AI-Based Insulin Decision System Matches Physician-Guided Therapy for Type 2 Diabetes
Key Highlights
- AI-assisted insulin titration achieved glycemic control noninferior to that of senior endocrinologists.
- No increase in adverse events, including hypoglycemia or hyperglycemia, was observed.
- Physicians reported high satisfaction with the AI tool’s clarity, safety, and efficiency.
Researchers conducting a randomized clinical trial found that an artificial intelligence (AI)–based insulin clinical decision support system (iNCDSS) provided glycemic outcomes comparable to standard therapy administered by senior endocrinologists. Published in JAMA Network Open, the study by Ying and colleagues evaluated whether iNCDSS could match the performance of experienced endocrinologists in adjusting insulin dosages for hospitalized patients with type 2 diabetes.
The study authors from Zhongshan Hospital, Xuhui Central Hospital, and Shanghai Fifth People’s Hospital enrolled 149 adults (mean age, 64.2 years) with glycated hemoglobin levels between 7.0% and 11.0%. Participants were randomly assigned to receive insulin titration guided by either the iNCDSS or senior endocrinology physicians for 5 consecutive days.
Continuous glucose monitoring was used to assess glucose levels, while insulin dosing recommendations generated by the AI system were reviewed daily by physicians. The primary endpoint was the proportion of time spent in the target glucose range (70–180 mg/dL), with a noninferiority margin of 6 percentage points.
Study Findings
The mean time in range was 76.4% in the iNCDSS group versus 73.6% in the physician-guided group (difference, 2.7%; 95% CI, −2.7% to 8.0%), meeting the criterion for noninferiority. Mean sensor glucose levels and glucose variability were similar between groups. The iNCDSS group required a slightly lower median daily insulin dose (27.2 units vs 30.2 units; P = .01), though this difference was attributed to baseline dosage variation.
There were no significant differences in the rates of hypoglycemia, hyperglycemia, or other adverse events. No severe hypoglycemia or diabetic ketoacidosis occurred in either group. Among the 10 physicians who used the AI tool, overall satisfaction averaged 4.1 of 5, with particularly high scores for clarity (4.6), safety (4.4), and time savings (4.2).
Clinical Implications
According to the study authors, the findings suggest that iNCDSS can safely and effectively support insulin titration for hospitalized patients with T2D. The tool’s real-time recommendations and integration into clinical workflows may help reduce physician workload and improve efficiency in glycemic management. The authors also noted that the system could be further developed for broader clinical use, including outpatient or general ward settings.
The investigators acknowledged several limitations. All participating sites were Chinese hospitals, which may limit generalizability to other populations. The study duration was short, covering only 5 days of inpatient intervention, and did not assess long-term glycemic outcomes. In addition, participants were hospitalized for glucose optimization in specialized endocrinology wards, so results may not extend to patients with acute illness, stress-induced hyperglycemia, or those receiving corticosteroids.
Expert Commentary
“In this randomized clinical trial of an AI-based iNCDSS, we developed a system that provided timely and personalized insulin dosage titration recommendations,” the researchers concluded. “The system demonstrated efficacy and safety in insulin dosage titration, showing noninferiority compared with the performance of senior physicians in treating hospitalized patients with type 2 diabetes in specialized endocrinology wards.”
Reference
Ying Z, Fan Y, Chen C, et al. Real-time AI-assisted insulin titration system for glucose control in patients with type 2 diabetes: a randomized clinical trial. JAMA Netw Open. Published online May 7, 2025;8(5):e258910. doi:10.1001/jamanetworkopen.2025.8910