associated with a decreased risk of all-cause dementia, especially among APOE ε4 noncarriers (31,32). In contrast, studies conducted at midlife found that moderate to high intake of saturated fats in relation to an increased risk of dementia/AD was only detected or more pronounced among APOE ε4 carriers (33,34). A German cohort study of individuals aged 75 + found there was no difference in the association of meat and sausage consumption with incident AD risk between APOE ε4 noncarriers and carriers (30). In addition, a cohort study from eastern Finland showed that the APOE ε4 genotype did not modify associations of egg and cholesterol intakes with risk of incident dementia and AD over ∼22 y of follow-up (35). Inconsistency in these and our study results may reflect particular cohort characteristics; in particular our participants were younger (50-68 y) and this may have led to our insignificant interactions between APOE genotype and meat intake with dementia risk in this population. It is also possible that APOE ε4 carriage is an independent process from dietary aspects in relation to dementia risk.
The underlying reasons for the inconsistent associations between different meat types in relation to dementia risk are not understood. High levels of protein in meat may potentially explain the link between unprocessed meat intake and a lower risk of dementia; adequate protein intake has been linked to a reduced risk of mild cognitive impairment and dementia in the elderly (36). High iron levels in unprocessed red meat may be protective, with iron deficiency being associated with decreased cognitive and attentional processes. Studies in animals have shown a negative impact of iron deficiency on myelination (37). On the other hand, as people age, iron deposits in the brain may impair normal cognitive function. Abnormal iron metabolism triggers oxidative stress, a major contributor to neurodegeneration (38). Processed meat contains nitrites and Nnitroso compounds, which may result in oxidative stress, lipid peroxidation, and activation of proinflammatory cytokines or other mechanisms potentially involved in the development of dementia (39). In addition, as meat consumption increases, intake of saturated fatty acids increases, which has been associated with a higher risk of dementia (40). Processed meat is often high in sodium, and rats fed a long-term high-salt diet had a marked increase in systolic blood pressure linked to reduced regional cerebral blood flow, and potentially linked to cognitive deficit (41). These differences in nutritional composition may explain why consumption of processed meat was associated with a higher risk of dementia rather than unprocessed poultry and unprocessed red meat. These potentially beneficial and negative effects of different meat types on risk of dementia may exist simultaneously, leading to the inconsistent associations seen with meat in this study. A major strength of the current study is that the prospective study with large sample sizes ensured sufficient statistical power. To our knowledge, this is the first study to estimate specific meat types in relation to several dementia outcomes with additional exploration of interactions with the APOE ε4 allele. Other strengths include use of multiple data linkages to maximize capture of incident dementia outcomes, and consideration of reverse causation in analyses. Nevertheless, our study has several limitations. Firstly, the baseline touchscreen brief FFQ only covered some commonly consumed foods and was not suitable to assess total energy or nutrient intakes; systematic bias from self-reported measures at recruitment and low responses to the more detailed repeated 24-h dietary assessments with less than half participants may limit generalizability. Secondly, the UK Biobank cohort study does not have a long follow-up (∼8 y). This will limit our ability to distinguish between reverse causation and causality for risk factors for dementia, as indicated in the Whitehall II cohort study (42). Thirdly, use of linkages to electronic health records may be high in specificity but low in sensitivity; moreover, without linkage to primary care data in our study, milder cases of dementia may have been missed (43). The percentage of AD out of all-cause dementia cases was low in our study (35%) compared with the report of the WHO (50-70%) (2); it is possible that some cases had not been clinically classified by type of dementia, which may attenuate associations between meat consumption and risk of AD. In addition, taking dates of hospital admission and death registry as proxy of diagnosis dates of incident dementia could have resulted in measurement errors; some incident cases might actually be prevalent cases diagnosed prior to hospital admission. Therefore, electronic linkages to accurate primary-care data should be taken into consideration for dementia ascertainment in future research.
Our findings suggest that consumption of processed meat may increase risk of incident dementia, and unprocessed red meat intake may be associated with lower risks, independent of APOE ε4 carriage. On the basis of the findings of this study, more specific public health guidance could be indicated differentiating between types of meat. However further research is recommended to confirm these results. Overall, the research adds to the growing body of evidence linking meat, especially processed meat consumption, to increased risk of a range of noncommunicable diseases.