Researchers use machine learning to target diabetes, predict changes in ICU patients

A global organization focused on diabetes research is using machine learning to analyze years of data among pediatric patients, while new studies out of MIT indicate machine learning could be a critical tool for physicians caring for ICU patients.

On Friday, JDRF, a global organization that funds Type 1 diabetes research, announced a new partnership with IBM to uncover factors that lead to Type 1 diabetes in children by using machine learning to comb through years of research data. Data scientists plan to tap into three different datasets to identify themes and patterns that reveal triggers for diabetes.

RELATED: Digital health companies are targeting obesity and diabetes, but will it work?

“JDRF supports researchers all over the world, but never before have we been able to analyze their data comprehensively, in a way that can tell us why some children who are at risk get [Type 1 diabetes] and others do not,” Derek Rapp, JDRF President and CEO said in an announcement. “IBM’s analysis of the existing data could open the door to understanding the risk factors of T1D in a whole new way, and to one day finding a way to prevent [Type 1 diabetes] altogether.”

Meanwhile, MIT researchers released two new research papers evaluating the benefits of machine learning that analyzes ICU patient data. In one paper that studied ICU data sources like labs, vitals and demographic data, researchers found a deep learning system could use previous cases to make recommendations to physicians while also providing the reasoning behind those decisions—an issue that some have highlighted as a possible roadblock to AI in healthcare.

RELATED: DOD researchers want to peek into the black box of AI, which could impact adoption in healthcare

“The system could potentially be an aid for doctors in the ICU, which is a high-stress, high-demand environment,” lead author Harini Suresh, a Ph.D. student at MIT, said in a release.

A second paper studied a model that uses natural language processing to align data from different EHR systems to ensure machine-learning models can maintain their predictive firepower across various platforms. Researchers found that the model could accurately predict the length of stay and mortality among ICU patients, even if that patient is transferred to a hospital that uses a different EHR system.