Meat consumption and risk of 25 common conditions: outcome-wide analyses in 475,000 men and women in the UK Biobank study

We used data from the UK Biobank study, a cohort of 503,317 men and women from across the UK [17]. Potential participants were recruited through the National Health Service (NHS) Patient Registers and invited to attend one of the 22 assessment centres between 2006 and 2010. Participants joining the study completed a baseline touchscreen questionnaire, provided anthropometric and biological data, and gave informed consent for their health to be followed up through linkage to electronic medical records.

Dietary intake was assessed using a touchscreen dietary questionnaire administered to all participants at baseline that included 29 questions on diet, assessing the consumption frequency of each listed food. Responses to the five questions on meat (unprocessed beef, unprocessed lamb/mutton, unprocessed pork, unprocessed poultry, and processed meat) were assigned values for frequency per week (never = 0, less than once per week = 0.5, once per week = 1, 2–4 times per week = 3, 5–6 times per week = 5.5, and once or more a day = 7). We then collapsed these meat intake frequencies into three or four categories to create approximately equal-sized groups (see Additional file 1: Methods 1 for additional detail).

Participants recruited after 2009, as well as participants who provided UK Biobank with an email address and agreed to be re-contacted, were additionally invited to complete the Oxford WebQ [18], an online 24-h recall questionnaire. Participants were asked to select how many portions of each food item they consumed over the previous 24 h, enabling calculation of mean grams per day by multiplying frequencies of consumption by standard portion sizes. Similar foods were then grouped together into meat types to match the touchscreen dietary questionnaire. We then assigned the mean WebQ meat intakes in participants who had completed at least three WebQs to each touchscreen meat category defined for all participants. Using these assigned means, we calculated trends in risk across categories of baseline meat intakes [4, 19]. This approach uses repeat measurements to estimate usual mean meat intakes in each category of meat intake, thereby reducing random error in the assessment of usual meat consumption (see Additional file 1: Methods 1 for additional detail).

The outcomes of interest in this study were incident cases of 25 common conditions. The conditions selected were those identified as the 25 leading, well-defined causes of non-cancerous hospital admission in this cohort based on the primary International Classification of Diseases (ICD) 10 diagnosis codes recorded during admission. Some of the commonest causes of hospital admission in this cohort (e.g. nausea or heartburn) were not considered to be separate conditions, because they were not well-defined and/or were likely to be associated with a diverse range of underlying conditions. Moreover, although diabetes was not among the 25 most common primary diagnoses associated with admission, it is a common secondary reason for admission and therefore any diagnosis of diabetes was included among the 25 common conditions examined (see Additional file 1: Table 1 for selected conditions and relevant diagnosis, and procedure codes).

Participant information on cause-specific in-patient hospital admissions and deaths (primary cause for all outcomes except diabetes which also included any diagnosis for hospital admission or mention on the death certificate) was obtained through linkage to the NHS Central Registers. For participants in England, Hospital Episode Statistics (HES) and information on date and cause of death were available until the 31st of March 2017; for participants in Scotland, Scottish Morbidity Records and information on date and cause of death were available until the 31st of October 2016; and for participants in Wales, the Patient Episode Database and information on date and cause of death were available until the 29th of February 2016. We also obtained information on cancer registrations (including date and cancer site) from the NHS Central Registers (see Additional file 1: Methods 2 and Additional file 1: Table 1 for information on exclusion, diagnosis and procedure codes).

Of the 503,317 recruited participants, 28,332 were excluded due to study withdrawals, prevalent cancer (except non-melanoma skin cancer, ICD-10 C44), or because their genetic sex differed from their reported gender, resulting in a maximal study sample of 474,985 (94%). Participants with a relevant diagnosis or procedure prior to recruitment, ascertained through the touchscreen questionnaire, nurse-guided interviews, and hospital admission data, were excluded for each condition (see Additional file 1: Table 1 for details about the exclusions for each outcome). Participants who did not report their meat intake in the touchscreen questionnaire or reported ‘prefer not to say’ or ‘do not know’ were classified as missing and excluded for the respective exposure analyses (see Additional file 1: Fig. 1 for participant flowchart and Additional file 1: Tables 6, 7, 8, 9, and 10 for total numbers for each exposure and outcome).

We used Cox proportional hazards regression models to assess associations between meat consumption and risk for incident cases separately for each disease or condition, calculating trends using the mean meat intakes calculated using the WebQ questionnaires for each category from the touchscreen questionnaire and the trend test variables. Participants’ survival time in person-years was calculated from their age at recruitment until their age at hospital admission, death, loss to follow-up, or administrative censoring. All analyses were stratified by sex, age at recruitment, and geographical region (Model 0). In Model 1, we estimated hazard ratios (HRs) and 95% confidence intervals (CIs) adjusted for race, Townsend deprivation index [20], education, employment, smoking, alcohol consumption, and physical activity, and in women, we additionally adjusted for menopausal status, hormone replacement therapy, oral contraceptive pill use, and parity. In Model 2, we further adjusted for total fruit and vegetable intake, cereal fibre intake score (calculated by multiplying the frequency of consumption of bread and breakfast cereal by the fibre content of these foods [21]), oily fish intake, and non-oily fish intake. For Model 3, we added adjustment for body mass index (BMI). Missing data for all covariates was minimal (< 10%) and thus a ‘missing’ category was created for each covariate (see Figs. 1, 2, 3, and 4 footnotes and Additional file 1: Methods 3 for full adjustment description with definitions of categories)

