Tuesday, October 11, 2011

New Study: Biomarkers are Cool, but Nearly Useless for Predicting Alzheimer's

Everybody would like to find a brain scan or a blood or spinal fluid test to predict Alzheimer's disease. Such objective tests seem inherently more reliable than the clinical interview.

Recently we've seen various studies identifying biomarkers for predicting who will develop Alzheimer's. The problem is that most of the studies are merely suggestive, though they are so complex that they create a veneer of definitiveness. The usual technique is a sort of statistical fishing expedition. First, you identify a group of people who are at risk of developing Alzheimer's. Second, you collect boatloads of data, such as from brain scans, blood draws, or spinal fluid specimens. Third, wait a couple of years to see which patients develop Alzheimer's. Finally, have a computer sort through dozens--sometimes hundreds--of possible biomarkers until you find one, or a combination  of several, that correlate with developing Alzheimer's.

There are many variations on this theme, of course, but this is basically how biomarker research is done. The problem is that the more biomarkers you test, the higher the risk that you'll find a correlation that isn't a valid biomarker--but is just a random finding that has little to do with Alzheimer's.  In order to test the utility of biomarkers, you have to choose a few likely candidates, and see if they actually predict Alzheimer's...and you have to compare the predictive power of the expensive biomarkers with the cheap old fashioned method--the clinical interview.  Amazingly, almost no such studies have been done--and not a single study has compared multiple types of biomarkers with the clinical interview. That is, until now. 

An article published in the latest issue of the Archives of General Psychiatry compared specific neuroimaging findings and spinal fluid markers with a clinical interview. Over a two period, the researchers found that only three variables were able to significantly predict which patients would develop Alzheimer's: two clinical measures of memory, and one neuromaging finding--the thickness of a part of the temporal lobe. By far the most important predictor of cognitive decline was a combination of a brief survey of activities of daily living and 5 minute paper and pencil test. These two combined explained 50% of the variance in outcomes, which is a very large number in these kinds of studies.

The bottom line is that if you want a test to predict whether a patient is going to progress from mild cognitive impairment to Alzheimer's, stay away from the needle.  A careful clinical interview, repeated over time, is overwhelmingly more effective than these expensive tests.

5 comments:

Focus Groups said...

I agree, these random samples give results differently.

John M. Nardo MD said...

Great point. We're about to be bombarded with biomarkers from iSPOT and EMBARC - the studies looking for antidepressant response predicters. I predict your point here will carry the day there too - much ado about nothing...

Anonymous said...

The type of investigation you describe is the mathematical equivalent of astrology.

And like astrology it will generate a lot of money-making scams.

JLH - general practice said...

This is why I participate in, and even pay into, Dr Carlat's activities. A service to the field of mental health.

Sean Paul said...

I agree with this post partially. If we can correlate the findings with clinical course and use medications to treat preemptively, could we not slow the progression?