by Prateek Chopra | June 28, 2026 | Diabetes Conferences | ADA 2026
Innovation Overview
- An agentic artificial intelligence (AI) system was developed to support complex treatment decisions in type 2 diabetes (T2D).
- The system generates structured, evidence-cited recommendations aligned with diabetologist workflows.
Key Functionalities
- Patient triage and problem identification.
- Medication recommendations.
- Treatment strategy development.
- Dose adjustment guidance.
- Monitoring plans and patient education.
Evaluation Approach
- Synthetic T2D cases were developed and validated by three senior diabetologists.
- Twelve diabetologists performed double-blind evaluations of AI-generated recommendations across 48 synthetic cases.
- Assessments were conducted using a Delphi-validated 29-item framework with a 5-point Likert scale.
Clinician Acceptability Ratings (Score ≥4)
- Patient triage and problem listing: 96.1%.
- Medication recommendations: 90.9%.
- Treatment strategy: 85.4%.
- Dose adjustments: 89.2%.
- Monitoring and patient education: 87.8%.
- Reasoning reliability: 100%.
- Clinical utility: 97.0%.
- Real-world feasibility: 90.9%.
Key Findings
- High levels of agreement and acceptability were observed across all evaluated domains.
- Reasoning reliability received the highest clinician ratings.
- No domain received predominantly unfavorable evaluations.
Clinical Implications
- AI-assisted decision support may help reduce practice variation in T2D management.
- Evidence-cited recommendations could facilitate more consistent and guideline-aligned care.
- The system has potential to support clinicians in managing increasingly complex treatment decisions.
The agentic AI system generated evidence-based T2D treatment recommendations with high clinician acceptability across multiple decision-making domains. These findings highlight its potential to support standardized, reliable, and evidence-informed diabetes care in clinical practice.