Table 1 shows baseline characteristics of participants by categories of unprocessed red meat and processed meat intake. Around one-third of participants consumed unprocessed red and/or processed meat once or more daily. On average, participants who consumed unprocessed red and processed meat regularly (three or more times per week) were more likely to be men, older, of White European race, retired, have higher BMI, smoke and consume alcohol, and consume less fruit and vegetables, fibre, and fish and more poultry meat; they were also less likely to have attained a tertiary education, and among women to have two or more children, not use oral contraceptives, use hormone replacement therapy, or be postmenopausal compared with participants who consumed meat less than three times per week (P < 0.001 for heterogeneity between meat intakes for all baseline characteristics). Participants who consumed higher amounts of unprocessed red meat were more likely to consume higher amounts of processed meat and poultry meat (see Additional file 1: Table 3). Baseline characteristics in relation to poultry meat consumption were somewhat different (see Additional file 1: Table 5).

Figures 1, 2, 3, and 4 present the numbers of incident cases for 25 common conditions and their HRs and 95% CIs per unit higher intake of meat for the multiple-adjusted model (Model 3) over an average follow-up of 8.0 years (standard deviation 1.0). Risks by categories of meat intake at baseline for Models 0–3 can be found in Additional file 1: Tables 6, 7, 8, 9, and 10. Overall, many of the positive associations were substantially attenuated, and in some cases were no longer statistically significant, with the additional adjustment for BMI (Model 3). Here we describe the results for Model 3 that were robust to correction for multiple testing. Risks for total meat intake (unprocessed red, processed, and poultry meat combined) did not yield any additional associations and these results are therefore only presented in Additional file 1: Table 6 and Fig. 7.

Similar to our findings, a recent meta-analysis of prospective studies [6] and a recent prospective study from the Pan-European EPIC cohort which included over 7000 IHD cases [9] reported positive associations between unprocessed red meat and processed meat consumption and risk of IHD. For stroke, previous meta-analyses of prospective studies [22, 23] and a recent prospective study from the EPIC cohort [24] both reported null associations for unprocessed red and processed meat intake and haemorrhagic stroke; this is consistent with our main findings but not with our findings in participants diagnosed after 4 or more years of follow-up, although this might be a chance finding due to shorter follow-up. Processed meats contain high amounts of sodium [25], a risk factor for high blood pressure [26], which is a causal risk factor for IHD and stroke [27]. Furthermore, unprocessed red meat and processed meat are major dietary sources of saturated fatty acids (SFAs) which can increase low-density lipoprotein (LDL) cholesterol, an established causal risk factor for IHD [28]. It is also possible that the positive association we observed for unprocessed red meat intake and IHD risk might relate to gut microbiota metabolism, for example through the production of trimethylamine-N-oxide [25‐27], but the importance of this potential pathway is uncertain.

Higher consumption of unprocessed red and processed meat was associated with a higher risk of pneumonia. To the best of our knowledge, these associations have not been shown previously, except for one recent study that found that higher intake of red meat (both processed and unprocessed) was associated with a higher risk of death due to respiratory disease, which included pneumonia [21]. It is possible that the observed association might reflect a causal link, for example related to the high availability of iron in unprocessed red and processed meat (see further discussion below in relation to anaemia), since excess iron has been found to be associated with a higher risk of infection [29] and increased availability of iron for pathogens [30]. It is also possible that hospital admission for pneumonia is a marker for co-morbidity and overall frailty [31]; therefore, residual confounding might operate (see further discussion on residual confounding below).

Few prospective studies have examined the risk for diverticular disease [32, 33], but consistent with our findings, the Health Professionals Follow-up Study (HPFS) observed increased risks of incident diverticulitis with higher consumption of unprocessed red and processed meat [32]. The HPFS did not observe an association for poultry meat, but had lower power than the current study. Meat consumption might affect the risk of diverticular disease via the intestinal microbiome, by altering microbial community structure and metabolism [34].

A recent meta-analysis of prospective studies reported that unprocessed red and processed meat consumption was positively associated with the risk of colorectal adenomas [35], which is consistent with our findings for colon polyps. Unprocessed red meat is a source of heme iron and processed meat usually contains nitrite and nitrates; these can increase the formation of N-nitroso compounds [36], which are mutagenic and have been associated with a higher risk of colorectal adenomas [37].

