Wednesday, March 9, 2011

When science gets it right: a smokin' prediction

Science, as practiced by scientists, has lots of flaws and fallibilities.  Methods and inertia and vested interests sometimes drive what's done and how it's done.  When inappropriate designs or methods are used to answer a question, or when an idea (or belief) is so strong that it can hardly be falsified by scientific evidence, science deserves criticism.

But when science gets it right and for the right reasons, this should be recognized as demonstrating that causation does actually occur in this world and can be identified when the situation is clear enough, by the methods we know how to use.  Often, success comes when a single cause is strong on its own, and predictive of an effect.

There have been decades of very good evidence that smoking causes lung cancer.  One can predict that a certain amount of smoking should lead to a certain amount of cancer.  It's not precise, but it's clear, and shows, at least statistically (since not even most smokers get lung cancer), that smoking is a causative agent.  Given what we knew of male smokers and cancer rates decades ago, information gathered when most smokers were males, it was predictable that when women started thinking that a smoke was cool they'd start joining their men friends in the cancer wards.

Women began smoking in large numbers around 25-50 or more years ago, and a new study demonstrates that it's catching up to them.  Also reinforcing the causal connection, men had quit smoking in large numbers at about the same time in the past, and their rates of lung cancer have been declining as would have been--as was--predicted.

Lung cancer rates have more than doubled for women over 60 since the mid-1970s, figures show.
Cancer Research UK figures say the rate rose from 88 per 100,000 in 1975 to 190 per 100,000 in 2008, the latest year for which statistics are available.
Lung cancers in men fell, and CRUK say this is linked to smoking rates.
The proportion of male smokers peaked before 1960. But women had rising rates in the 1960s and 1970s, which would have an effect on those now over 60.
Overall, the number of women diagnosed with lung cancer has risen from around 7,800 cases in 1975 to more than 17,500 in 2008.
Figures for men went from 23,400 over-60s diagnosed in 1975, falling to 19,400 in 2008, with rates showing a similar large drop.

Strong evidence, to go with laboratory and molecular/biochemical evidence about the nasty ingredients in smoke and what it does to DNA to transform nice, pink healthy lung cells to charred, ugly cancerous one (anybody who's taken a gross anatomy class in a medical school has probably seen the coal-bag lungs of cadavers of former smokers).

Famous people, most notoriously RA Fisher, one of the founders of modern statistics, have tried to find reasons why this association was due to confounding--some true cause other than Virginia's finest, but that was correlated with smoking. But the evidence has piled up the other way (despite the effects of other exposures).

Other predictions
So this is prediction about the future made from past observations.  But what about the other kind of prediction?  A scientific theory can be really convincing if it can make some additional predictions that would be a consequence of the hypothesis.  So, what if you go not to smokers and non-smokers and follow their exposure rates, but go to lung cancer wards and ask whether the patients were smokers? You'd expect to find that most of them were, and that is what the evidence shows.  Even with twists, such as in Utah, where the population is heavily Mormon.  Mormons don't believe in smoking, but the cancer wards in Utah suggest that Mormon lung cancer patients had apparently not adhered to their religion's teaching.

Understandably, attention is on the gruesome outcome of lung cancer.  But we can make another prediction, and we guess some of the data are probably already in hand.  In many studies, perhaps largely of men since they were the main smokers, a high fraction of smoking-attributed death and disease was not due to lung cancer, but involved many other systems--heart attack, emphysema, and many others.  Lung cancer is only a minority, perhaps a small minority of these consequences.  So we can predict that these traits have diminished in men (we think they have), but should be increasing, along with lung cancer, in women.  If that turns out not to be the case, then we have to revisit much that we think we know about smoking.

Given both the prospective prediction and retrospective assessment, our ideas about cause and effect receive strong, persuasive scientific support.  No weakling GWAS evidence here!  Yet, why given this strong and clear support for smoking as a sledge-hammer kind of risk factor, do so many people--even college students who learn about these facts in a reasonably rigorous way, still smoke?  It raises questions about the efficacy of education, about understanding of statistics and risk, and of the impact (or not) of scientific knowledge.

Because today, only the tobacco industry would still claim that smoking was just plain innocent fun.

No comments: