Combining non-parametric models with logistic regression: an application to motor vehicle injury data.

Abstract

To date, computer-intensive non-parametric modelling procedures such as classification and regression trees (CART) and multivariate adaptive regression splines (MARS) have rarely been used in the analysis of epidemiological studies. Most published studies focus on techniques such as logistic regression to summarise their results simply in the form of odds ratios. However flexible, non-parametric techniques such as CART and MARS can provide more informative and attractive models whose individual components can be displayed graphically. An application of these sophisticated techniques in the analysis of an epidemiological case-control study of injuries resulting from motor vehicle accidents has been encouraging. They have not only identified potential areas of risk largely governed by age and number of years driving experience but can also identify outlier groups and can be used as a precursor to a more detailed logistic regression analysis.