Red meat, poultry and fish consumption and risk of diabetes: a 9 year prospective cohort study of the China Kadoorie Biobank

Diabetes is a major public health problem affecting almost half a billion people worldwide [1]. In China, the incidence of diabetes has increased rapidly since the early 1980s, currently affecting ~10% of adults [2]. About half of this increase can be attributed to increasing adiposity [3] and other lifestyle factors (e.g. smoking [4], reduced physical activity and certain dietary habits [5]), which may influence diabetes risk either independently or through adiposity.

Several prospective studies, conducted chiefly in western populations, have reported that higher consumption of red meat is associated with a higher risk of diabetes [6, 7]. Potential underlying mechanisms may include obesity and iron overload, which may lead to pancreatic beta cell dysfunction and impaired insulin sensitivity [8, 9]. In addition, other components of red meat, such as cholesterol, saturated and trans fatty acids, and amino acids, may also influence diabetes risk directly or through impacts on gut microbiota [6, 10]. Thus far, the only reported prospective study from China observed no such positive association [11]. People in China tend to have a lower mean BMI and consume mainly plant-based diets with lower amounts of red meat compared with people in western countries [12]. Moreover, Chinese people tend to consume pork, which contains lower amounts of iron than beef and lamb, which are more commonly consumed in western countries [13]. Therefore, reliable evidence about the association between red meat intake and risk of diabetes is particularly needed in China and other populations with relatively low body iron status [14].

The lower fat content of poultry and fish when compared with red meat has led to them being considered relatively healthier for cardiovascular disease [15, 16]. However, existing evidence does not support a clear overall association of either poultry or fish intake with diabetes risk [6, 17, 18]. We therefore investigated the associations of red meat, poultry and fish consumption with incidence of diabetes in the prospective China Kadoorie Biobank (CKB) study. In addition to the overall associations, we also examined the role of adiposity and whether sociodemographic and lifestyle factors, which are potentially related to nutritional status, might modify the associations [19].

The mean (SD) baseline age of the study participants was 51.2 (10.5) years, 59% were women and 58% resided in rural areas. At baseline, 28.7% of the participants consumed red meat on a daily basis and 17.3% reported <1 day/week consumption (Table 1). Those participants who consumed red meat more frequently were younger, male, urban residents, and had higher education and income levels. They were also more likely to be regular smokers and regular alcohol drinkers. Similar associations between age, sex, SES, and alcohol intake and poultry and fish consumption were also found (ESM Table 1). Except for whole grain and preserved vegetables, consumption of other dietary variables was positively correlated with red meat, poultry and fish consumption, particularly fresh fruit, dairy products and eggs.

The estimated usual mean daily consumption was 55.1 g for red meat, 14.4 g for poultry and 23.1 g for fish, higher in men and in urban areas (ESM Fig. 2). Fish intake showed the largest urban vs rural difference among the three food groups investigated, with Qingdao and Haikou (two coastal urban areas) having the highest mean usual consumption and Gansu and Henan (two inland rural areas) having the lowest (data not shown).

Red meat consumption was positively associated with BMI (ESM Fig. 3), with men and women who consumed red meat daily having a 0.3 and 0.7 kg/m² higher BMI, respectively, compared with those consuming red meat less than once weekly. After accounting for BMI, red meat consumption was not clearly associated with waist circumference and BF%. Likewise, fish consumption was also positively associated with BMI in both men and women, while poultry consumption was positively associated with BMI only in men.

During a mean follow-up of 9 years (~4.5 million person-years), 14,931 incident diabetes cases were recorded at age 35–79 years (incidence rate ~3300 per 10,000). After adjustment for all above-mentioned covariates (including the other two food groups under investigation but not BMI), consumption of red meat and fish was positively associated with risk of diabetes with HR 1.19 (95% CI 1.15, 1.23) and 1.15 (95% CI 1.08, 1.23), respectively, for the highest vs lowest consumption category (Table 2); no such association was noted for consumption of poultry. After correcting for regression dilution bias, each 50 g/day increase in consumption was associated with HR of 1.19 (95% CI 1.11, 1.28) for red meat and an HR of 1.12 (95% CI 1.06, 1.19) for fish. Additional adjustment for BMI attenuated the HR to 1.11 (95% CI 1.04, 1.20) for each 50 g/day increment in usual red meat intake and 1.06 (95% CI 1.00, 1.13) for each 50 g/day fish intake, respectively, corresponding to 36.8% (26.1%, 55.8%) and 46.1% (28.4%, 78.4%) attenuation. Poultry consumption was not associated with diabetes risk before or after adjusting for BMI.

