Tuesday, August 31, 2010

Booze and Death

Here's a writeup of a study which suggests that heavy drinkers live longer than those who have never consumed alcohol at all, even after taking into account socioeconomic status and all sorts of other possible explanations for differential death rates. The moderate drinkers were found to live longest, then the heavy drinkers, then the teetotalers.

Studies do generally find that moderate drinking is not bad for your health. But heavy drinking being better than not drinking at all?

Before we all run out to get really, really drunk, I'd like to draw your attention to one possible problem with the study:

The sample of those who were studied included individuals between ages 55 and 65 who had had any kind of outpatient care in the previous three years. The 1,824 participants were followed for 20 years. One drawback of the sample: a disproportionate number, 63%, were men. Just over 69% of the never-drinkers died during the 20 years, 60% of the heavy drinkers died and only 41% of moderate drinkers died.

Having a sample so biased towards men is pretty odd, because it suggests something nonrandom at the initial sampling stage. We'd expect a random sample from the general population to have a little more than fifty percent women.

But there's another reason to carefully scrutinize that paragraph: The initial sampling frame consisted of people who had had outpatient care in the previous three years. I haven't read the actual study so I am not quite sure what it means by "outpatient" care, but I would interpret this frame as meaning that people were selected for the study from a group which came to the attention of the health care system because of ill health or accident.

But consider the groups of teetotalers and heavy drinkers: What sort of reasons might they have to seek outpatient care, reasons which differ by the simple fact that the latter group consumes alcohol?

Perhaps heavy drinkers are more likely to enter care because of an accident? A fall or something similar? Could it be that this sampling frame will over-sample non-drinkers who are not well at the beginning of the study? Or, if you like, over-sample fairly healthy heavy drinkers who contacted the system because of a fall or something else associated with drinking itself?

These are pure speculations from my part, of course. Still, a more representative sampling frame for the study would have been good.