Afaik he does mostly videos and dont do blogs. Anyways one argument presented in the video (but dont know if its relevant to this post) is that the datasets used sometimes dont differentiate between medical error and medical harm. Medical harm is where you need to do a particular intervention, but the possible side effects of the intervention leads to the patients death. Furthermore the line between medical harm and medical error is fuzzy especially when you dont have all the information at hand when doing the procedure. The datasets for the papers mentioned in the video were retrospective