Development and Validation of an Interpretable Model for Predicting Multiple Postoperative Complications

Researchers developed and validated a Markov-embedded multilabel model to predict the risks of various postoperative complications among surgical inpatients. The model outperformed other existing models, achieving the highest average area under the receiver operating characteristic curve (AUC) for eight outcomes, including cardiac complications, neurological complications, and mortality. The model identified important preoperative variables, but the interaction between complications was found to contribute more to the risk. The interpretability of the model provides valuable insights for clinical decision-making and the identification of high-risk patients.