An observational study published in BMC Medicine looks at the dietary patterns of British adults, and the incidence of cardiovascular disease and death in middle-age.
Prof Kevin McConway, Emeritus Professor of Applied Statistics, The Open University, said:
“This study is complicated to describe, and the findings are complicated to interpret, but generally it is statistically respectable in my view. As is made clear in the paper and the press release, it’s an observational study. This means that people with different diets will differ from one another in other ways too, and so the study can’t establish for sure that the differences in diet are what actually cause the observed differences in cardiovascular disease (CVD) risk and all-cause mortality. For instance, they could differ in terms of smoking, or physical activity levels, or levels of education, which are known to be related to CVD risk and also in some ways to diet. It’s possible to try to allow for these other differences by making statistical adjustments, for factors on which data are available. The researchers did a pretty thorough job of adjusting, but nobody has data on everything relevant, and you can never be sure that everything important has been taken into account. Sometimes, as more information comes in from more and more observational studies, a picture is built up so that researchers can become more confident about what is causing what – but not many studies on nutrition have yet been done using the methods that were used here, so we really can’t be sure that the findings show that the different diets described are causing the differences in the health risks.
“Often, research on the effect of food on health has looked at just a few food types at once, or at just certain nutrients or substances in the food. That’s been problematic, partly because it’s difficult to get a clear picture of how people’s whole diet can affect their health, and partly because dietary advice on specific nutrients in the food (like saturated fats, or sugars) can be difficult to translate into what foods one should actually eat. This research takes a different line. The researchers decided in advance that they wanted to work with four measures of what one might call the quality of what is in people’s diets, for which there is evidence from previous research that they might be related to CVD risk and the risk of premature death. There was a measure of how dense in energy (calories or kJ) people’s food was, as well as of how much saturated fat was in it, how much free sugar, and how much fibre in relation to the amount of energy in the food. Then they looked at the detailed data they had on which foods were eaten by over 100,000 participants in a long-term UK study (UK Biobank), who were middle-aged or older, and used a statistical technique called reduced rank regression (RRR) to find patterns of diet, in terms of how much of which foods were eaten, that were as closely related as possible to the four measures of diet quality. This resulted in the participants being given scores in relation to two different dietary patterns, with, in each case, a high score corresponding on average to relatively poor diet on the four measures they were using. Then the participants were followed up over time (for eight years on average), and the researchers recorded which of them were diagnosed with CVD (heart disease, strokes, etc.), died from CVD, or died from any cause at all. A big advantage of using the participants’ diets as well as the four measures of diet quality is that the findings can be related to the foods people eat, rather than to some diet measures that make sense to nutritionists but not necessarily to others. The association between what foods one eats and the measures of diet quality will be different in different places, because of cultural differences in what exactly people eat – but that’s an advantage of this study for those of us in the UK, because the food patterns come from the diets of a large set of UK people.
“To get a high score on the first dietary pattern, one would have to eat a lot of chocolate, sweets, other sugary foods and drinks, butter and white bread, and low amounts of fresh fruit, vegetables and whole grain foods. Obviously a low score on that pattern would require the opposite kind of diet (lots of fruit and vegetables, not many sweets and not much butter. But the score was based on all the foods that were recorded, and people who (for example) ate not many vegetables but also not much chocolate would also get a score, probably somewhere between the extremes. There were clear links between the score on this dietary pattern and the health risk measures, with people with higher scores having a higher risk of CVD or of dying. These risks were not all that high in people like these participants (because they weren’t so old on average). Across the whole group, about 47 per year in every 10,000 of them had a new diagnosis of CVD, about 9 per year in every 10,000 died of CVD, and about 40 per year in every 10,000 died of all causes. But comparing the risks of these events in people who were in the lowest fifth of scores on this first dietary pattern with the risks in the highest fifth, the risks were between 30% and 40% higher in those with the highest scores. I’ll remind you, though, that we can’t tell from this study whether it is the differences in diet that actually cause these increases in risk.
“The second dietary pattern is quite a lot different. To get a high score, one would have to consume a lot of sugary drinks and fruit juice, eat a lot of sugar or jam, and eat low amounts of butter and high fat cheese. But again, there’s an association with the measures of health risk, though not so strong as with the first dietary pattern. Again comparing the people in the lowest fifth of scores on this second dietary pattern with those in the highest fifth, the risks of CVD diagnoses, CVD death or all deaths during follow-up were between 10% and 20% higher – though again we can’t tell from this study whether those increases in risk are actually caused by the differences in diet.
“This second dietary pattern does look quite a bit different from the first one – in particular, eating a lot of butter and (to a lesser extent) high fat cheese would give someone a high score on the first pattern but a low score on the second, other things being equal. In a way it’s not surprising that there are differences like this – that’s essentially just the way that the RRR statistical procedure works, in that when it decides on the second pattern, it looks at patterns that are statistically unrelated to the first pattern in a particular sense. But it does make things a bit difficult in terms of basing dietary advice on the findings. What should the recommendation be on eating fatty foods like butter and cheese? Or should recommendations just not be made about individual food types like this, and in that case, what kind of recommendations should be made instead? The researchers recommend making further investigations.”
