[Next] [Up] [Previous]
Next: Modelling the risk
Up: A guide to the
Previous: A guide to the
The QAMC project began in 1995 with the primary aim being to explore
artificial neural networks for obstetric risk prediction. This
investigation required neural networks to be meaningfully compared to
alternative risk models and a detailed description of this research is
given in [6,7]. The initial motivation
behind the project was
- to try to develop systems that could provide clinicians early
warning of cases with a high risk of having specific adverse
pregnancy outcomes
(APO).
- to see whether useful models of risk could be constructed from
large databases of routinely collected perinatal information;
- to address theoretical issues, such as how to select
good predictors of outcome using the information contained in large
medical databases.
As the project evolved, it became clear that there was an additional
reason for doing this research: to report on the factors which
currently restrict the usefulness of medical risk prediction systems.
As we shall show, these factors range from difficulties in data
collection and problem specification, through to medico-legal issues
and interpretability of results.
Before we discuss these matters, we shall describe the QAMC project's
risk prediction web page in detail. The page provides access to the
observed and estimated incidence of failure to progress in
labour
within a portion of the
SMR2 database.
D.R. Lovell
Mon Sep 15 18:08:31 BST 1997