Correspondence on ‘Evaluation of the impact of NOAC underdosing and exploration of bleeding risk factors in elderly patients with atrial fibrillation: artificial intelligence-based approach’ by Protzenko et al

Protzenko et al address a question that hospital pharmacists encounter regularly: in older and frail patients with atrial fibrillation, how should concerns about bleeding influence dosing of non-vitamin K oral anticoagulants (NOACs)? Their local analysis is therefore clinically relevant. The point that merits a more cautious interpretation, however, is not the observation that underdosing occurred, but the strength of the inferences drawn from a small retrospective cohort and the extent to which the predictive model can be regarded as clinically actionable.1