Chapter 16 - The Development and Execution of Medical Prediction Models

Abstract

Multivariable regression models are firmly established as the standard method in medical literature for obtaining adjusted estimates and adjusted tests of association. Prediction of individual patient outcome is a different area than testing for associations. The purpose of this chapter is to present the strategy developed by others that is used for building and implementing several regression-based prediction models. One of the software products is the most common prognostic tools in cancer for the personal digital assistant, according to a survey by the American Society of Clinical Oncology. One of the specific messages conveyed here is that building prediction models require a different strategy even though the statistical models are the same. This chapter emphasizes on achieving high predictive accuracy as the goal rather than building a model with statistically significant factors. It also reviews the importance of model validation and calibration as indispensable steps before finalizing a predictive model.It focuses on simple and effective communication of the results. The tool used for this purpose is the nomogram. This graphical depiction of a multivariable model has been used for a long time but not as widely as one might expect, given its advantages.