Ultra-low risk epidemiology
A new meta-analysis highlights what junk science it is
It is safe to assume that the number of epidemiologists in the world greatly exceeds the number of risk factors that are left for them to discover. Every year, thousands more pour out of the universities and they all have to find something to do. The easiest - although most boring - thing for them to do is carry out a meta-analysis, ploughing through the garbage left behind by earlier epidemiologists. One such meta-analysis was published this week by ten Chinese academics. It is titled ‘Maternal smoking, consumption of alcohol, and caffeinated beverages during pregnancy and the risk of childhood brain tumors: a meta-analysis of observational studies’.
Is there any particular reason to think that the use of tobacco, alcohol or caffeine during pregnancy increases the risk of childhood brain tumours (CBTs)? Not really, but that is not the epidemiologist’s problem. They are only interested in correlations and, as the authors of this meta-analysis perceptively note…
Smoking, alcohol consumption, and consumption of caffeinated beverages have become common lifestyles for people.
No flies on them!
It must have come as some disappointment when the only statistically significant association with childhood brain tumours in the meta-analysis turned out to be with caffeine (OR 1.16, 95% CI 1.07–1.26). Caffeine is the only one of the three substances that is not currently in the public health doghouse, so the authors don’t mention this finding until the end of the abstract. Instead they lead with what they call ‘smoking’.
The results suggested a borderline statistically significant increased risk of CBTs associated with maternal smoking during pregnancy (OR 1.04, 95% CI 0.99–1.09).
Another way of saying ‘borderline statistically significant’ is ‘not statistically significant’. Not only does maternal smoking not have a statistically significant relationship with CBT, but what they are measuring is not ‘smoking’ as any normal person would understand the word. For reasons best known to themselves, the combined the results for smoking and passive smoking. (Describing exposure to ambient tobacco smoke as passive smoking was a brilliant PR move by the anti-smoking lobby back in the day as it implied that inhaling 1/1,000th or 1/10,000th as much smoke as a smoker was tantamount to being a smoker. This is where it has led us.)
When the results are broken down they become rather interesting:
We found that passive smoking (OR 1.12, 95% CI 1.03–1.20), rather than active smoking (OR 1.00, 95% CI 0.93–1.07), led to an increased risk of CBTs.
The dose makes the poison. If the risk doesn’t increase when exposure to the risk factor increases, it’s probably not a risk factor. If heavy exposure does not increase the risk while minimal exposure does, it is not science. It is homeopathy. The authors write that “it cannot be ruled out that women might have a higher tolerance for active smoking”, but I’m going to go ahead and rule that out.
In a saner world, no one would be the slightest bit interested in a 12% increase in risk for a rare health condition and no one would take relative risks of less than 1.25 (i.e. a 25% increase) from observational epidemiological seriously for anything, especially when there is no obvious biological mechanism. The only useful purpose studies like this serve is to highlight how useless ultra-low risk observational epidemiology is.
This is not the first time that the purported risks of passive smoking have appeared to be as strong or stronger than the risks of smoking. It has happened with breast cancer studies, for example (despite decades of trying, epidemiologists have struggled to find a link between smoking and breast cancer). The fact of the matter is that chance, residual confounding and (if smoking is a genuine risk factor) undisclosed smoking can easily produce spurious relative risks in the region of 1.05-1.30 for passive smoking. Studies like this inadvertently demonstrate that. I would be surprised if the association with caffeine isn’t equally spurious.
Alternatively, you can believe that children are more likely to get a brain tumour if their mother drinks coffee and their father smokes than if their mother smokes and drinks alcohol but doesn’t drink coffee.
As far as I can see they also didn’t test for a difference between passive and active smoking - just assumed they were different because one result was significant and the other was non significant. Despite having overlapping confidence intervals.
I can do you a nice epidemiological meta analysis in 10 hours. People who live in places with single payer health systems are fatter than those who live in universal health systems based on co-payments by the patient or SHI. And that imposes a cost on the societies with single payer systems in terms of lack of incentives and responsible behaviours.
Furthermore every country with a single payer system has dropped down world league tables of prosperity, but correlation ain't causation so maybe there's another underlying cause like say collectivism as a government philosophy.