To our knowledge, this is the first prospective study of meat consumption and risk of GERD and gastritis and duodenitis. We found a positive association between poultry meat intake and GERD risk, whereas the available cross-sectional evidence suggests a null association for meat (total) [38‐41]. We also found a positive association between poultry meat consumption and risk of gastritis and duodenitis. Helicobacter pylori, a bacterium that increases the risk of gastritis [42], has been previously detected in raw poultry meat [43]. Therefore, it is possible that the observed association might relate to inappropriate handling or cooking of poultry meat, but additional research is needed.

Some published studies have found evidence of an association between higher unprocessed red and processed meat consumption and gallbladder disease which remained after BMI adjustment [16, 17], whereas in our analyses this association was greatly attenuated and not significant after adjusting for BMI. In the present study, BMI was calculated from standardised measurements of weight and height, whereas previous studies used self-reported weight and height. Therefore, it is possible that adjusting for BMI in this study explained a larger proportion of the observed associations; high BMI has been consistently shown to be associated with a large increase in the risk of gallbladder disease in both observational and genetic studies [18‐20]. We observed a novel association between poultry intake and gallbladder disease, though additional research is needed to assess this association.

We found an inverse association between the consumption of unprocessed red meat and poultry meat and risk of IDA. Some previous evidence from prospective studies [44] supports these findings and has also shown a positive association between unprocessed red meat [45] and total meat [46‐48] consumption and indicators of body iron stores. Moreover, previous cross-sectional work from the UK Biobank has shown that people who did not consume meat were more likely to be anaemic [49]. This association is likely related to the high availability of heme iron in meat, which is more easily absorbed than non-heme iron [50].

Similar to our findings, meta-analyses of prospective cohort studies have consistently reported a positive association between unprocessed red and processed meat consumption and risk of diabetes [7, 51, 52]. We also found a positive association between poultry meat consumption and risk of diabetes, which has been reported in some [14] but not all prospective studies [13, 53]. Obesity is the major risk factor for diabetes, and the association for unprocessed and processed meat intake (combined) and diabetes in the present study was substantially attenuated (by ~ 60%) after adjusting for BMI, suggesting that the remaining association with meat may be entirely due to higher adiposity. It is also possible that meat could affect risk independently of adiposity; for example, high intakes of heme iron and greater iron storage may promote the formation of hydroxyl radicals that damage the pancreatic beta cells, thereby impairing insulin synthesis and excretion [54, 55].

In the present study, most of the positive associations between meat consumption and health risks were substantially attenuated after adjusting for BMI, suggesting that BMI was a strong confounder or possible mediator for many of the meat and disease associations. BMI is an important risk factor for many of the diseases examined (e.g. diabetes [7]). BMI was highest in participants who consumed meat most frequently, and some previous studies have found that high meat consumption is associated with weight gain [56, 57], but it is unclear whether this indicates any specific impact of meat or an association in these populations of high meat intakes with high total energy intakes. The associations of meat with disease risk reported here which remain after adjustment for BMI might still be due to higher adiposity, because BMI is not a perfect measure of this characteristic; we observed similar effects when adjusting for waist circumference (results not shown), but, as with BMI, waist circumference is not a perfect measure of adiposity and there could still be residual confounding.

As far as we are aware, this is the first outcome-wide study of meat intake and risk of 25 common conditions (other than cancer). Additional strengths of this study include the large size of the cohort, its prospective design, and the comprehensive array of confounders considered. This allowed us to investigate a large number of common conditions and thus avoid outcome selection bias, while simultaneously controlling for confounding. Additionally, we used national record linkage to ascertain information on disease incidence, which is objective and minimises selective loss to follow-up. Nevertheless, some potential methodological issues should be considered when interpreting our findings. Some measurement error would have occurred while measuring meat consumption at baseline; however, we reduced the impact of random error and short-term variation in diet by using the repeated 24-h recall WebQ data and applying corrected intakes to each category of the baseline intakes. Another limitation was that the touchscreen dietary questionnaire only included a subset of food groups and food items and therefore total dietary intake could not be calculated, and confounding by energy balance could not be directly accounted for. We addressed this by adjusting for BMI, physical activity, and other dietary factors [24]; however, there might still be some residual confounding by energy intake. Likewise, participants who consumed high amounts of unprocessed red meat also consumed high amounts of processed meat. Therefore, we could not mutually adjust the meat types, and there may be residual confounding. Multiple testing might have led to some spurious findings; we addressed this by using Bonferroni correction, but this is a stringent approach and it is possible that some real associations did not meet the Bonferroni threshold. Another consideration is the use of hospital records for incident case ascertainment. Some conditions might only require hospital use at later stages (e.g. diabetes), and therefore, some admissions might reflect prevalent and/or more severe cases. Finally, given the observational nature of this study, it is possible that there is still unmeasured confounding, residual confounding, and reverse causality. For instance, in analyses restricted to never smokers, some of the adjusted risk estimates were lower than in the main analysis (e.g. for unprocessed red meat intake and diabetes and for processed meat intake and IHD), suggesting that even after adjustment for smoking there may be residual confounding. However, most of our results were similar after excluding participants who smoked or formerly smoked and after excluding the first 4 years of follow-up.