Hi. I have a coxph model of variables linked to survival in medulloblastoma. The data were collated from various publications and not all authors quoted all variables. I'd like to internally validate the model and have tried bootstrapping it using boot.stepAIC but it fails because of the NA values. If I remove all samples with any missing values my dataset falls from n=227 to n=65 and, not- surprisingly, only the most glaringly obvious variables remain in the final model. Is there a way to get the boot-strapping and variable- picking process to include samples with NA values, even though it's not statistically the most correct approach? Martin