Thursday 17 March 2011

Idiotic statistics

Tobacco Control recently published an 'e-letter' (known to the rest of us as an 'e-mail') from Konstantin Krasovsky, Head of Tobacco Control Unit in Ukraine, about the WHO's measure of "selected smoking-related causes". Bear with me, because it's more interesting than it sounds.

The WHO has a handy website for accessing health and demographic data. One of things you can look up is the smoking-related mortality per 100,000 people. But, as Krasovsky points out, these estimates seem to have no basis in fact.

In the case of Ukraine, to consider just one example, the WHO HFA database shows that in 2004 male "smoking-related mortality" was 1081 per 100,000 population. However, all-cause male mortality for the same year was 1920. So 56% of all deaths were to be considered smoking-related deaths. The respective figures for women were the following: 586, 978, and 60%.

Even if we ignore the fact that female smoking prevalence in Ukraine was about four times lower than among males, it is obvious that smoking could not cause so many deaths even for men.

He's got a point. According to this database, Armenian women—of whom just 1.5% are smokers—have a smoking-related mortality rate of 138 per 100,000, which is 16.3% of all deaths.

This is not a great deal lower than the rate for Armenian men—52% of whom are smokers—who have a rate of 259 per 100,00, or 19.6% of all deaths.

In Kyrgyztan, the smoking rate for women is just 1.7%, and yet the smoking-related mortality is 607 per 100,000. Since the total female mortality rate is only 931 per 100,000, this means that a habit practised by less than 1 in 50 women is responsible for 65% of all female mortality.

For men in Kyrgyztan, the smoking rate is 41%, but smoking is responsible for 60% of male deaths (934 per 100,000 out of total mortality of 1532 per 100,000).

No one expects absolute precision from these kinds of estimates, but what we have here are figures that are so wildly implausible they don't even pass the basic test of common sense.

The WHO quietly acknowledges their uselessness on the page of definitions which almost no one will ever read:

The mortality from combined, selected causes of death which are known from literature to be related to smoking. It has to be pointed out that it is relatively rough indicator and it is NOT the estimate of tobacco-attributable mortality, which is more complex and difficult to calculate.

In other words, they have taken all deaths from any cause that is related to smoking and bundled them together even if the person didn't actually smoke. Nonsmoker dies of a heart attack? Smoking-related death. Nonsmoker dies from throat cancer? Smoking-related death. Seriously, this is what it's come to.

So remember, "smoking-related mortality" is not the same thing as "tobacco-attributable mortality". One is only vaguely related to smoking and has no basis in fact, the other is too "complex and difficult" for the World Health Organisation to bother with.

I'm sure that campaigners, journalists, broadcasters and politicians will understand that these figures are essentially worthless and will explain that carefully to the public before citing figures that might as well have been plucked out of the air. And I'm sure that no serious researcher would ever use these meaningless figures as a basis for making real-world estimates.

Yes sir, that will definitely never happen.

8 comments:

Anonymous said...

"No one expects absolute precision from these kinds of estimates, but what we have here are figures that are so wildly implausible they don't even pass the basic test of common sense."

They don't pass the test of logic, actually. In the long run, if 100% of smokers die of a smoking-related cause, the percentage of smoking-related death will equal the percentage of smokers. It won't go further.

If it does, that means at least one of those things:
- they assume that one person can die of several causes;
- they assume that non-smokers can die of smoking-related causes;
- they assume that non-smokers are punished for the sins of smokers.

View from the Solent said...

Must be from all those second-, third-, fourth-, etc-hand smoking effects.

Anonymous said...

The people making-up these numbers,as well as most other people, seem to ignor the probability of 'random co-incidence/co-occurrence.

For instance, in a group of 57 people, there is a 99% probability that two of them have the same birthday.

% of people having condition A times the % of people having condition B times the total population gives the number of people that will have both A and B due to random co-iincidence/co-occurrence.

In America, 20% of the adults are current smokers and lung cancer deaths are 0.00068% of the adult population.

.2 X .00068 = 0.00017 X 230 million adults = 31,400 adult current smokers dying from lung cancer due to nothing more than random co-incidence/co-occurrence.

Anti-tobacco claims there are about 32,800 smokers deaths from lung cancer 'caused' by their smoking.

As shown, most of those deaths are random co-incidence/co-occurrence and are not 'caused' by smoking.

Gary K.

Ken said...

So according to WHO's definition, you could have a lot of 'smoking-related deaths' in a population in which no one smokes at all?

Christopher Snowdon said...

Ken,

Oh yes!

Mark Wadsworth said...

Good find. I suppose they might as well just classify all deaths as smoking related and have done with it.

Somebody dies in a car crash? Probably trying to light a cigarette while driving. Somebody drowns? Clearly short of breath because of smoking. Murdered? Probably the burglar was a smoker.
---------------------
But I think Gary K has pushed the logic a bit too far. Let's say (for example) 20% of the population ride bikes and 1% of the UK population (60 million) has a cycling accident each year.

So there are 12 million cyclists and 600,000 cycling accidents, or one cycling accident per 100 population.

It would be incorrect to use his logic and say that:

0.2 x 0.01 = 0.002 X 60 million = 120,400 cyclists having a cycling related accident due to nothing more than random co-incidence/co-occurrence.

Mark Wadsworth said...

Should be '120,000 cyclists' obviously.

Anonymous said...

Dear Mark W.
Non-smokers do die from the diseases said to be caused by smoking; but,unless the cyclists ran into them, why would non-cyclists be involved in cycle accidents?

Gary K.