Confounding

I know what you're thinking. You're thinking that this post has something to do with the logic behind anti-vaccine groups. Confounding doesn't even begin to describe what they're up to. No, this post is about a concept in research studies and epidemiology called "confounding". Simply defined, confounding is what happens when you can't see the forest for the trees... When you assume that A and B are related without knowing that A and C are related and C and B are related, but A and B are not really related. Confusing? Confounding? Let's try to clarify it after the jump...

It's not the fall that kills you. It's the sudden stop.

COFFEE AND CANCER
The best example of confounding I can remember from school is from a study back in the day that concluded that coffee was linked to pancreatic cancer. From March, 1981, in the New York Times:

A statistical link between the drinking of coffee and cancer of the pancreas, the fourth most common cause of cancer deaths among Americans, was reported yesterday by scientists of the Harvard School of Public Health. The discovery was unexpected, and its significance is not yet clear. ''If it reflects a causal relation between coffee drinking and pancreatic cancer,'' the report said, ''coffee use might account for a substantial proportion of the cases of this disease in the United States.'' The findings need to be pursued with further research, said the report in today's issue of the New England Journal of Medicine.

If you read the story, you'll find out that even the researcher quit drinking coffee when he saw the results. Of course, now in 2011, we still consume coffee at a really good clip. What happened? Well, let's recap: Coffee (A) was statistically linked with a higher risk of developing pancreatic cancer (B).

WHAT ABOUT "C"?
Glad you asked. As it turns out, someone figured out that something was up with that association between coffee and cancer. Armed with the knowledge that cigarette smoking is associated with other cancers, someone decided to stratify the results. What's stratifying? It's a technique used to control for confounding (and to look at different factors individual of each other). Stratifying is done by splitting up the study subjects by a factor like gender, age, etc. In this case, it was done by smoking vs. non-smoking. (Smoking, in this case, is "C".) Guess what happened?

SURPRISE!
It turned out that non-smoking coffee drinkers were no more likely to get pancreatic cancer than non-smoking non-coffee drinkers. On the other hand, smoking coffee drinkers were more likely to get pancreatic cancer than non-smoking coffee drinkers and non-smoking non-coffee drinkers. Finally, smoking coffee drinkers were just as likely (within a margin of error) to get pancreatic cancer than smoking non-coffee drinkers. So was it the coffee?

Nope. It was the smoking.

RECAP
Confounding is a quirk of the analysis of studies when you fail to understand the relationship(s) between one factor and a series of other factors. It may very well not be your fault. Just like getting lost in a forest and walking in circles without knowing that a road is nearby, you can totally overlook some factor that is confounding your findings. In the example above, researchers didn't see smoking as something to take much into account, for some reason. Eventually somebody did, and baristas everywhere breathed a sigh of relief.

"It's all about the Franklins and the Benjamins!"

In other words, coffee seemed to be associated with pancreatic cancer until you took smoking into account. Once you did that, you could write off coffee and focus on smoking.

CONTROL?
How do you control for confounding? It's hard to do until the analysis portion of the study because you randomize people into groups. If you try to control for confounders at the beginning, you may very well bias your study in other ways. It's a total "damned if you do, damned if you don't" type of situation. In the example above, knowledge of the biology and epidemiology of cancer pretty much saved the day... Even if it's now a story that will be told countless times by epidemiology teachers to their students.

NEXT TIME
I'm somewhat excited about the next lesson, but it's not for the right reasons. It will be an intro to outbreak investigations. I'm excited because there is a certain medical doctor and pediatrician to the stars who thinks that the current outbreak of measles in Minnesota is not a big deal. He's even joked about it on Twitter. While the outbreak post won't be a personal attack on him or his abilities, I'll use him as a prime example on why public health officials and workers have such a hard time containing outbreaks... Hint: it's ignorance. So I'm excited to bring some clarity to what is and what isn't an outbreak, and what we stay up late into many nights to do about them.

Thank you for your time.

BlogThis!