Fair question. The model we comment on both suffer from the problem described in the article but also a more severe problem: The developers sampled obvious cas

Fair question. The model we comment on both suffer from the problem described in the article but also a more severe problem: The developers sampled obvious cases og hypotension and nonhypotension, and trained the model to distinguish those. And also validated it on data that was similarly dichotomous. In reality the outcome is often between these two scenarios. But worse, they also introduce a more severe problem where as range of an important predictor is only available in the hypotension outcome.