Using ‘integrated learning’ to strengthen COVID-19 outcome prediction

Mount Sinai researchers have published one of the first studies using a machine learning method called “federated learning” to examine electronic health records to better predict how COVID-19 patients develop onwards. The study was published in the Journal of Medical Internet Research – Medical Informatics on 18 January.

The researchers said the emerging innovation promises to create more robust machine learning models that extend beyond a single health system without compromising patient privacy. These models can, in turn, help patients improve the quality of their care.

Integrated learning is a method of training an algorithm across multiple devices or servers that holds local data samples but avoids the collection of clinical data, which is not desirable for reasons including patient privacy issues . Mount Sinai researchers implemented and evaluated integrated learning models using data from electronic health records at five individual hospitals within the Health System to predict mortality in COVID-19 patients. They compared the performance of an integrated model with those constructed using data from each individual hospital, called local models. After training their models on an integrated network and testing local model data at each hospital, the researchers found that the federated models showed improved predictive power and outperformed the local models. at most hospitals.

“Machine learning models in healthcare often require robust and large-scale diverse data and the ability to translate outside the population of trained patients,” said the corresponding author of the study , Benjamin Glicksberg, PhD, Associate Professor of Genetics and Genomic Sciences at the Icahn School of Medicine at Mount Sinai, and a member of the Hasso Plattner Institute for Digital Health at Mount Sinai and the Mount Sinai Clinical Information Center. “Federal learning is gaining traction within the biochemical space as a way to learn models from multiple sources without revealing any sensitive data about patients. In our work, we show that this strategy may be particularly useful in situations such as COVID-19. “

Machine learning models built within a hospital are not always effective for other patient numbers, in part because models are trained on data from one group of patients who do not represent the patient. whole population.

Machine learning in health care still suffers a reproductive crisis. We hope that this work highlights the benefits and limitations of using integrated learning with electronic health records for a data-deficient disease in an individual hospital. Models constructed using this integrated approach perform better than those constructed separately from remote hospital sample sizes. It will be encouraging to see the results of larger campaigns of this nature. “

Akhil Vaid, MD, First Author of the Study, Postdoctoral Fellow, Department of Genetics and Genomic Sciences, Icahn School of Medicine, Mount Sinai, and Member of the Hasso Plattner Institute for Digital Health at Mount Sinai and Mount Sinai Clinical Information Center

Source:

Mount Sinai Health System

.Source