Mount Sinai researchers have published one of the first studies with a machine learning technique called “continuous learning” to examine electronic health records to better predict the progression of patients with COVID-19. The study was published in Journal of Medical Internet Research – Medical Informatics January 18th.
The researchers said the emerging technique promises to create more powerful machine learning models that go beyond a single healthcare system without compromising patient privacy. These models, however, can help triage patients and improve the quality of their care.
Collaborative learning is a technique that teaches an algorithm across multiple devices or servers that have local data patterns, but avoids aggregating clinical data, which is undesirable for reasons including patient privacy issues. Mount Sinai researchers conducted and evaluated federal learning models using data from electronic medical records in five separate hospitals in the health system to predict mortality in patients with COVID-19. They compared the performance of the federal model with a model built on data from each hospital separately, called local models. After training their models in the federal network and testing local model data in each hospital, the researchers found that the federal models in most hospitals showed greater predictive power and outperformed the local models.
“Machine learning models in healthcare often require that diverse and comprehensive data be reliable and transferable beyond the patient population in which they were trained,” said study author Dr. Benjamin Glicksberg, assistant professor of genetics and genomic sciences at Icahn Medical School on Mount Sinai and a member of the Hasso Plattner Institute of Digital Health on Mount Sinai and the Mount Sinai Clinical Intelligence Center. “Continuous learning is becoming increasingly embedded in the biomedical space as a way for models to learn from multiple sources without disclosing sensitive patient data. In our work, we demonstrate that this strategy can be particularly useful in situations such as COVID- 19. ”
Hospital-built machine learning models are not always effective for other patient populations, in part because of models trained based on data from one group of patients that is not representative of the entire population.
“Machine learning in healthcare continues to suffer from repeatability,” said the study’s first author, Akhil Vaid, MD, a postdoctoral fellow in the Department of Genetics and Genomics at Icahn Medical School on Mount Sinai and a member of the Hasso Plattner Institute of Digital Health at Mount Sinai and the Mount Sinai Clinical Intelligence Center. “We hope this paper demonstrates the benefits and limitations of using federated learning with electronic health records for a disease that has a relative lack of data in an individual hospital. Models developed with this federated approach go beyond those built separately from limited sample sizes of isolated hospitals. will see the results of major initiatives of this kind. “
About Mount Sinai Health System
Mount Sinai Health System is New York City’s largest academic medical system, comprising eight hospitals, a leading medical school, and a wide network of outpatient practices throughout the greater New York region. Mount Sinai is a national and international source of unparalleled education, translational research and discovery, and shared clinical leadership that provides the highest quality care – from prevention to treatment of the most serious and complex human diseases. The healthcare system includes more than 7,200 physicians and includes a strong and continuously expanded network of multispecialist services, including more than 400 outpatient clinics in five counties of New York, Westchester, and Long Island. Mount Sinai Hospital ranks 14th as the best hospital in the country and Icahn International Medical School ranked 20th as the best medical school in the country. Mount Sinai hospitals are consistently ranked by region, and our physicians are among the top 1% of all physicians nationwide, according to US News & World Report.
For more information visit https: /
Disclaimer: AAAS in EurekAlert! we are not responsible for the accuracy of press releases posted on EurekAlert! by contributing institutions or using any information through EurekAlert.