Creating a Nomogram with Multiply Imputed Datasets


My team is hoping to create a nomogram with multiply imputed data using the nomogram function from the rms package.

We can create a full nomogram with a single dataset, and we can create a partial nomogram from the imputed dataset by pooling results using fit.mult.impute from the Hmisc package. However, we can’t figure out how to create a pooled survival curve for the multiply imputed data that we can use for our nomogram.

Looking around online, I wasn’t able to find any solid answers, but I found some emails with Dr. Harrell from 2013 that seemed to indicate that there wasn’t a set method for how to combine results across survival curves for each imputed dataset. Given that that post is more than 10 years old at this point though, I wondered if there has been any movement on that front that I haven’t been able to find.

So, I wanted to ask if there is a way to create a nomogram from multiply imputed data in the rms package, or, if not, if there is a way to compute a survival curve from multiply imputed data either with or without rms and use that survival curve to complete the nomogram created using fit.mult.impute.

Thank you for your time,

It’s still a good question. While you can average survival curve estimates, stacking of all the imputed datasets and doing one large model fit, ignoring the falsely low standard errors and p-values, will probably work best. I do some other types of analysis with this approach in Regression Modeling Strategies - 24  Bacteremia: Case Study in Nonlinear Data Reduction with Imputation .