A team of researchers at Cedars-Sinai Medical Center in Los Angeles has developed an artificial intelligence model able to identify hospitalized patients at risk of hypoglycemia up to 24 hours before it occurs, according to a study published in the journal npj Digital Medicine.
The model, based on a long short-term memory (LSTM) neural network, was trained and validated using data from nearly 143,000 adult hospital admissions, using medications, lab values, diet orders and the percentage of meals consumed across four-hour windows over five days as variables.
In-hospital hypoglycemia is the most common adverse event during diabetes treatment while admitted, and is linked to longer stays, higher costs and worse outcomes, including seizures, coma or arrhythmias in the most serious cases. Anticipating it a day in advance could help clinical staff intervene before symptoms appear.