Answers are grounded in this paper and cite their sources.
Coronary heart disease (CHD) remains the major cause of premature death in developed countries. Many of the risk factors for CHD are well known and interventions exist for some of them, e.g. statins for reducing cholesterol and drugs for hypertension. In the paper by Reinhard Voss and colleagues, a new approach for predicting those at high absolute risk (>20%) of a coronary event in the next 10 years using neural network models is compared to a method using a standard logistic regression model.1 The success of the multi-layer perceptron neural network (MLP-NN) is quite remarkable: 74.5% of coronary events were predicted compared to 45.8% by the logistic regression model. Furthermore, this was achieved by predicting a smaller number of men to be at high risk (7.9%) compared to the logistic regression model (8.4%) so that less men would be eligible for treatment overall, but there would be a substantial increase in the number of coronary events potentially preventable by treatment from 15% with the logistic regression model to 25% with the MLP-NN model. This commentary looks at two issues: firstly how might prediction of CHD be further improved using neural networks, and secondly how transportable are these models and can they be used as clinical tools?
No takes yet. Share an insight, caveat, or question.
MARGARET T. MAY (2002) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: