A few interesting points
The results - not surprisingly, by now - confirm the previous US, German and Canadian population studies. There is no statistically significant correlation between mortality and BMI - a person with a BMI of 19 and a person with a BMI over 40 have virtually identical chances of dying in any year. (Chapter 6)
There does seem to be some difference in causes of death - respiratory conditions are less common causes of death in "obese" people, and endocrine and metabolic conditions are more common - but as the German study pointed out - that just means we have a better idea of what statistically kills fat people. It doesn't make thin people any less dead.
Health service usage
The study also includes a very interesting look at health service usage by BMI in Chapter 5. Fat people do use more health services than thin people - to a statistically significant degree for some but not all services - but generally not by a very large margin. For instance, for visiting GPs, were the campaigns to get men to visit the doctor to succeed, bringing male GP visit rates up to the same as female GP visit rates, this would increase visits considerably more than if everyone visited the GP as often as a government-approved weight person of their gender.
Comments on the Big Fat Blog piece point out some reasons why correlation may well not be direct causation here - misdiagnosis of fat people necessitating repeat visits and weight gain caused as a side effect of medical treatment - and the German study also notes underdiagnosis in thin people of conditions popularly associated with obesity as another possibility.
The study doesn't cover health care costs directly - but they did do a brief literature review. Emphasis mine.
Studies of health care costs (not analysed in this report) show significant positive associations with BMI level (Andreyeva et al., 2004; Borg et al., 2005; Raebel et al., 2004; Thompson et al., 2001). This is consistent with the higher use of some healthcare services as shown in this report, though the differences appeared to be higher in those studies than results from this study would suggest. Some of this may be due to differences in the context and costs of healthcare, as most of these studies were done in the United States.
Or, paraphrased, if your health system is massively inefficient, expensive to use, and encourages people to only seek treatment in emergencies, you might well have very distorted costs. That doesn't mean that countries with more sensible health provision will have comparable costs.
Unlike many previous studies, this one recognises the arbitrariness of the BMI measures. After describing the World Health Organisation categories:
These categories were created by examining relationships between BMI, mortality, and morbidity. However, the cutoffs remain somewhat arbitrary and have changed over time: a BMI of 27 used to separate the low risk from the high risk categories [...]. Furthermore, these groupings may not be equally useful for older adults [...] or different ethnic groups, including Aboriginal peoples [...]. Therefore, analyses in this study used continuous BMI values whenever possible, with summarized data for the standard groups shown above as well.
I've mentioned the "step change" attitude that public health seems to take - it's good to see a study that explicitly takes the opposite approach, and doesn't take the WHO-set boundaries as definitionally accurate.
They also acknowledge many problems with BMI as a measure of weight categories (including assumptions of white European as default) - but note that they need to use something and
[...] it is the only measure available in existing data sources that covers a large and representative sample of the population (excluding residents of First Nations)
Changes over time
There are a couple of graphs in Chapter 2 noting that weight distributions have been basically static since 2000 in Canada (very similar to the situation in the UK). One of the things they note is that "overweight" is more common in men than women.
I suspect that a partial explanation for this may be due to flaws in the BMI measurement. BMI is based on a ratio of weight and height2. But that's not a ratio found in nature1 - scaling something up generally involves a ratio of weight to height2.5, because of the way circulatory and nervous systems work at small scales.
So the simplification - which makes BMI possible to calculate without a scientific calculator or a log table - means that for people of the same build, BMI will increase with height.
Men tend to be taller than women, so will tend to have higher BMIs for the same build. (Also, there's not actually any reason, given that men and women have statistically different anatomies, that the BMI distributions shouldn't be different)
Studies like this leave governmental "obesity" strategy not as "correlation equals causation" but as "lack of correlation equals causation". There's no evidence that heavier weight harms life expectancy (though it may have some statistical effects on cause of death), and the effect on health care costs is likely to be trivial. It will be interesting to see if the Manitoba government, which commissioned this study, changes policy as a result.
Unfortunately the popular belief that "obesity" is both morally disgusting and voluntary2 means that research into "how obesity is bad" gets far more attention and support than research into "is obesity actually bad for health?"
1 If you can view images, there's an unintentional "uncanny valley" effect in these images, where simple rescaling of the image of a person gives a slightly unusual effect. The relative sizes of body parts aren't supposed to be identical at each size, so it gives a strange effect.
2 Even aside from the studies suggesting that weight and build are to a large part genetic, and that the "diet and exercise" mantra is so oversimplfied it's almost useless, this ignores the concept of constrained choices and so also contains a lot of classism, disablism, and other forms of discrimination.