This study looked at patients with Pseudomonas aeruginosa infections to see if a machine-learning model could predict carbapenem resistance early on. The team used routine complete blood count data to train the model and then tested it on separate groups of patients. The goal was to help doctors decide on antibiotics before final lab results were available, which usually take 48 to 72 hours.
The model showed strong performance in predicting resistance. In the main group, it correctly identified resistant cases with high accuracy. When tested on a separate group of patients from another setting, the model remained reliable. It was particularly good at confirming resistance when the model said it was present, with very few false alarms.
However, the study has important limits. If the model said a patient was not resistant, doctors cannot safely rule out the infection based on that alone. The tool is meant to work alongside standard practices, not replace them. This research suggests a low-cost way to help guide early treatment, but more study is needed before it is widely adopted.