Identifying prognostic factors in binary outcome data: An application using liver function tests and age to predict liver metastases

Publisher Website
Google Scholar
Abstract
A frequent issue confronting the medical scientist is the detection of important prognostic variables, or covariates, that affect the outcome variable. In this paper we propose specific guidelines for the analysis of binary outcome data based on recent developments in statistical methodology and on less formal graphical techniques. By incorporating these methods into the process of fitting the standard logistic regression model, one can assess covariate selection and model specification more thoroughly and obtain a more balanced view of a model's predictive capabilities. We apply the methods to the evaluation of non-invasive liver function tests and age to predict the presence of liver metastases in patients with small cell lung cancer.