Does anyone understand the protocol set forth in Sauerbrei's and Royston's Multivariable Modeling book on fitting interactions in fractional polynomials? The first step on p. 156 says to apply the FP algorithm to the predictors. Clear enough but in what cases does one include the continuous covariate and when is it excluded? Does the second step merely include refitting the model developed in the first step this time including continuous covariate? Why in step 5 would one refit a model already developed in step 2? Clearly I'm missing something here. _________________________________________________________________ owitworks_022009 [[alternative HTML version deleted]]