Using data mining to improve clinical decision support

In this paper we report that it is important to train clinical decision support systems on as large a data set as possible. We describe how the diagnosis of heart disease, from 12-lead ECGs, may be improved by up to 32%, with respect to diagnosis using a small data set, by the use of data mining algorithms trained on a sufficiently large data set. We also describe how these suitably trained and chosen, data mining algorithms appear to be capable of provision of clinical decision support by virtue of the short time, 1-3 seconds, required to perform patient state classification.