Type 2 Diabetes Mellitus (T2DM)
Conditions
Brief summary
This study tests whether an artificial intelligence (AI) tool can help doctors choose better diabetes medicines for their patients. Type 2 diabetes is very common, but there are far more patients than diabetes specialists, so many patients are treated by doctors who are not diabetes specialists. The researchers built an AI consultation system that gives doctors real-time suggestions and predictions about diabetes medicines while they are prescribing. The doctor always makes the final decision. In this trial, patients with type 2 diabetes whose blood sugar is not well controlled will be placed by chance (randomly) into one of two groups. In one group, the doctor uses the AI system when deciding on diabetes medicines. In the other group, the doctor prescribes as usual, without the AI system. All medicines used are already approved in Taiwan and given at approved doses. The study follows each patient for 12 months, with check-ups at the start and at 3, 6, 9, and 12 months. The main goal is to compare how much the patients' long-term blood sugar level (HbA1c) improves between the two groups after one year. The researchers also look at how many patients reach their blood sugar target, how often low blood sugar happens, and whether any side effects occur. The aim is to find out whether using the AI tool leads to better blood sugar control.
Interventions
DEVICEAI-assisted antidiabetic drug consultation system
A machine-learning based clinical decision support software that provides non-specialist physicians with real-time, interactive antidiabetic prescribing recommendations, a drug-prioritization order, and outcome predictions (e.g., the predicted likelihood of reaching glycemic targets and responder/non-responder status for individual drugs). The system was developed and validated using the NTUH integrated medical database platform. It provides advisory recommendations only; the treating physician retains full control over the final prescribing decision. All recommended medications are approved in Taiwan and within approved dose ranges.
OTHERManual antidiabetic prescribing (without AI)
Antidiabetic medications prescribed manually by non-specialist physicians according to usual clinical practice, without using the AI consultation system. All medications are approved in Taiwan and prescribed within approved dose ranges.
Sponsors
National Taiwan University Hospital
Study design
Allocation
RANDOMIZED
Intervention model
PARALLEL
Primary purpose
TREATMENT
Masking
SINGLE (Subject)
Intervention model description
Prospective, single-center, parallel-group randomized controlled trial. Eligible patients are randomly allocated 1:1 to AI-assisted prescribing versus conventional prescribing, using stratified randomization balancing age, sex, and baseline HbA1c. Follow-up is 12 months.
Eligibility
Sex/Gender
ALL
Age
19 Years to 80 Years
Healthy volunteers
No
Inclusion criteria
* Adults aged 18 to 80 years * Diagnosis of type 2 diabetes for at least 6 months * HbA1c above 8% within the past 3 months * Currently using one or more oral antidiabetic drugs * Able to understand and provide written informed consent
Exclusion criteria
* Pregnancy or breastfeeding * Recent participation in another interventional clinical trial * Cognitive impairment precluding understanding of the study * Active cancer treatment within the past 6 years * Use of systemic steroids
Design outcomes
Primary
| Measure | Time frame | Description |
|---|---|---|
| Change in HbA1c from baseline to 12 months | Baseline and 12 months | The between-group difference in the change in glycated hemoglobin (HbA1c) from baseline to 12 months, comparing the AI-assisted prescribing arm with the manual prescribing (control) arm. HbA1c reflects long-term glycemic control. The primary analysis uses analysis of covariance (ANCOVA) adjusting for baseline HbA1c, following the intention-to-treat principle. |
Secondary
| Measure | Time frame | Description |
|---|---|---|
| Proportion of participants achieving HbA1c < 7.0% at 12 months | 12 months | The proportion of participants reaching the glycemic target of HbA1c below 7.0% at 12 months, compared between arms using chi-square tests. |
| Incidence of hypoglycemia over 12 months | Up to 12 months | Incidence of hypoglycemic events over the 12-month follow-up, graded by severity: Level 1, glucose \< 70 mg/dL (3.9 mmol/L) and ≥ 54 mg/dL (3.0 mmol/L); Level 2, glucose \< 54 mg/dL (3.0 mmol/L); Level 3, severe hypoglycemia requiring assistance of another person regardless of glucose value. Compared between arms using Poisson regression. |
| Incidence of prespecified adverse events over 12 months | Up to 12 months | Incidence of prespecified adverse events over the 12-month follow-up, including urinary tract infection, lower-limb edema, signs of heart failure, fractures, nausea, vomiting, and diarrhea. Compared between arms using Poisson regression. |
| Change in HbA1c from baseline at 3, 6, 9, and 12 months | Baseline, 3, 6, 9, and 12 months | — |
Countries
Taiwan
Contacts
CONTACTYi-Cheng Chang, M.D.
CONTACTPan Hou Che, PhD student
Outcome results
None listed