Statistical Methods for the Analysis of Prognostic Factor Studies
Lisa M. McShane, Richard M Simon · TNM Online · 2003
Abstract Prognostic factor studies in cancer relate covariates describing clinical features and tumor characteristics to clinical endpoints such as therapy response, disease recurrence or progression, or survival. Examples of covariates in cancer prognostic factor studies include tumor size, nodal status, presence of metastases, and tumor‐specific measurements such as expression levels of certain proteins, presence of chromosomal aberrations, or gene mutations. Typically, the relationship between the covariates and the clinical endpoints is examined by development and testing of a prognostic model. The modeling approaches discussed in this chapter include logistic regression for binary endpoints, Cox proportional hazards regression for survival outcomes, and neural networks and recursive partitioning for both binary and survival outcomes. In the final section, techniques for model validation are discussed.