September 8, 2024

expert reaction to unpublished conference abstract on eating fruit, oats, and rye in childhood and type 1 diabetes risk

An unpublished conference abstract presented at the European Association for the Study of Diabetes (EASD) Annual Meeting looks at the association between oat, rye, and fruit intake in childhood and risk of developing Type 1 diabetes. 

Prof Kevin McConway, Emeritus Professor of Applied Statistics, Open University, said:

“It’s pretty frustrating to try to assess the quality of a new piece of research from such limited information. We have just a brief summary (abstract) of the work and a press release, around 1,100 words in all, and much of the content of the abstract is repeated in the press release. The full research paper covering this work will inevitably be much longer and contain far more detail. So far, though, the research won’t yet have gone through a full process of peer review by other scientists in the field.

“The research may make sense as a preliminary look at whether there could be associations between children’s intakes of different foods and the risk of diagnosis of Type 1 diabetes (T1D), or of conditions likely to lead to T1D later. But, as the researchers rightly say, it’s far too early to say whether the findings should lead to dietary recommendations. That’s clear even from the limited information we have so far.

“One important reason for that is that the study is observational – the children weren’t assigned to particular diets by the researchers, but just ate what they would eat anyway. The researchers observed and recorded what they ate, and observed T1D diagnoses and related health outcomes.

“The issue here is that the children who ate different foods would also have differed in many other respects, though we have no detail on exactly which respects. Maybe the correlations between consumption of different foods and the health outcomes, that the researchers reported, were not caused by the foods but by one or more of the other differences. We just can’t tell.

“The researchers did make some statistical adjustments to allow for some possibilities for other differences (energy intake in food, an aspect of the children’s genotype, their sex, and whether there was diabetes in their family). But there are likely to be other important differences that could not be adjusted for.

“This doesn’t rule out that eating different foods could actually cause differences in the risk of T1D, but it does mean that we can’t be at all certain from this study that the foods are the cause of difference in T1D risk.

“It’s also worth remembering that all the children who participated in this study were known to be more susceptible to T1D than average, because of their genetic make-up. That would in any case make it hard to give general dietary recommendations for all children, because things may be different in children who are not susceptible in the same way as the study participants.

“It does, however, make sense to involve only children who are susceptible in a preliminary study like this, which seems to be aimed at finding correlations to investigate further in other research. There were almost 5,700 children in the study, but only 94 of them actually had a T1D diagnosis. That’s rather a small number of T1D diagnoses for investigating associations with food intake (or with anything else really). If the participants had not been more susceptible to T1D than average, the researchers would have had to involve far more children in total, to get enough T1D diagnoses to give meaningful results. The researchers may or may not discuss this point in the full research paper, but we haven’t got that yet.

“There are some other awkward aspects, where more information is needed to make full sense of the findings. We know there were 34 food groups in all, but we don’t know what they all were. Presumably because (for T1D) higher fruit consumption was correlated with higher disease risk, while higher berry consumption was correlated with lower risk, the researchers were not treating berries as part of the fruit group – but what exactly was in the fruit group? What were the findings in all the food groups that aren’t mentioned in the abstract or press release?

“The information we have doesn’t say very much about the statistical methods that the researchers used. They may all be standard and well-accepted, or novel but good, or maybe there’s some issue with them that casts doubt in the conclusions. There’s no way I can check.

“It would be good to know how large the associations between food intake and diabetes risk actually are. Knowing the size of the risks is part of what would be needed to see how important changes in diet might be. The press release gives no figures for this. There are some figures in the research abstract, which gives estimates of the so-called hazard ratio. You can think of as a measure of how different the risk of diagnosis is, in a given period of time, for children who consume large amounts of a food group compared to those who consume smaller amounts.

“But these are hazard ratios are impossible to interpret more precisely, because we don’t yet know what the researchers are counting as large and small amounts of consumption. So actually it was wise to leave all those numbers out of the press release. But to assess the research and its meaning properly, we do need to know much more about the actual numbers.

“One final, very geeky, statistical point is as follows. The researchers looked at three different health outcomes, and at the association of each of them with consumption of 34 different food groups – so they are looking at over 100 different possible associations or correlations between a food group and a health outcome. For just one of these possible associations, even if there’s no true association, random variability between children could make it appear that there is an association. That would be a kind of false positive – it looks as if there is an association but in fact there isn’t really. If one is not careful in analysing the data, and if one looks at lots of associations, the chances that you will turn up false positive associations somewhere in the results can become high. There are several good statistical ways of dealing with this awkward issue, and maybe the researchers used one of these methods. But this is a point of detail that’s not mentioned in the information we have. If the researchers did deal with this properly, that’s great. If they did not deal with it properly, then maybe it’s quite likely that some of the associations they report do not really exist at all – but we still don’t know which ones. That’s maybe another reason that no dietary recommendations should be based on these findings – but it is certainly another reason why it’s impossible to evaluate the work properly on such limited information.”

The abstract ‘Food consumption associated with the risk of islet autoimmunity and type 1 diabetes’ by S.M. Virtanen et al was presented at the European Association for the Study of Diabetes Annual Congress. It was under embargo until 23:01 BST Sunday 8 September.

Declared interests

Prof McConway: “I am a Trustee of the SMC and a member of its Advisory Committee. My quote above is in my capacity as an independent professional statistician.”