Data-Mining a Diagnosis in the ICU
Data-mining has found its way into hospital Intensive Care Units (ICUs):
- Predictive Medical Technologies is claiming their ICU monitoring software can predict adverse events like a cardiac arrest up to 24 hours before they happen.(See Algorithms are the new medical tests.)
- IBM and the University of Ontario Institute of Technology (UOIT) are researching the data-mining of ICU monitoring data to detect changes in the condition of critically-ill premature infants. (See First-of-a-Kind Technology to Help Doctors Care for Premature Babies.)
I confess unease upon reflecting on Predictive Medical Technologies monitoring software, as the phrase "secret sauce" leads me to think of trade secrets – and trade secrets are in opposition to the tradition of open research. The only defense I can come up with is that their "secret sauce" might be in how quickly they process the information (as a process that takes 2 weeks to process 8 hours of results would be useless except as a proof-of-concept). On the other hand, it sounds like the IBM/UOIT collaboration should result in standard, open research results.
Pretty cool stuff all around, though.
Perl/CPAN user since 1992.
Coming from a somewhat related field (computational linguistics) I think in practice it'll turn out to be mostly a non-issue. Inference is a well studied subfield of statistics, and both computational linguistics and computational biologists mostly spend their time stealing methods from the statisticians. And usually the best methods are the simple ones, like logistic regression and Markov chains.
Selecting clever features can sometimes be good secrets to keep, but they do tend to be fairly transparent.