DESIGN, CONSTRUCTION AND EVALUATION OF SYSTEMS TO PREDICT RISK IN OBSTETRICS

We present a systematic, practical approach to developing risk predict ion systems, suitable for use with large databases of medical informat ion. An important part of this approach is a novel feature selection a lgorithm which uses the area under the receiver operating characterist ic (ROC) curve to measure the expected discriminative power of differe nt sets of predictor variables. We describe this algorithm and use it to select variables to predict risk of a specific adverse pregnancy ou tcome: failure to progress in labour. Neural network, logistic regress ion and hierarchical Bayesian risk prediction models are constructed, all of which achieve close to the limit of performance attainable on t his prediction task. We show that better prediction performance requir es more discriminative clinical information rather than improved model ling techniques. It is also shown that better diagnostic criteria in c linical records would greatly assist the development of systems to pre dict risk in pregnancy. (C) 1997 Elsevier Science Ireland Ltd.