Publication: AI Model Enhances Glucose Control Post-Cardiac Surgery

Managing blood glucose levels in post-cardiac surgery patients is a critical challenge in intensive care units (ICUs), where fluctuations can lead to severe complications. Traditional insulin dosing protocols rely on clinician judgment and standardized guidelines, which, while effective, can lack precision for individual patient needs. A new AI-driven model, GLUCOSE, developed using distributional offline reinforcement learning, aims to bridge this gap by providing personalized insulin dosing recommendations based on real-world patient data.

The research, conducted on a dataset of over 5,000 cardiac surgery patients and validated on nearly 1,600 more, demonstrated that GLUCOSE outperforms conventional methods in maintaining optimal glucose levels. By leveraging historical patient data, the AI model predicts the most effective insulin adjustments, reducing the likelihood of hypoglycemia and hyperglycemia. The model’s ability to adapt to individual patient responses makes it a promising tool for improving post-operative care.

Study overview. a) Schema for model development, testing, and selection. b) Overview of clinician validation study.

One of the key advantages of AI in this context is its ability to learn from a vast array of patient cases, enabling it to refine recommendations beyond human capability. Unlike rule-based approaches, which apply generalized guidelines, GLUCOSE continuously refines its predictions based on diverse patient profiles. This advancement not only improves glycemic control but also reduces the cognitive load on ICU staff, allowing clinicians to focus on broader aspects of patient care.

GLUCOSE performance. a) OPE counterfactual estimated performance of the model computed by FQE (solid lines) compared to the returns by the treating clinicians (dotted lines) with 95% CI. b) Comparison of TIR and average glucose relative to insulin dosing differences with 95% CI. c) Average insulin doses across several glucose ranges with 95% CI

As artificial intelligence continues to integrate into medical practice, models like GLUCOSE highlight the potential of machine learning to enhance decision-making in critical care. While further validation through clinical trials is necessary, the study marks an important step toward AI-assisted medicine that is both precise and adaptive. By improving glucose management in high-risk patients, this technology could pave the way for broader applications of AI in intensive care settings, ultimately leading to better patient outcomes and reduced healthcare burdens.

Source: GLUCOSE: A Distributional Reinforcement Learning Model for Optimal Glucose Control After Cardiac Surgery. Jacob M. Desman, Zhang-Wei Hong, Moein Sabounchi, Ashwin S. Sawant, Jaskirat Gill, Ana C Costa, Gagan Kumar, Rajeev Sharma, Arpeta Gupta, Paul McCarthy, Veena Nandwani, Doug Powell, Alexandra Carideo, Donnie Goodwin, Sanam Ahmed, Umesh Gidwani, Matthew Levin, Robin Varghese, Farzan Filsoufi, Robert Freeman, Avniel Shetreat-Klein, Alexander W Charney, Ira Hofer, Lili Chan, David Reich, Patricia Kovatch, Roopa Kohli-Seth, Monica Kraft, Pulkit Agrawal, John A. Kellum, Girish N. Nadkarni, Ankit Sakhuja.