Looking for the Signal Posted on November 15, 2007 Eric A. Youngstrom (bio) encourages a closer clinical-statistical relationship.

ROC analysis is an interesting case in point. It was developed as military technology back in World War II as a way of looking at radar and getting a sense of, “Can we tease apart a signal, like an enemy airplane, from all of the background noise?” And it’s been picked up in medicine and used to say, “Can we separate — can we look at an MRI or a X-ray image and separate the signal, a tumor, from the background noise that’s going on in the human body?”

And what’s happening now in psychiatry is people are starting to apply the technique looking for the signal, bipolar disorder, and teasing it apart from all of the background noise. And it’s like listening to popcorn pop, because the technique’s been around since the ’40s, and there are one or two examples of it getting applied in the literature in the ’70s and ’80s.

And then in the ’90s, the paper that I saw first using it was out of Boston: Chen, Faraone, Biederman, Tsuang. And I read it several times and thought about, “This is cool.” And as you look at the literature now, it’s like the popcorn is starting to pop a lot more rapidly.

It still needs to get into the training programs. We need to have psychology and psychiatry programs offering that as part of the core content. And when that happens, it will be exciting, because then people, as part of their training, will be getting exposed to the statistics that reconnect the research and the practice. And so I think that that’s going to be a very powerful feedback loop when we get there. So that all’s neat.

The other part, thinking about the statistics, that I didn’t appreciate at the time was just also how marketable a skill it is, because so many people don’t enjoy it or don’t see the connection. But I’m not advocating becoming a biostatistician, although the world needs more of them for sure.

But what I’m really advocating is becoming a clinical scientist who’s also comfortable with statistics, because it’s the rare person that can get excited about ROC analysis, but also sit with the family and do the interviews and then learn about, “You know, I’m not hearing them describe elated mood the same way. Does it look the same when I do a logistic regression? Does it look the same?”

And so that being able to connect what you’re hearing with the families with what you see in the data really only comes if you have a commitment to the clinical training and to the statistics.

« Back to Article