Prof Gunter Kuhnle, Professor of Nutrition and Food Science, University of Reading, said:
“Nutrition research is focusing increasingly on dietary patterns as they are considerably more important than individual foods. In this context, a study investigating the association between specific dietary patterns and health is very welcome. There have been many studies that have looked into specific dietary patterns such as the ‘Healthy Nordic Diet’ or the ‘Mediterranean Diet’ – in contrast to these, this study uses a method commonly used in machine learning to identify different dietary patterns and investigate their impact on health.
“The study is based on a large, well-characterised population with very good outcome data, and has been well conducted. The results show that specific dietary patterns are associated with increased risk of heart disease and early death. The foods that dominate the patterns have been known for some time to affect health: vegetables, fruits and high fibre-foods are generally considered to be “healthy”, while low-fibre, sweet and fat foods are generally considered to be “unhealthy”. A dietary pattern dominated by those “unhealthy” foods is therefore likely to be associated with poorer health outcomes. Of note is the ambiguous role of butter in these patterns – which reflects well the current discussion on the role of dairy.
“The interpretation of these results is difficult: people who consume the main “unhealthy” dietary pattern are more likely to be obese, but they were also more likely to be of lower socio-economic status. The impact of social deprivation on health is well known, and although the authors did adjust for socio-economic status, the results are likely affected by this. Thus while the data clearly support current dietary recommendations, they also highlight the importance to address health inequalities in the UK, in particular access to food.”
Dr Duane Mellor, Registered Dietitian and Senior Teaching Fellow, Aston Medical School, Aston University, said:
“This is an interesting and large study looking at how diet may influence cardiovascular risk and mortality. It has some strengths, in that it considered dietary intake on more than one occasion, which is not always the case in this type of nutritional study. However, the dietary intake is based on 24 hour recall which relies completely on a person’s memory and willingness to say what they eat, and it can depend on time of year and day of the week the survey is done. A strength is also how the cardiovascular disease and deaths were recorded, as it used hospital and death records, so this likely to be very reliable.
“The way dietary patterns were decided was based on 50 food groups, which although better than the 5 or 6 in the EatWell Guide, are not entirely similar foods and is still reductionist and can miss a lot of detail of how people actually eat. This could also explain some of the findings related to cheese and dairy products, as other research has found fermented cheese and dairy products can reduce risk of cardiovascular disease. This can also be seen with the grouping of chocolate with confectionary, where some very high cocoa solid dark chocolates (often specially made for clinical trials) have been linked to reduced cardiovascular risk in both clinical trials and observational studies. However, it is likely that splitting these food groups up, due to the limitations of the 24 hour recall method, despite this being a huge number of people may lack the numbers to be statistically meaningful. The finding of the effect of fruit juice and sugary drinks as a predictor of risk is interesting, this perhaps happened as free sugars were a selected criteria to define the study’s dietary patterns. The inclusion of fruit juices separately from sugary drinks is useful, as it avoids the problem seen with cheeses and chocolate. Perhaps what is important to consider is quantity, for some groups, especially those with less to spend on food, the recommended amount of 150ml per day can be a valuable source of vitamin C. Especially, where food storages and access can be limited.
“The development of the dietary patterns, the authors argue, is more likely to represent a UK diet unlike other patterns used before in this type of study such as the Mediterranean diet. It is based on nutrient intakes that have been hypothesised and shown in other studies to increase risk of cardiovascular disease, these were then mapped to the 50 food groups to determine which foods predicted the most change from what was considered to be an evidence based optimal nutrient intake. This statistical approach is interesting to generate hypothesis, but it highlights only a strong association although the authors suggest there is evidence of mechanism. The problem is this type of study cannot show how an individual chooses the food and the food combinations they eat. The authors claim that this approach mirrors the culinary use of foods, but it perhaps does not achieve this as although this type of work looks at dietary pattern, it does not show how we actually choose to eat. Therefore, to gain a more complete understanding of the effect of dietary patterns, we need to consider why and how people chose to eat and not just link nutrients to food groups which can be reductionist and does not consider the effect of food choice behaviours.”
‘Associations between dietary patterns and the incidence of total and fatal cardiovascular disease and all-cause mortality in 116,806 individuals from the UK Biobank: a prospective cohort study’ by Min Gao et al. was published in BMC Medicine at 01:00 UK time on Thursday 22 April 2021.
Prof Kevin McConway: “I am a Trustee of the SMC and a member of its Advisory Committee. However, my quote above is in my capacity as an independent professional statistician.”
Prof Gunter Kuhnle: “I have received an unrestricted grant from Mars and we collaborate on a research project regarding flavanols.”
Dr Duane Mellor: “Dr Duane Mellor undertook his PhD investigating the effects of high polyphenol chocolate on glycaemia and cardiovascular risk in people with type 2 diabetes. This work was supported in an unrestricted way by Barry Callebaut and Nestle.”