I woke up on the morning of Thursday 6th February 2014 to a number of tweets informing me that headlines were claiming “yoghurt is key to beating diabetes.” The on-line version of the Express moderated the headline to “How low-fat dairy food can help to cut your diabetes risk” but the story was all over the news from Australia to India.
This study in the journal Diabetologia was the source article. It is available for free download (as of the time of writing), which is helpful.
There is a study going on across Europe called the European Investigation into Cancer Study (EPIC). It started in 1991 and I just happen to be one of the UK members. I get questionnaires every few years asking me to share health details and asking me what I ate over the previous year. This blog post has a link to the questionnaire.
The Norfolk part of the EPIC study (this is called a cohort) has 25,639 men and women. They were aged 40-79, when they were recruited into the study between the years 1993-1997. The numbers in the Diabetologia paper are confusing at times. It appears that this diabetes/dairy study has randomly selected 4,127 participants – 753 with known cases of type 2 diabetes from the full cohort and 3,502 as a subcohort. This 3,502 group also contains cases of type 2 diabetes (as a valid random selection will reflect the original group). There are 128 cases of type 2 diabetes in the subcohort of 3,502. (Incidents of type 2 diabetes were recorded up to 31 July 2006. There were 892 in the whole study – 25,639 people – giving an incident rate of 3.5%.)
Don’t worry about any of this. It will soon become irrelevant.
I then tried to work out when the dietary questionnaire was undertaken. Is this diabetes/dairy study relying upon the initial food diary questionnaire completed on entry? Or one undertaken during the study? The paper says “Baseline dietary intake data were collected using a 7 day food diary [Reference 15].” Reference 15 is dated 2001 and describes the general method for the Norfolk EPIC study. Does this mean the study is relying upon a food diary from between 1993 and 1997 – baseline entry to the study?
I read on a bit and realised that this too is irrelevant. And this is why…
The characteristics of the subjects
Table 2 is the important one on the study. It puts the 3,502 subcohort people into three groups – those with total dairy intake under 183 grams per day; those with total dairy intake between 184-312 grams per day and those with total dairy intake over 313 grams per day. These may seem strange groups – but they’ve just split the 3,502 into equal numbers of people and seen where the lowest, middle and highest thirds of dairy intake fall.
Table 2 tells us that we ended up with no age differences between the three groups (all had an average age of 59). However that’s where the similarities ended.
|TABLE 2 in the paper||Tertile 1||Tertile 2||Tertile 3|
|g/day||≦ 183||184-312||≧ 313|
|People in group||1,168||1,167||1,167|
|Mean (average) age||59||59||59|
|Total dairy intake – mean g/day||116||245||447|
|Men (% of group)||40.5%||41.1%||48.5%|
|Related factors lifestyle:|
|Smoking (% current smokers)||14.1%||10.7%||9.6%|
|Physical activity (% active)||15.6%||19.9%||21.8%|
|Waist circumference (men) – mean in cm||97.5||95.7||93.9|
|Waist circumference (women) – mean in cm||82.4||81.4||81.3|
|Saturated fat (% total energy)||12.9%||12.9%||13.1%|
|Mononunsaturated fat (% total energy)||12.3%||11.9%||11.6%|
|Polyunsaturated fat (% total energy)||6.7%||6.6%||6.2%|
|Calcium intake (mg/day)||624||807||1,077|
|Magnesium intake (mg/day)||256||289||330|
|Vitamin D intake (mcg/day)||2.38||2.85||2.98|
The Abstract of the paper (the summary) states:
“Results: Total dairy, high-fat dairy, milk, cheese and high-fat fermented dairy product intakes were not associated with the development of incident diabetes. Low-fat dairy intake was inversely associated with diabetes in age- and sex-adjusted analyses (tertile [T] 3 vs T1, HR 0.81 [95% CI 0.66, 0.98]), but further adjustment for anthropometric, dietary and diabetes risk factors attenuated this association.”
In simple terms this is saying:
– We found NO association (positive or negative – no association at all) between the incidence of type 2 diabetes and total dairy OR high-fat dairy OR milk OR cheese OR high-fat fermented dairy products. NO ASSOCIATION AT ALL.
– The claim is then that low-fat dairy intake was inversely associated with diabetes when we only adjusted for age and men/women i.e. a higher intake of low-fat dairy was associated with a lower incidence of diabetes with virtually no data adjustment. This still says nothing about causation. It merely says we observed two things together when we didn’t adjust for things that we need to adjust for.
– The final part says that – when we did properly adjust for all the things that were different, the observation was “attenuated”. This is so disingenuous it’s scandalous. It should have said from the outset, when we adjust for all the differences between the groups there is no association whatsoever between low-fat dairy and diabetes. There is nothing to observe. We have found nothing and there need be no newspaper headlines.
There are 3 models in the paper:
Model 1 adjusts for age and sex and nothing else – that’s the one they use to claim an inverse association between low-fat dairy and incidence of diabetes. Table 2 tells us that there is no difference in age – all groups have an average age of 59. There are more men in the higher dairy group, but so what?
Model 2 adjusts for BMI, family history of diabetes (which is not in table 2 for some unknown reason), smoking, alcohol, physical activity and other lifestyle attributes. We can see from Table 2 that the lower dairy group have 1.6 times the alcohol unit intake of the higher dairy group. The lower dairy group has one and a half times the percentage of current smokers. The lower dairy group has 70% of the physically active people in the higher dairy group. The lower dairy group have higher BMIs and larger waistlines. All of these favour the higher dairy intake having less incidence of diabetes and this having nothing to do with dairy, but lots to do with alcohol, smoking, activity, weight and waist size. And sure enough – when the data is adjusted to compensate for these stark differences, the association disappears.
Model 3 then further adjusts for dietary differences – energy intake, vegetables (higher in the higher dairy/lower diabetes group), processed meat (lower in the higher dairy/lower diabetes group)etc and this also eliminates, sorry, attenuates, any associations.
I was amused by the fact that group 3 – with the higher dairy intake and lower incidence of diabetes also had the highest saturated fat intake, lowest monounsaturated fat and lowest polyunsaturated fat intake, but that didn’t make the headlines. Less amusing is the difference in calcium, magnesium and vitamin D intake between the three groups. What impact did that have on health?
If you like stats, table 3 in the paper shows you all of this. This table takes the lower dairy tertile as being the baseline of 1 and then looks at the middle intake of dairy and the higher intake of dairy relative to this – for models 1, 2 and 3.
Each row gives a mean (average) and then the 95% confidence interval in this kind of format: 0.93 (0.74-1.18). This means that the average was 0.93 but that the confidence ‘range’ around this average lies between 0.74 and 1.18. a) That’s quite wide and b) the number 1 falls within this range so this is not seen as significant – because the comparator of 1 falls within both groups.
You can use this principle to see at a glance – as the paper reported – that nothing is significant for total dairy, high-fat dairy, milk, cheese or high-fat fermented dairy. Low fat dairy is only significant for model 1 – with none of the adjustments that need to be made. There is no association when the lifestyle factors are properly taken into account. Ditto – the association with fermented dairy disappears as soon as lifestyle factors are properly taken into account.
Only one factor nudges out of the significant range. Look at the last line in table 3. The upper limit of the standard deviation is 0.99 – just a notch away from touching the line of no significance and this is for low-fat fermented dairy. And it’s association, not causation. And it’s about relative, not absolute risk. Do you think that was worth today’s headlines?!