By Kelsie Sandoval
Los Angeles June 27: Cedars-Sinai Health Sciences University investigators developed an AI-based model that can identify hospitalized patients at risk of low blood sugar up to 24 hours before the condition occurs. The long short-term memory (LSTM) model, described in npj Digital Medicine, could help clinicians intervene earlier and prevent complications, including, in severe cases, seizures, coma and long-term heart arrhythmias.
The model addresses a longstanding challenge in hospital care. Low blood sugar, also called hypoglycemia, is a common and potentially life-threatening complication among hospitalized patients, including those receiving diabetes treatment, those who are fasting before procedures or those in critical care. However, there are no widely used tools for predicting which hospitalized patients may develop hypoglycemia.
“Today, most hospital care for hypoglycemia is reactive, and we respond after a patient’s blood sugar drops,” said Roma Gianchandani, MD, senior author of the study and vice chair of Quality and Innovation in the Department of Medicine and program director for Diabetes.
The AI model developed by Cedars-Sinai investigators analyzes patterns in medications, lab results, meals and other data from patients’ electronic health records. It collects the information in four-hour intervals over a five-day period and uses it to predict whether a patient will develop hypoglycemia within the next 24 hours.
Researchers developed and tested the model using data from more than 143,000 adult hospital admissions across three Cedars-Sinai Health System hospitals between 2014 and 2025. Investigators also tested the tool using prospective hospital data to confirm their initial findings.
“The AI model is designed to alert patient care teams before a patient experiences low blood sugar and identify the key factors driving that risk,” said Amanda Momenzadeh, PharmD, lead author of the study and a project scientist in the Meyer Research Lab at Cedars-Sinai. “By offering actionable insights to care teams, it also aims to support hospital diabetes management programs.”
Researchers estimate the tool could help prevent about three to four cases of low blood sugar at a large hospital each day. Extrapolating across all hospital beds worldwide, the impact could be substantial.
“What’s exciting is that this isn’t just a theoretical model, but instead, it is built and validated to work prospectively in real time using data hospitals already collect,” said senior author of the study Jesse Meyer, PhD, assistant professor in the Department of Computational Biomedicine at Cedars-Sinai. “By identifying patients at risk earlier, we have an opportunity to reduce preventable complications and improve patient safety.”
If widely adopted, the model could lead to more proactive, data-driven care for hospitalized patients with diabetes and other conditions that affect blood sugar.
Additional Cedars-Sinai authors: Caleb Cranney, Dennis Chen, and Elizabeth Nguyen.
Funding: NIGMS R35GM142502, NIH National Center for Advancing Translational Science (NCATS), and UCLA CTSI Grant Number UL1TR001881
Disclosure: Jesse Meyer, Amanda Momenzadeh, and Caleb Cranney are listed as inventors on a patent application related to this AI tool.