Associations of cereal grains intake with cardiovascular disease and mortality across 21 countries in Prospective Urban and Rural Epidemiology study: prospective cohort study

Study participants

The PURE study is a large prospective cohort study conducted in countries across the regions of North America and Europe, South America, Africa, the Middle East, South Asia, South East Asia, and China. The design and sampling for the study have been published elsewhere.1516 Data on people aged 35 to 70 years at baseline were collected from 21 countries—five low income countries, five lower middle income countries, seven upper middle income countries, and four high income countries, as classified by the World Bank in 2006, covering urban and rural locations. Participants recruited from January 2003 and followed up until July 2019 were included in the analysis.

Data collection procedures

Exposures

The main exposures were the consumption of refined grains, whole grains, and white rice. For this analysis, we categorised grains into three groups: refined grains, whole grains, and white rice. We examined white rice separately from all other refined grains because more than 60% of the PURE population reside in Asia where rice is a staple food. The three grain food categories were mutually exclusive. The definitions used for the grain categories were as follows.

We defined refined grains as wheat grain products or flours that have been modified to remove the bran (the aleurone layer) and germ and therefore have a low fibre content. Products in this category are made with refined (white) flour, including white bread, pasta/noodles, breakfast cereals, crackers, and bakery products/desserts containing refined grains. For bakery products and desserts, we counted only the proportion of refined grains. Ready to eat breakfast made from corn was also included in this group.

The whole grains group contained two types of grains. The first was whole grain flours such as those made from wheat, rye, triticale, oats, barley, maize, finger and pearl millet, sorghum, or buckwheat. Wheat products made from flours ideally contain all the components of the intact grain (whole meal). This applies to all flours that are produced as “stone ground flours”—that is, where millstones, the traditional method of making grain flours, are used to grind flours. With the onset of industrialisation, flours were produced by steel roller mills. In this case, a “white” flour is produced and the germ (lipid) and aleurone layer that contains fibre and protein are separated. The germ, containing the lipid, is then heat treated to prevent rancidity. The bran (that is, the aleurone fibre layer), the heat treated germ, and the “white” flour are recombined in various proportions to reflect their proportion in the intact grain (for example, 25%, 50%, 75%, or 100% of the germ and aleurone layer). This percentage is the extraction rate. As the method of processing is unknown (for example, stone mill or steel roller), we refer to all flours with some increased fibre, over and above the “white flour” content, as whole grain flour. The second type of grain in this group is intact or “cracked” whole grains and whole grain porridges—whole grains that are still associated with their fibre, such as pumpernickel breads, cracked wheat, bulgur, steel cut oats, barley, and oat and maize porridges. The whole grain definition also includes whole corn and cornmeal and dark bread but excludes corn products such as popcorn. Intake of popcorn as a snack was recorded only in Brazil and Canada.

In white rice, the grain has been left recognisably intact but processed to remove the protein and fibre aleurone layer and the germ to make it “white” but visually “intact.” In other words, it is milled rice from which the husk, bran, and germ have been removed. Ready to eat breakfast made of rice was also included in this group. In the PURE study, we were unable to differentiate between regular white rice and parboiled white rice, which contains more of the nutrients associated with brown rice and is more slowly digested and accordingly has a lower glycaemic index. Brown rice consumption was captured only in Argentina and Brazil, where its intake was very low (2 g/day in Argentina and 16 g/day in Brazil). Therefore, brown rice was not included in any grain group.

For each centre, all food containing grains or mixed dishes with grains were identified. All food frequency questionnaires combined gave a total list of 566 refined grain, 207 whole grain, and 176 white rice items. For mixed dishes, we considered the proportion of grains constituting the cooked portion in a recipe while calculating the grain intake in grams for each person. We grouped participants on the basis of their range of intakes into five categories for refined grains (<50, 50 to <150, 150 to <250, 250 to <350, and ≥350 g/day), four categories for whole grains (0, <50, 50 to <100, and ≥100 g/day), and five categories for white rice (<50, 50 to <150, 150 to <300, 300 to <450, and ≥450 g/day). The categorisation for each grain was based on a 100 g increase, except for white rice, for which an increase of 150 g per day (a cup of cooked rice) was considered. Furthermore, for the categorisation of white rice, as a large proportion of participants reported intakes below 150 g per day, intakes below 150 g were separated into two categories. For refined grains, the highest category of 350 g/day or more represented about seven to 10 servings per day mainly as white bread, whole grain intake of 100 g/day or more represented at least three servings per day, and 450 g/day or more of white rice represented at least six servings of cooked rice per day.

To calculate the raw value of whole grain, we calculated dry matter for whole grain foods at the region level. The calculations for dry weight used a coefficient for the water content. Thus, for North America/Europe, South America, and the Middle East, it was 30%, as the main source of whole grain was whole wheat bread.18 It was 50% for Africa, with porridges mostly being consumed, 35% for South Asia and South East Asia, and 40% for China (based on some of the main whole grain recipes from these regions). For dry weight, we categorised whole grains as 0, less than 10, 10-19.9, 20-29.9, and at least 30 g/day. The serving sizes for each country were also calculated as per the US Department of Agriculture.

We used the lowest category of intake for refined grain and white rice and the highest category for whole grains as the reference categories. Participants with plausible energy intakes (500-5000 kcal/day) were included in the analysis.4

Outcomes

The primary outcome of this analysis was the composite of mortality or major cardiovascular disease events (defined as death from cardiovascular causes, non-fatal myocardial infarction, stroke, or heart failure). Secondary outcomes were total mortality, and major cardiovascular disease (fatal and non-fatal myocardial infarction, fatal and non-fatal stroke, heart failure, and cardiovascular mortality). Other outcomes measured were blood lipids (total cholesterol, high density lipoprotein cholesterol, low density lipoprotein cholesterol, total cholesterol to high density lipoprotein cholesterol ratio, triglycerides, apolipoprotein A1, apolipoprotein B, apolipoprotein B to apolipoprotein A1 ratio) and blood pressure measured at baseline.

