Tuesday, 12 January 2010

Unable to escape their initial assumptions

Via a comment at Shapely Prose (in a piece well worth reading itself) I was directed to BMI and Mortality: Results From a National Longitudinal Study of Canadian Adults (full text, for once, public!), yet another study of the correlation between BMI and lifespan, similar to the US and German studies previously mentioned.

It largely repeats the results of the previous studies: a BMI below 18.5 or above 35 is associated with a statistically significant (though small, especially for "above 35") increase in risk of death compared with a BMI in the 18.5-25 range, and BMIs between 25 and 30 are associated with a statistically significant (though again small) decrease in risk of death.

Let's take as read for now, because it's been covered many times before, that:

  • most importantly, there is no moral value associated with healthiness or having a particular BMI anyway. While governments might understandably want a longer-lived and healthier population, because it makes their jobs easier, and while many individuals understandably might themselves like to live long and healthy lives if possible, that doesn't mean that there's any requirement on individuals to be healthy - it's often out of their control, and even when it is within their control, there may be other things they think are better uses of their time, energy and money;
  • BMI as a usable statistical population measure does not make it a useful individual measure for a whole range of reasons;
  • there is no known way to reliably effect a long-term change in weight;
  • correlation is not causation: just because people with a particular BMI are more likely to die in a given time period doesn't mean that this is caused by being in that BMI range, nor does it imply that magically increasing or decreasing their weight would make them live any longer.

The way the paper is written is quite interesting, and indicates that the authors have a deep belief that fat is bad that they're trying to preserve in the face of the evidence.

Firstly, there's the way that they refer to the 18.5-25 BMI range. This may just be standard terminology for epidemiologists in Canada, but referring to it as the "acceptable" range is unusual. It's what "normal", it's more common name, means, of course, but it's unusual to see someone come right out and say it.

This is an important public health message, because while overweight may not be a risk factor for mortality, becoming overweight is a necessary step between being of acceptable weight and becoming obese.

This is key evidence of their assumptions. There's an initial assumption that being "obese" (categories 30-35 and 35+) is unhealthy. The relative risk of the 30-35 category is, within the bounds of statistical significance, exactly the same as the "acceptable" 18.5-25 category. The 25-30 category has less risk than either (though this is barely significant).

One could therefore equally phrase this in the opposite direction. Let's take a similar sentence with the same risk pattern to show just how absurd the conclusion is.

This is an important public health message, because while being hydrated may not be a risk factor for mortality, becoming hydrated is a necessary step between being acceptably dehydrated and dying of water poisoning.

The sentence only makes any sort of sense due to the power of "acceptable BMI" as a default, but even with that, it's still nonsensical: "we won't encourage behaviour that might be healthy, because if people do it too much, it's unhealthy again" (mind you, the same logic, if applied consistently, would also discourage attempts to lose weight, so maybe it's not all bad.)

Another bit that makes a similar lack of sense is this bit. As background, if you didn't read the entire paper yourself: they took people who reported their height and weight in 1994/5, and checked if they were still alive in 2006/7. They then used their 1994/5 BMI, among other factors, to determine the risk factors. This quote follows immediately on from the last one about public health:

Other analyses using the National Population Health Survey data demonstrated that almost a quarter of Canadians who had been overweight in 1994/1995 had become obese by 2002/2003 and Canadian adults within all BMI categories continue to gain weight.

They present this as a bad thing, of course, but of course they're not following through the logic enough. These people have gained weight relative to their starting weight, often by a sufficient margin to move them into a different category. They haven't died. This is not evidence that any of these are dangerous except for the prior assumption - which the data isn't supporting - that being "obese" is dangerous.

If the sample are generally gaining weight (and we'll assume, since it's a sample of adults, that most are not significantly changing height over the period of the study), that suggests that this effect is already accounted for in the relative risk. Looking at the age breakdown in Table 2 suggests this further: the relative risk associated with less than 18.5 BMI rises, and the relative risk associated with above 25 BMI generally falls, for the older part of the sample, which is what would be expected if gaining weight over time was common (though only one of these figures differs from the reference category by enough to be statistically significant).

It's worth noting, of course, from Table 1, that the people over 60 in Table 2 have a relative risk factor due to their age of between 10 and 100 compared with those under 60, which is not surprising, and would completely swamp any of the BMI risk factors - which as has already been said, are either barely significant or statistically insignificant - for any age band.

So we have a study that shows being "overweight" appears not to be harmful, and being "obese" largely appears not to be harmful, and a group of researchers trying to suggest other ways it might be harmful because they can't believe that the "acceptable" weight band could be anything other than the "best". It's a common pattern in weight-related research.

The power of the default and its assumptions are strong indeed, and so despite an increasing volume of studies showing that weight has virtually no effect on health, many of the major public health programs are trying to combat an illusory problem, with no effect other than to demonise a significant part of the population.