Balancing machine learning with human application: the importance of interpretability and context

We read with interest the study by Chew et al, who applied an innovative machine learning (ML) pipeline to a dataset of 17 200 patients who underwent total knee arthroplasty (TKA) at a single center.1 These investigators identified two distinct postoperative pain archetypes and developed a predictive model able to classify patients in either high or low pain clusters with 64% accuracy.1 We commend the authors for sharing their detailed methodology and clinically meaningful findings and wish to offer a few additional perspectives.