Researchers at Cedars-Sinai Health Sciences University have built and prospectively validated an AI model that identifies hospitalized patients likely to experience a dangerous blood sugar crash as much as a full day before it occurs. The model, described in npj Digital Medicine this week, was trained on more than a decade of electronic health records and tested in a live clinical environment at Cedars-Sinai hospitals. Every year in the United States, more than 7.8 million hospital discharges involve adults with diabetes, and studies show that roughly 10 percent of them will develop hypoglycemia. Research has documented that hospitalized patients who experience hypoglycemia face an average of 4.1 additional days in the hospital compared to those who do not. "Today, most hospital care for hypoglycemia is reactive, and we respond after a patient's blood sugar drops," said Roma Gianchandani, MD, senior author and vice chair of quality and innovation in Cedars-Sinai's Department of Medicine. The system uses a long short-term memory, or LSTM, architecture.