Kernkonzepte
Federated learning models can accurately predict major postoperative complications using electronic health record data from multiple institutions, while preserving data privacy.
Zusammenfassung
The study developed preoperative and postoperative federated learning models to predict the risk of nine major postoperative complications, including prolonged intensive care unit (ICU) stay, mechanical ventilation, neurological complications, cardiovascular complications, acute kidney injury, venous thromboembolism, sepsis, wound complications, and hospital mortality.
The key highlights and insights are:
The federated learning models achieved area under the receiver operating characteristic curve (AUROC) values ranging from 0.81 for wound complications to 0.92 for prolonged ICU stay at the University of Florida Health (UFH) Gainesville center, and from 0.73-0.74 for wound complications to 0.92-0.93 for hospital mortality at the UFH Jacksonville center.
The federated learning models achieved comparable AUROC performance to central learning models, except for prolonged ICU stay, where the performance of federated learning models was slightly higher than central learning models at UFH Gainesville center, but slightly lower at UFH Jacksonville center.
The federated learning models obtained comparable performance to the best local learning model at each center, demonstrating strong generalizability.
Subgroup analysis showed that the performance of federated learning models was independent of patient sex and race, but was affected by age, with a more pronounced effect at the larger data provider (UFH Gainesville).
Sensitivity analysis of downsampling the data from the larger center (UFH Gainesville) to equalize the sample sizes between the two centers resulted in a pronounced performance decline at the larger data provider and a slight performance increase at the smaller data provider (UFH Jacksonville).
Statistiken
Approximately 1 million patients die during or immediately after surgery every year worldwide.
Over 15 million major, inpatient surgeries are performed in the United States, and at least 150,000 patients die within 30 days after surgery each year due to postoperative complications.
Postoperative complications occur up to 32% of surgeries.
Zitate
"Federated learning is shown to be a useful tool to train robust and generalizable models from large scale data across multiple institutions where data protection barriers are high."
"Federated learning models achieved comparable AUROC performance to central learning models, except for prolonged ICU stay, where the performance of federated learning models was slightly higher than central learning models at UFH GNV center, but slightly lower at UFH JAX center."