Wednesday, 29 April 2015

Are 35% of deaths due to a 'poor diet'?

Graham MacGregor, the chair of Action on Sugar, is in the BMJ today complaining that the government hasn't compelled food companies to reduce salt as much as he wants. (MacGregor is also the chair of Consensus Action on Salt. What a small world.) He begins with this claim...

Poor diet is now the biggest cause of death and ill health in the United Kingdom and worldwide.

This may or may not be true, but neither of the references he cites even mention the UK so we can't tell. The first is an op-ed by Marion Nestle about 'Big Food'. The second is a webpage about a study about the USA. Aren't peer reviewers supposed to spot these errors?

MacGregor's claim is modest compared to that of Aseem Malhotra and friends in their astonishing op-ed last week (see here for a reminder). They assert that:

According to the Lancet global burden of disease reports, poor diet now generates more disease than physical inactivity, alcohol and smoking combined.

This is a surprising claim to anyone who is familiar with the statistics. Although the authors name the Lancet report, they do not include it in their citations, perhaps because they would then feel obliged to name the page in which the study made the claim. No such page exists because the claim isn't in the study.

Nevertheless, the notion that "poor diet" is worse than smoking, drinking and physical inactivity combined soon told hold. A nifty graphic appeared on the Telegraph website and was soon doing the rounds on Twitter.

These figures are not mentioned in the Lancet report either, but the credit to Simon Capewell gives us a clue as to how they were arrived at. The 34.6 per cent figure for "poor diet" is wrong. Badly wrong.

Here's what Capewell did. He went to the Lancet report and looked at the graph that shows the percentage of disability-adjusted life years attributed to different factors. He added the diet-related factors together and came up with a big number.

I know that's what he did because the slide below comes from a presentation he gave in 2013.

(Incidentally, I doubt Malhotra and friends would be happy with the content of this slide presentation. It's all about how evil saturated fat is. It was, however, created before the Year Zero of 2014 when Action on Sugar was launched and Capewell declared that "sugar is the new tobacco".)

I can't be certain, but I would bet a few quid that this is the source of Malhotra's claim. When the press release for his study was being prepared, either he or Capewell simply subtracted the 5.4 per cent for alcohol to get the 34.6 per cent seen in the Telegraph graphic.

What's wrong with that? Well, pretty much everything. For a start, he's included every death from high blood pressure, high body-mass index and high cholesterol under "diet". These factors can be influenced by diet, to be sure, but there are many other causes. It is quite wrong to group them together in this way. (As for including alcohol use under "diet" in his original slide, this is unconventional, to say the least.)

Moreover, you simply can't add these figures together. One of the counter-intuitive aspects of epidemiology is that it creates attributable fractions for risk factors which, if added together, would suggest that there are many more deaths than there are people dying. If this seems peculiar, I sympathise, but that's how it is and researchers in the field are quite sanguine about it (Carl Phillips gives an explanation for the layman here). 

The authors of the Lancet report make this clear at the very start of the section on 'results', saying:

The effects of multiple risk factors are not a simple addition of the individual effects and are often smaller than their sums, especially for cardiovascular diseases, which are affected by several risk factors (eg, table 2). The sum of the individual effects of just the metabolic risk factors at the global level is 121% and the summation of all the risks is greater than 400%.

In other words, if you added all the risk factors in their report together (there are many), you would conclude that there are four times as many deaths as there actually are. You would be double-counting all over the place and that's why you don't do it.

But even if you did try to compare different risk factors, you wouldn't use Capewell's method. It is bizarre that he chose to work out the percentages from the graph above (presumably using a ruler!) when the exact figures are listed in Table 2 and Table 3. If you look at those figures, you'll see the following estimates for disability-adjusted life years:

Dietary risk factors and physical inactivity: 254,333,000

Alcohol use: 136,063,000

Tobacco smoking (including second-hand smoke): 156,838,000

Physical inactivity: 69,318,000

You only need to subtract the figure for physical inactivity from the figure for 'dietary risk factors and physical inactivity' to get the total DALYs attributed to a 'poor diet'. It is 185,015. This is much lower than the combined figure for alcohol, tobacco and inactivity, which is 362,219.

You can do the same with the figures for mortality in Table 2:

Dietary risk factors and physical inactivity: 12,503,370

Alcohol use: 4,860,168

Tobacco smoking (including second-hand smoke): 6,297,287

Physical inactivity: 3,183,940

Again, subtracting the inactivity total from the top total gives 9,319,430 for dietary risk factors. This is much less than the 14,341,395 for alcohol, tobacco and inactivity combined.

Adding alcohol, tobacco and inactivity together entails some of the problems of confounding discussed above, but these risk factors are more likely to independent of one another and therefore less problematic than bundling a bunch of dietary risk factors together (drinking doesn't cause lung cancer, smoking doesn't cause liver cirrhosis, etc.)

But if you start adding 'diet low in fruits' to 'diet low in fibre', for example, you begin the process of double, triple and quadruple counting. Sure enough, if you add together all the deaths from 'diet low in calcium', 'diet low in milk' and so on together, the total exceeds 18 million, which is fifty per cent more than the figure for all dietary risk factors and physical inactivity combined in the Lancet report itself. This is why—if I may labour the point—you don't do it.

To an epidemiologist, this really is a schoolboy error, but it is exactly what Capewell did. He compounded the error by adding in a huge numbers of deaths that are not diet-related and are not considered to be diet-related in the original report. This misinformation was then passed onto Malhotra and friends who used it in their op-ed, a graphic was created and a meme was born.

But if "poor diet"—which, by the way, at the global level usually means malnutrition rather than excessive sugar and salt consumption—does not cause 34.6 per cent of all disease, how much does it cause?

You don't need to crunch numbers or estimate percentages from graphs to work this out. All you need to do is go to page 2,253 of the Lancet report and you will see these words...

We have expanded the set of components of diet included from a combined category of fruits and vegetables in the 2000 comparative risk assessment to 15 components in GBD 2010; together these dietary risk factors account for a tenth of global disease burden.

So it's 10 per cent, not 34.6 per cent (and the researchers also say that "we stress that these results should still be interpreted with caution"). This is not a trivial difference, by any means. It amounts to claiming that a quarter of all deaths worldwide—some 14 million—are due to poor diet when they are not. And since more than 10 per cent of deaths are attributed to smoking, alcohol and physical inactivity, this means that the claim in the Malhotra op-ed is also wrong.

May I humbly suggest that the media start asking questions of these people and that peer reviewers start doing their job?

No comments: