[Feature] Stratified evaluation of prediction models

[lhjohn]

During the coagulopathy and dementia studies at Erasmus MC, we found that it can be useful to not only evaluate models on the full test/validation set, but also in sub-populations of it.

Particularly, evaluation in certain age and sex strata showed significant limitations of a model that were initially masked by evaluating in the full test/validation set. Those limitations had impact on the final recommendations of how to effectively use the model.

I will attempt to implement a simple function to evaluate in custom age and sex strata of a given test/validation set.