![Russia's new stealth fighter to challenge for supremacy of the skies [video]](http://i2.zdnetstatic.com/blogs/thumbnail/_jkqSU.jpg)
Dana's common sense approach to why the study is flawed is, well
for lack of a better word, flawed. The the whole idea behind a
prospective cohort study, a study where a researchers follow a
group of people, keeping track of exposures (alcohol intake,
vitamin E etc) and outcomes (CVD, obesity etc) is that you can
track multiple exposures and multiple outcomes. This holds true
even if there is a primary exposure-outcome relationship that the
study was set up to look at, and you are looking at something different. As long as the data collected matches the data you're analyzing, the study has some validity. So, if we look at Wang et
al's study, we see they use this design. The fact that the data are
old is a little bit misleading, the study started in the 1990's but
continues to this day. The data that Wang et al used stopped being
collected in 2004. So we must ask ourselves, have women's behavioral patterns when it comes to food, alcohol and weight
changed in the last 6 years?
So let us look at the hypothesis that the amount people reported
drinking is wrong completely wrong? The answer is yes and no. The
amounts reported are probably different than those consumed, but it
would only matter if women who were gaining weight under-reported
more than those who's weight was stable. And since the participants
didn't know what the goal of the study was, there would be no
indication to the women gaining weight to under-report. Might they
have anyway? Yes. That leads us to to two issues; one, welcome to
observational research, there are no guarantees that your data are
good. And two, of those women gaining weight, the vast majority of
them would have to had underreported to get the reduction in risk
of obesity reported (up to a 2.5 fold decrease).
Finally, Dana's push for electronic records. He states that in the
future "we will be able to look at data on millions of women and
parse it carefully so we get their actual intake," and I hope that
this is true. But it will only work if health care providers ask
and record the data similarly. Would I take the millions data
points from electronic systems over this study? Probably not, as
the data would come from thousands of different sources with
different methods. This study used a standardized questionnaire.
So should we accept Wang et al's study carte blanche? No. But
instead of looking at old data, and complex study design and saying
that it is just bad because common sense tells us so, we should ask
one simple question. Is the relationship between alcohol and weight
gain different for 19,000 middle aged, normal weight, female nurses
different from the general population of women in America?
The media is filled today with the glorious news that women who drink don’t get fat. Well they don’t get as fat as their non-drinking friends.
But before you run to the bar, let’s look at how this conclusion came about. (Bottles of wine from CBS.)
The study, from the Archives of Internal Medicine, is a re-examination of data from the Womens’ Health Study, which ended last year. The goal of that study had been to look at low-dose aspirin and Vitamin E in preventing heart attacks.
What Lu Wang and her colleagues at Brigham and Womens Hospital did was look at two fields in that database, self-reported alcohol use and weight gain over the course of the study.
They concluded that, while men tend to pile alcohol on top of their regular diets, women may use a drink in the evening in lieu of food. This limits their food intake and keeps the pounds from adding up.
Well, maybe. But this is a study of about 19,000 women. And this is self-reported alcohol intake.
I may estimate how much I drink to a doctor, but I don’t count the drinks each night and give an accurate total. I suspect most people don’t.
That’s the limit of all these studies which take old data and seek new interpretations from it. You’re taking something made to prove X and getting it to prove Y. You’re taking a sample that may be valid in size for one purpose and turning it into something else.
It’s not just true here. I don’t want to be seen as looking down at Dr. Wang. This is how we have to do things right now. Our data stores are limited to those created for specific purposes in the past. It reminds me of using a chemical process to mine old gold tailings.
There are two points in the health IT revolution. One point is to avoid mistakes and treat you more efficiently. The other is to collect data, vast mines of data, for studies like these.
Some time in the not-too-distant future, we will be able to look at data on millions of women and parse it carefully so we get their actual intake, not an estimate, and a firm result on this question.
But that time is not now. Even though my lovely wife does enjoy a glass of red with dinner. I just see it makes her happy. She’ll always be a size two to me.
