Estimating breast cancer incidence using multiple imputation with chained equations (MICE)

Breast cancer is not one disease but many different subtypes. When estimating breast cancer incidence in the population, we use routine registry data. Information on breast cancer subtype is sometimes missing in these registry data, and such missingness is more common in certain patient groups and thus not random. Hence, it is appropriate to use multiple imputation with chained equations (MICE) when estimating subtype-specific breast cancer incidence. I will give examples on how we have applied MICE to Swedish breast cancer data, which choices we made in order to build an imputation model (using mi impute), as well as challenges in combining the imputed estimates using Rubin's rules (using mi estimate).