> paradoxical risk, where successful treatments unexpectedly lower the risk of higher-risk patients to below that of untreated lower-risk patients.

This seems perhaps tautological whenever the treatment intensity is binary, and it's an effective treatment. Someone at the threshold that receives treatment would necessarily do better than someone at the threshold not receiving the treatment.

It's a pretty good argument against any binary treatments, or at least to set the threshold low enough that improvement with treatment at the threshold is minimal.