MIT researchers have developed a machine-learning model that groups patients into subpopulations by health status to better predict a patient’s risk of dying during their stay in the ICU. This technique outperforms “global” mortality-prediction models and reveals performance disparities of those models across specific patient subpopulations. By training on patients grouped by health status, neural network can better estimate if patients will die in the hospital.

In a paper recently presented at the Proceedings of Knowledge Discovery and Data Mining conference, MIT researchers describe a machine-learning model that functions as the best of both worlds: It trains specifically on patient subpopulations, but also shares data across all subpopulations to get better predictions.

Read More