We congratulate Xu et al on developing a machine learning-derived triage tool for major incidents.1 The use of this technology, together with artificial intelligence, is likely to increase rapidly, and a triage tool that relies on binary questions, rather than quantitative measures or calculations, is intuitively advantageous. However, the proposed tool, while statistically sound, suffers from a lack of face-validity.

One of the three questions in the tool is ‘chest injury present?’. The study used retrospective calculation of Abbreviated Injury Scale scores, which would not be available in the field. In practice, determination of possible chest injury would depend on rapid clinical observation and could include injuries which are later determined to be …