The association of red meat intake with diabetes was more pronounced in men than in women (HR 1.23 [95% CI 1.09, 1.39] vs HR 1.06 [95% CI 0.97, 1.15] per 50 g/day usual consumption) and was more pronounced in urban areas than in rural areas (HR 1.25 [95% CI 1.11, 1.40] vs HR 1.03 [95% CI 0.94, 1.13]) (p_{heterogeneity} = 0.05 and 0.01, respectively; Fig. 1). In addition, the association was stronger in those with higher education level (p_trend = 0.03), those who were current alcohol drinkers and those with family history of diabetes, although the heterogeneity test was not statistically significant for the latter two. BMI significantly modified the association between red meat consumption and diabetes risk; the association was only significant in the overweight group (not the other two groups). Similarly, the association between fish consumption and risk of diabetes was stronger in residents of urban vs rural areas, in those with higher education and in those with a family history of diabetes (Fig. 2).

Stratified analyses by sex and area (urban vs rural) showed that both red meat and fish were positively associated with diabetes risk in urban men and women, but not in those from rural areas (Figs 3, 4). Compared with participants who reported consumption of red meat on <1 day/week (the lowest group) at baseline, daily consumption (the highest group) was associated with an HR (95% CI) of 1.32 (1.25, 1.41) in men from urban areas, 1.18 (1.12, 1.25) in women from urban areas, 1.11 (1.02, 1.20) in men from rural areas and 0.96 (0.89, 1.04) in women from rural areas. For each 50 g/day increment in usual red meat intake, the corresponding HR (95% CI) was 1.42 (1.15, 1.74), 1.18 (1.03, 1.36), 1.11 (0.95, 1.30) and 0.98 (0.88, 1.10), respectively, in these four subgroups (Fig. 3). For fish consumption, the HR (95% CI) for diabetes risk in men and women from urban areas and men and women from rural areas reporting the highest consumption was 1.33 (1.19, 1.49), 1.10 (1.00, 1.21), 1.00 (0.85, 1.19) and 0.91 (0.76, 1.08), respectively, compared with those who never or rarely ate fish (Fig. 4). Each 50 g/day increase in usual consumption of fish was associated with an HR (95% CI) of 1.15 (1.02, 1.30), 1.11 (1.01, 1.23), 1.04 (0.91, 1.20) and 0.92 (0.81, 1.05), respectively, (Fig. 4).

Across five urban and five rural areas, the association between red meat intake and diabetes risk varied a little but there was significant heterogeneity across five individual urban areas for fish intake in relation to diabetes risk (ESM Fig. 4).

Sensitivity analyses, including the exclusion of the first 2 years of follow-up, additional adjustment for other dietary factors (i.e. egg, dairy products, soybean, fresh and preserved vegetables, and whole grain staple foods) and other adiposity indices (i.e. waist circumference and BF%), and exclusion of participants who developed CVD and cancer during follow-up, did not materially alter the main results described above (ESM Table 2).

In this large prospective study of Chinese adults, consumption of both red meat and fish (but not poultry) was positively associated with the risk of developing diabetes, with each 50 g/day increase in usual intake associated with 11% and 6% higher risk, respectively. These associations were independent of other dietary factors and adiposity and were more pronounced in urban areas than rural areas and in men than in women.

Red meat is a major source of valuable proteins, essential amino acids, vitamins (e.g. vitamin B₁₂) and minerals (e.g. zinc and iron) [31]. Consumption of red meat, however, has been associated with higher risks of several chronic diseases, including diabetes, in western populations. In two large meta-analyses of prospective cohort studies involving >442,000 adults and >17,000 individuals with diabetes, each 100 g/day increase in red meat consumption was associated with 13% and 19% higher risk of diabetes, respectively [6, 7]. To date, only one prospective study from China, involving ~75,000 middle-aged women and ~2000 individuals with diabetes, has been published on this topic, reporting no clear association between red meat consumption and risk of diabetes (HR 0.94 [95% CI 0.80, 1.10] for top vs bottom quintile) [11]. By contrast, in a similar-sized prospective study of Chinese adults living in Singapore (57% women), the HR (95% CI) for diabetes comparing highest vs lowest consumption quartile was 1.23 (1.14, 1.33) [32]. The present study included more incidences of diabetes than these two previous Chinese studies combined and found a statistically significant positive association between red meat consumption and diabetes, particularly in men and women living in urban areas. Moreover, our study findings showed that over one-third of the positive association could be explained by adiposity, which is the most important risk factor for diabetes [3].

