I was just doing some personal griping about glass ceilings for people like me without a formal education, when I noticed that the top of the next page had a University of Phoenix ad on it. I've done enough natural language processing, analytics, and machine learning to understand how little was really needed to target me for that ad. For the algorithm to work, it needs to do the following:
1. Identify me as a good candidate for the ad based on a few key words,
2. Offer it to me (and others like me a few thousand times), in a way that is attractive but not offensive,
3. Deal with the fact that it helps maybe 1 in 100,000 people who see it.
One of the "problems" of CDS is identifying which clinical decision support algorithms are applicable, and when to use them.
The same set of techniques could also be used in an EHR, so long as they are equally unobtrusive.