Saturday, August 29, 2009

Evaluating Health Canada smoking statistics

Many readers will be aware that I have little faith in the smoking related statistics postulated by the anti-smoker brigade. I believe they've been grossly exaggerated and manipulated to the point where they're totally unreliable.

Last year, two long-time members of the housing co-op in which I live were diagnosed with lung cancer. Both are in their early seventies. As might be expected, the prognosis is not good for either of them.

One is male, a never smoker who was not exposed to secondhand smoke to any significant degree. He worked most of his life in a smoke-free environment (a school) and his now grown children never smoked in the house. He took care of himself, and up until a year or so back, still rode his bicycle on a regular basis.

The other, a female, has been smoking since she was a teenager in Northern Ireland many years ago. Her sister, also a resident of the co-op and a heavy smoker, died a few years back from the same dread disease. She continues to smoke. In fact, when doctors suggested she move into a palliative care facility, to make her remaining life a little easier, she refused. “Not bloody likely,” she told them, “they'll take me fags away.”

One of these two people will become a statistic; a smoking attributable death. As a smoker, she will be stigmatized in death, as she was in life. But what about the non-smoker, the male who never smoked a cigarette in his life? Will he too become a statistic? The answer, apparently, is yes.

Health Canada determines smoking attributable deaths by calculating an SAF (Smoking Attributable Fraction) which is applied to all lung cancer deaths whether the deceased was a smoker, a former smoker or a never smoker. The last report compiled for Health Canada (based on 2002 data from StatCan) indicates that the SAF for males was 88.6%.

So, if, or perhaps more accurately, when, my neighbours succumb to lung cancer, the man who never smoked a cigarette in his life will have 88.6% of his death attributed to smoking. The lifelong smoker will have 62.5% of her death attributed to smoking (the SAF for women). At an individual level, it's an absurd proposition, since only one of these lung cancer deaths can reasonably be associated with smoking.

Using an SAF means that for every 1,000 male smokers who die of lung cancer, 886 can be attributed to smoking. Likewise, for every 1,000 former smokers who die, 886 will be attributed to smoking. And, yes, for every lung cancer death experienced by those who have never smoked, the same number, 886, will be attributed to smoking. Uh-huh. 886 out of every 1,000 lung cancer deaths among those who have never smoked will be attributed to smoking.

Of course, this means that 2, 658 (3,000 X 88.6%) of those three thousand lung cancer deaths would be attributed to smoking, even though the maximum number that could reasonably have died as a result of smoking was 2,000, the combined total of smokers and former smokers. The numbers are inflated by statistical chicanery.

First, they estimate the relative risk through epidemiological studies. The Lung Cancer Alliance claims smokers face a risk of contracting lung cancer that is 23 times higher than those who have never smoked. Former smokers carry a Relative Risk of 9 according to the Lung Cancer Alliance.

Then, using a complicated mathematical equation, they calculate a smoking attributable fraction which is then applied to all lung cancer deaths, regardless of the actual smoking status of the deceased. But, doing so raises the possibility of serious error in the calculation of smoking related deaths, as noted in the 3,000 deaths scenario. Using this kind of statistical model can lead to an inflated, and misleading, number of deaths attributed to smoking.

A statistician would likely argue that, if the relative risks estimated in the epidemiological studies are accurate and the smoking prevalence rates estimated in telephone surveys are accurate, then the statistics derived from these data should provide an accurate snapshot of the number of deaths that could be attributed to smoking population wide.

And, being neither an epidemiologist nor a statistician I would be hard pressed to argue the fine points of statistical analysis. But . . .

First off, it's not really the mathematical precision of the calculations which concerns me. But, the reality is that the smoking related statistics Health Canada postulates as fact have never been verified. They are statistical deaths, based on assumptions that may, or may not, be accurate.

For example, up until November, 2007, estimates of smoking attributable mortality were based on 1998 mortality data. SAM (Smoking Attributable Mortality) was estimated to be 47,581, or according to Health Canada, 21% of all deaths recorded in that year. In 2007, Health Canada quietly revised these figures downwards to 37, 203, or 16.6% of total deaths.

Overnight, with the simple stroke of a pen, there were over 10,000 fewer deaths attributable to smoking. But, there was no change in the number of actual deaths. The change was in the statistical deaths attributed to smoking.

The reason – faulty data.

Health Canada had been using data from CPS-II, an American study, which they had been criticized for using for many years. Said Health Canada of the modification: “Thus, direct application of a large US survey to the Canadian population may not be appropriate.“ Uh-huh.

Knowing that inappropriate figures from a US study were used to generate what turned out to be grossly inflated numbers of smoking related deaths does not exactly inspire confidence in Health Canada's statistics. And, knowing that 88.6% of lung cancer deaths among those who have never smoked will be attributed to smoking also contributes to a degree of skepticism.

But, I'll have more to say on their damnable statistics in my next post.

1 comment:

mbakken said...

WOw! This page should be mandatory reading for everyone, especially those who make laws that affect everyone!