The object of this chapter is to summarise some key aspects of the design of prognostic factor studies. These studies are usually based on regression models that help determine which of the (usually many) candidate variables, often in combination, are truly prognostic for outcome. We point out that choice of a model depends on subjective judgement, ranging from the choice of the variables to consider to the precise statistical methods used in the variable selection process. Consequently, emphasis is placed on the need to first develop a model with an index group of patients which is then validated in an entirely separate group of patients. We use a survival time endpoint to illustrate the methodology, although the same issues relate to other types of endpoint variables.
Was this article helpful?