Use of Generalized Additive Models in studies of association, prediction, and clasification

Publications in many biomedical fields have shown an interest in the application of Gen-eralized Additive Models (GAMs, Hastie and Tibshirani, 1990) since this type of modelsconstitute a good compromise between flexibility and interpretability, while avoiding thecurse of the dimensionality. GAMs including interactions can be adapted adequately inbiomedical studies of association, prediction, and classification. Based on GAMs, flexibleeffect curves such as the Odds-Ratio (for association purposes), and Receiver OperatingCharacteristic (ROC) regression curves (for the purposes of classification and prediction)can be readily obtained. The GAM-based statistical procedures may be extended appropri-ately to deal with some interesting extensions models, like the additive multi-state modelfor survival analysis. Finally, all the methods are illustrated with real data arising fromvarious biomedical fields, discussing the necessity of development of user-friendly software,to use this modern statistical methodology in practice. References[1] Hastie, T.J. and Tibshirani, R.J. , Generalized Additive Models. Chapman and Hall,(1990). ∗This work was partially supported by the Spanish Ministry of Science and Technology, Grant MTM2005-00818.