What is it about?
Federated learning (FL) has become a commonly used method to combine datasets together in settings where data privacy is important like healthcare. Still, questions remain about the utility of FL in taking care of differences between datasets and still performing well. This study looked at using health records from many hospitals to make predictions about the risk of people getting AKI and Sepsis. While we found that federated learning could mitigate heterogeneity across datasets, certain key points need to be kept in mind when using FL.
Featured Image
Photo by Marcelo Leal on Unsplash
Why is it important?
This paper is important because it addresses a critical challenge in healthcare: how to effectively use healthcare data from multiple sources while preserving patient privacy. Federated learning has shown promise in addressing this challenge, but data heterogeneity may affect its performance. By exploring the impact of data heterogeneity on predictive models for acute kidney injury and sepsis onset risk, this study sheds light on how to build accurate predictive models in a federated learning framework, and highlights the importance of understanding data heterogeneity in healthcare. This research could lead to better patient outcomes by improving the accuracy of disease risk prediction models.
Perspectives
We hope this collaborative effort of analyzing eICU data in a federated framework provides other bioinformaticians insights into the scenarios in which it's best to use federated learning. By doing so, we can develop models that truly aid patients in and out of the hospital.
Suraj Rajendran
Weill Cornell Graduate School
Read the Original
This page is a summary of: Data heterogeneity in federated learning with Electronic Health Records: Case studies of risk prediction for acute kidney injury and sepsis diseases in critical care, PLOS Digital Health, March 2023, PLOS,
DOI: 10.1371/journal.pdig.0000117.
You can read the full text:
Contributors
The following have contributed to this page
