Abstract:
|
The "proportion explained" is a concept that has been proposed to validate surrogate endpoints for clinical trials. We apply the concept to a statistical analysis of the factors underlying length of stay in hospital after coronary artery bypass grafting. Factors are divided into three groups according to the time of their observation: pre-surgery, during surgery, and post-surgery. We assume that a factor in one group can affect a factor in a later group but not vice versa, and divide the total effect of each factor into its direct and indirect parts. The "proportion explained" is the ratio of the indirect effect to the total effect. We do the same for groups of factors, expanding the definition of the proportion explained to the case of several variables. Variances of the estimates are obtained using the bootstrap method. We present tables showing the factors that most affect the length of stay, how much of the effect is direct, and how much indirect. The advantages and pitfalls of the method are discussed
|