Statistical analysis

We summarised continuous measurements by using mean and standard deviation if normally distributed; otherwise, we used medians and interquartile ranges. We present categorical observations as frequency and percentages. We created a wealth index by using information collected on household possessions such as electricity, car, computer, television, and telephone. We created a binary yes/no classification for each item and then used a principal component analysis to extract the component with the largest eigenvalue. We then assigned each household to a score based on factor loadings. We categorised education as none or primary school (first six years), secondary school (seven to 11 years), and college, trade school, or university (>11 years); smoking as never, former, or current; and physical activity as low (<600 MET-min per week), moderate (600-3000 MET-min per week), and high (>3000 MET-min per week) activity.

We evaluated the proportional hazards assumption by visual inspection of log-log plots, which were consistent with proportional hazards. We assessed the association between types of grain intake and the time to occurrence of events by using a Cox frailty model with random intercepts to account for clustering by centres (which simultaneously adjusts for region and country). We assumed that frailty followed a γ distribution with correlated observations within cluster. We estimated hazard ratios for both the univariate base model and the multivariable adjusted model and presented them as hazard ratios and 95% confidence intervals. We excluded participants with history of cardiovascular disease at baseline. For all analyses on associations, we adjusted total energy intake by using a standard multivariable model, as the study examined the association between food intake and clinical outcomes.1920 We identified covariates a priori, on the basis of the literature,172122232425262728 as those that were associated with at least one of the events listed and are potential confounders in the association of grains with events. We also included fruit and vegetable intake as a covariate, as these are a major source of fibre. We used one way analysis of variance to examine the association of wealth index, age, waist to hip ratio, and intakes of total energy, fruits and vegetable, and unprocessed red meat with grain intake. We used Pearson’s χ² test for categorical variables such as sex, location (urban/rural), education, smoking, physical activity, and history of diabetes (yes/no). As previous food frequency questionnaire validation studies have shown that the error of variance for sodium is high,29 we did not include dietary sodium as a covariate in our models. However, an analysis with adjustment for saturated fat and sodium is also presented in the supplementary material. We did not include alcohol intake in the model, as this was not captured in five Muslim countries (Bangladesh, Iran, Pakistan, Malaysia, and United Arab Emirates). Soft drink intake in low and middle income countries (except Argentina) was low, so intake of sugar sweetened drinks was not included as a covariate in the models. To evaluate whether history of diabetes could be a mediator, we analysed data with and without diabetes included in the model. Considering the biology and the causal pathway, no colliders were included.

The base model was adjusted for age and sex, with centre considered as a random effect. The minimally adjusted model included the base model with further adjustment for location (urban/rural), education, wealth index, smoking status, waist to hip ratio, physical activity, history of diabetes, energy intake, fruit and vegetable intake, and dairy intake. The fully adjusted model additionally included refined or whole grain intake as appropriate, as grains are eaten in combination. This also allows for adjustment for fibre from other cereal sources. The main results reported are on the basis of the fully adjusted models. We did an additional analysis using the fully adjusted model with medications included (statins and blood pressure lowering drugs). We used restricted cubic splines with four knots (at the 5th, 35th, 65th, and 95th percentiles) to explore the shape of the association of grains with events.

We did a sensitivity analysis of the fully adjusted model by excluding participants who had cardiovascular disease events in the first two years of follow-up to account for possible reverse causation. To account for the clustering of intakes within region, we performed the Cox proportional hazards models with frailty separately within high and low intake regions for refined grains. Low intakes regions were Africa and South Asia, where median intakes were below 100 g, and the remaining regions had high intakes. The whole grain intake in Africa was much higher than in the rest of the regions, so we did a sensitivity analysis that excluded data from Africa and explored the association of whole grains and outcomes. To further account for the clustering of grain intake by region, we estimated hazard ratios by region and obtained an overall estimate by meta-analysis. To examine heterogeneity of the associations of grains with major cardiovascular disease and mortality by geographical region, we did a random effects meta-analysis across all regions. Tests of heterogeneity used the I² statistic.

We could not do subgroup analyses for the Africa region for estimate of composite outcome, total mortality, and major cardiovascular disease events for refined grain, whole grain, and white rice consumption and for South East Asia for estimation of total mortality for refined grain consumption alone, as the sample size in these regions was insufficient for hazard ratio estimation. Additionally, we did a subgroup analysis in the Asian regions (South Asia, South East Asia, and China), where consumption of rice was higher than in non-Asian regions. We also present results with serving size per day as a continuous variable for each grain type. We did further subgroup analyses to assess associations of grains with major cardiovascular disease events and mortality by country income level (high, middle, and low income countries), body mass index (<25 and ≥25), and sex, with tests for interaction.

We also analysed whole grains alone with the raw quantities of whole grains calculated for each region. The results of analysis using serving sizes are also reported.

To examine the association of grain intake with blood pressure and blood lipids, we used multilevel linear regression with random effect models adjusted for centre. For all comparisons, the criterion for the P value to be considered significant was at the 5% level. We used Stata version 14 for all analyses.

Patient and public involvement

Patients and the public were not formally involved in setting the research question or the outcome measures; nor were they involved in developing plans for recruitment, design, or implementation of the study. No patients were asked to advise on interpretation or writing up of results. At each site, community leaders were informed about the study plans to motivate their community to participate in the study. Each participating centre provided results of clinical tests such as electrocardiography and spirometry at baseline and follow-up visits along with information about the progress of the study.