Avoiding C-hacking when evaluating survival distribution predictions with discrimination measures

In this article, we consider how to evaluate survival distribution predictions with measures of discrimination. This is non-trivial as discrimination measures are the most commonly used in survival analysis and yet there is no clear method to derive a risk prediction from a distribution prediction. We survey methods proposed in literature and software and consider their respective advantages and disadvantages.

Whilst distributions are frequently evaluated by discrimination measures, we find that the method for doing so is rarely described in the literature and often leads to unfair comparisons or 'C-hacking'. We demonstrate by example how simple it can be to manipulate results and use this to argue for better reporting guidelines and transparency in the literature. We recommend that machine learning survival analysis software implements clear transformations between distribution and risk predictions in order to allow more transparent and accessible model evaluation.

The code used in the final experiment is available at https://github.com/RaphaelS1/distribution_discrimination.