AI Automates Insulin Dosing for Better Blood Sugar Control in Hospitalized Patients


Hospitalized patients with complex dietary restrictions often face hyperglycemia, affecting about 25-50% of such patients and potentially leading to severe complications, particularly in those with diabetes. Managing blood sugar in a hospital setting is challenging due to factors like variable caloric intake, changes in organ function, surgeries, infections, and the difficulties of intensive glucose monitoring and insulin management.

To tackle these issues, researchers have developed a self-adjusting subcutaneous insulin algorithm (SQIA) that autonomously determines insulin doses, thereby reducing instances of both hyperglycemia and hypoglycemia and decreasing the frequency of new insulin orders from doctors. This algorithm was created by medical experts at the University of California, San Francisco (UCSF).

The SQIA, integrated into the medication administration record (MAR) within the electronic medical record system, has been fully implemented from September 2020 to September 2023. It has been used for thousands of patients with dietary restrictions like nothing by mouth (NPO), continuous tube feeds (TF), or intravenous nutrition (TPN). When physicians prescribe rapid-acting insulin under these conditions, they can choose between the SQIA and traditional insulin (CI) dosing orders. The SQIA requires physicians to input only an initial insulin dose, which the algorithm then adjusts automatically, unlike the CI method that necessitates manual updates.

During insulin administration, nurses input the patient’s current glucose level into the MAR, and the SQIA calculates the new insulin dose based on previous insulin and glucose levels, along with the current glucose reading. Ongoing refinements to the algorithm and user interface, based on feedback from nurses, pharmacists, and physicians, have enhanced its precision in titrating the correct insulin dose.

Research showed that the SQIA significantly reduced the number of insulin orders written by physicians by more than twelve times compared to CI dosing. It resulted in higher insulin doses for patients on NPO and TPN diets, decreased severe hyperglycemia rates, and no increase in hypoglycemia, suggesting that CI orders might under-treat some patients. Moreover, the incidence of severe hyperglycemia continued to decline throughout the study, indicating ongoing improvements in the SQIA’s efficacy. Currently, SQIA is the preferred insulin ordering method at UCSF hospitals, chosen by doctors for about 80% of eligible patients.

Read More: Click Here