I have written an online Applied Missing Data book for all researchers that have missing data. Click on continue reading to access the book.
With the psfmi package it is possible to force variables in the model during backward selection. Be aware that backward selection may result in overfitted and optimistic prediction models, see TRIPOD. Backward selection should therefore be followed by internal validation of the model.
External validation means that the performance of a prediction model is studied in a new (external) patient dataset that is not used to develop the model. To study the performance of the model measures for discrimination (e.g. Area Under the Curve) and calibration (e.g. Calibration curve, Hosmer and Lemeshow test (H&L)) are used.
I have written a Missing Data guideline as part of the Research Quality handbook for all researchers in and outside The Amsterdam Public Health research institute (collaborative research institute between the VUMC and AMC universities and Academic Hospitals in Amsterdam). This guideline may be useful for researchers.