We employ machine learning to predict the 30-days mortality after heart valve surgeries from demographic and preoperative parameters. We achieve AUC values of almost 84%, while the standard EuroSCORE I provides an AUC of only slightly more than 70% for the given cohort. These results indicate (1) that state-of-the-art machine learning is superior to traditional risk models and (2) that calibrating models to specific institutions and surgical procedures allows for more accurate predictions that have the potential to improve medical decision making.
|Original language||English (American)|
|Publication status||Published - Dec 2017|
|Event||NIPS Workshop on Machine Learning for Health - Long Beach, CA, United States|
Duration: 8 Dec 2017 → 8 Dec 2017
|Workshop||NIPS Workshop on Machine Learning for Health|
|Period||08.12.2017 → 08.12.2017|