AI-guided insulin dosing for hospitalized adults with type 2 diabetes on insulin pumps
Validation of Insulin Dose Prediction Model Based on Long Short- Term Memory Artificial Intelligence Algorithm
This will try an AI (LSTM) tool to set insulin pump doses and compare its blood-glucose results with doctors' usual dosing in hospitalized adults with type 2 diabetes.
Quick facts
| Phase | Not applicable |
|---|---|
| Study type | Interventional |
| Enrollment | 400 (estimated) |
| Ages | 18 Years to 75 Years |
| Sex | All |
| Sponsor | Sun Yat-sen University Academic / other |
| Locations | 1 site (Guangzhou, Guangdong) |
| Trial ID | NCT07066891 on ClinicalTrials.gov |
What this trial studies
This randomized controlled trial will enroll 400 hospitalized adults with type 2 diabetes who are treated with continuous subcutaneous insulin infusion (CSII) and randomize them 1:1 to insulin dosing guided by an LSTM-based AI prediction model or to usual clinician-directed dosing. Participants will receive CSII for about 1–2 weeks while wearing continuous glucose monitoring (CGM) and receiving standard diabetes self-management education. The study collects clinical data, CGM profiles (including 8-point glucose checks), and safety outcomes to compare efficacy and hypoglycemia risk between groups. The AI model was trained on a 20-year institutional CSII database and this trial is designed to validate its real-world performance against clinician experience.
Who should consider this trial
Good fit: Hospitalized adults with type 2 diabetes who are treated with CSII for at least 6 days (and under 30 days) and who do not have severe infections, acute uncontrolled complications, or advanced organ failure are ideal candidates.
Not a fit: People with diabetes types other than type 2, those using non‑CSII regimens, patients aged ≥75 deemed unsuitable for intensive insulin therapy, or those with severe hepatic/renal failure or unstable cardiovascular/cerebrovascular disease are unlikely to benefit from participation.
Why it matters
Potential benefit: If successful, the AI model could improve glucose control accuracy, reduce hypoglycemia, and help standardize insulin-pump dosing decisions.
How similar studies have performed: Related machine-learning insulin-dosing models, including LSTM approaches, have shown promise in retrospective and pilot work, but prospective randomized comparisons versus clinicians remain limited.
Eligibility criteria
Show full inclusion / exclusion criteria
Inclusion Criteria: 1. Meets the diagnostic criteria of type 2 diabetes mellitus in the Chinese Guidelines for the Prevention and Treatment of Type 2 Diabetes (2020 edition). 2. Insulin pump is used to control blood glucose during hospitalization, and the duration of CSII treatment period ≥6 days and \<30 days. Exclusion Criteria: 1. Diabetes other than type 2. 2. Age ≥75 years who is not suitable for intensive insulin therapy. 3. Hypoglycemic regimen other than CSII treatment, such as oral hypoglycemic drugs or multiple daily insulin injections during hospitalization. 4. With severe infection or uncontrolled acute complications (including ketoacidosis coma, hyperosmolar hyperglycemia, etc.) , or any condition that the researcher believes not suitable for the study. 5. Severe hepatic and renal insufficiency (ALT≥5 times the upper limit of normal, eGFR\<30ml/min/1.73m2) ), or patients at the acute stage of cardiovascular and cerebrovascular diseases considered unsuitable for study. 6. Pregnancy.
Where this trial is running
Guangzhou, Guangdong
- Department of Endocrinology and Diabetes Center, The First Affiliated Hospital of Sun Yat-sen University — Guangzhou, Guangdong, China (Recruiting)
Study contacts
- Study coordinator: Zhimin Huang, MD. & PhD.
- Email: hzhim@mail.sysu.edu.cn
- Phone: +86 13925057613
How to participate
- Review the eligibility criteria above with your treating physician.
- Visit the official trial page on ClinicalTrials.gov for the most current contact information and recruitment status.
- Contact the listed study coordinator or principal investigator to request pre-screening. Pre-screening is free and never obligates you to enroll.