The observed clear patterns of effect modifications by sex, area (urban vs rural), and education level suggest a plausible mediating role of iron overload in the association between red meat intake and diabetes [9]. That is because women, participants living in rural areas and people with lower SES tend to have relatively lower levels of body iron stores [33] and, as such, eating red meat may not lead to iron overload and diabetes. In addition, the much stronger associations seen in current alcohol drinkers and participants with a family history of diabetes might also relate to a higher level of iron storage because drinking alcohol and carrying certain genetic variants could lead to a higher rate of dietary iron absorption [34, 35]. However, directly measured iron status was not currently available in our study and future studies with this information are required to confirm or refute this hypothesis.

In the present study population, the mean level of consumption was much lower for poultry than red meat and its null association with diabetes is consistent with the overall available evidence to date [6]. Although the consumption of fish is generally recommended by most dietary guidelines for prevention of CVD [36], existing evidence overall does not support any clear beneficial association with diabetes. For example, in a meta-analysis of 13 cohort studies involving 481,489 participants and 20,830 cases of incident diabetes, fish intake was not in an overall significant association with diabetes incidence (RR 1.12 [95% CI 0.94, 1.34] per 100 g/day), with substantial heterogeneity across different cohorts [17]. However, in separate analyses by geographic location of study populations, fish intake was inversely associated with risk of diabetes (HR 0.89 [95% CI 0.98, 0.81] per 100 g/day intake) in five Asian cohorts (including two from China), involving ~7100 diabetes cases, but positively associated with diabetes risk (HR 1.38 [95% CI 1.13, 1.70]) in eight North American/European cohorts (~13,700 cases). Though less extreme, our findings, particularly among participants residing in urban areas, were broadly consistent with those from previous western studies conducted in North America and Europe [37, 38]. The potential mechanisms underlying this positive association are unclear but both iron [39, 40] and environmental contaminants (e.g. mercury) in fish might play a role [41]. In addition, higher consumption of animal-sourced foods (i.e. red meat and fish) is very likely associated with affluence-related dietary patterns, which may increase the risk of diabetes in China [42].

The present study has several major strengths. First, the large sample size in the CKB allows us to exclude diagnosed and undiagnosed diabetes as well as major chronic diseases (i.e. CVD and cancer) at baseline, thus limiting the potential influence of reverse causality. Second, the analyses controlled for a wide range of potential confounding factors and regression dilution bias caused by long-term variation and inevitable measurement errors in self-reported dietary exposure variables [43]. More importantly, the main exposure variables in the present study had good reproducibility and validity.

The study also has several limitations. First, we collected outcome information through linkage with hospitalisation records, so some non-hospitalised incidences of diabetes may have been missed. However, we have previously observed that diabetes prevalence based on the CKB re-survey population was reasonably consistent with nationally representative surveys [3]. On the other hand, such under-reporting in outcome measures would most likely be non-differential and thus not overestimate risk estimates. Second, our dietary questionnaire is relatively simple, collecting consumption data for only some of the major food groups instead of individual food items. Hence, it was not possible to adjust for total energy intake and other specific dietary factors (e.g. saturated fat, salt and dietary fibre). However, total energy intake should not play a major role in the observed associations because our main analyses were adjusted for both BMI and physical activity, which together could be considered as a good proxy for total energy intake. Third, we were unable to distinguish between unprocessed and processed red meat or between different types of red meat (i.e. pork, beef or lamb) and fish (e.g. fatty fish or lean fish). However, the nationally representative nutrition survey showed that unprocessed pork accounts for ~80% of total red meat consumption in China [44]. Fourth, the observed heterogeneities across BMI categories (for red meat) and across regions could not be properly explained. Last, as in all observational studies, the possibility of residual confounding cannot be ruled out and causality cannot be automatically assumed.

In summary, in this large study of a Chinese adult population, higher consumption of red meat was associated with higher risk of new-onset diabetes and this association was only partly explained by adiposity. The association appeared to be stronger in men, participants residing in urban areas and those with a higher education level. Our data do not support the inverse association between fish consumption and diabetes risk previously reported in some Asian studies. Further studies are warranted to understand the exact mechanisms linking red meat and fish consumption with increased risk of diabetes.