(140) An Agentic AI Virtual Assistant for Gestational Diabetes: Patient Engagement in the First Two Months

To quantify early patient engagement with an agentic AI virtual assistant for GDM.

Study Design:
Gestational diabetes mellitus (GDM) demands daily glucose checks, dietary adjustment, and timely nutritional guidance, but adherence often declines and educator capacity may not meet demand. We developed a patient-facing, agentic AI tool for GDM that received FDA approval as an investigational device to support self-management. The WhatsApp-based AI assistant coordinates six specialized agents (Glucose Logging, Glucose Analysis, Nutritional Advice, Daily Check-in, Vision, Physician Handoff). Functions include structured capture of patient-entered values, trend-based meal guidance, symptom triage, and automated physician handoffs. A secure dashboard allows clinicians to review real-time logs and AI-flagged alerts at-a-glance. Daily operations include a 09:00 am glucose summary and a supportive nudge after 22 h of disengagement (quiet hour rules enforced). Ten pregnant individuals with diet-controlled GDM enrolled 1 June–30 July 2025. Engagement metrics were extracted from every patient message; glucose targets were < 90 mg/dL fasting and < 120 mg/dL post meal.

Results:
Over 60 days, the assistant exchanged 2,170 total messages; 710 (33 %) originated from patients. Participants texted the coach on a median 17 separate days (range 1–59) and sent a median 3 messages per engaged day. At least one glucose value was logged on a median 24 days per patient, and 67% of those days contained ≥ 2 readings. Across 411 captured values, 71% met pregnancy targets. Two symptom texts (“nauseous”, “dizzy”) triggered clinician callbacks within 15 min; no hypoglycemia alerts were missed and no manual overrides were required.

Conclusion:
The agentic AI GDM Assistant sustained meaningful patient interaction and regular glucose logging during the critical first two months after GDM diagnosis, while safely triaging symptoms without added clinician workload. A randomized controlled trial comparing traditional nutritional counseling with agentic AI–based diabetes management is underway.