SUN-208 - Automated Transcription and Artificial Intelligence to Strengthen Adherence Related Support in the Tuitek Patient Support Program

Background:
The Tuitek Patient Support Program in Latin America offers a structured intervention and support led by specialized advisors, focusing on behavior-change and education directed at caregivers of pediatric patients who are initiating treatment with recombinant human growth hormone with Saizen®. The objective of the intervention is to identify early adherence risks among caregivers and provide personalized coaching to address these risks, thereby supporting adherence related behaviors. The program focuses on four adherence-related domains: disease and treatment coherence, emotional burden, treatment related anxiety, and self-administration. There is a need for tools that assist advisors in preparing for and reflecting on adherence-focused conversations.

Aim:
To pilot evaluate whether a workflow combining automated transcription and artificial intelligence (AI) can generate structured insights that may assist advisors during adherence-oriented interactions with caregivers whose children are receiving growth hormone treatment (GHT).

Methods:
Five caregivers previously identified as high risk in one of the four adherence related domains during the initial Tuitek questionnaire were selected. Audio recordings of the first accompaniment calls were transcribed using an automated speech recognition approach. The large language model based AI workflow generated short and extended summaries, verified whether the advisor addressed the identified high-risk domain, produced post conversation risk assessments with explanations, identified actions taken by the advisor, detected dominant emotions, screened for additional potential risk signals, and captured new topics and emotional cues outside the predefined domains.

Results:
Across all five cases, AI correctly linked each transcript to the documented high-risk domain and confirmed whether it was addressed during the call. The system generated coherent summaries, identified emotional and behavioral cues, and provided plausible post conversation risk reassessments. The workflow also revealed additional topics, including questions related to treatment administration, injection technique, and practical issues such as device management or supply needs.

Conclusions:
Automated transcription combined with AI can provide structured, clinically meaningful insights that support advisors in tailoring adherence-focused conversations with caregivers of children receiving GHT. Future work will scale this methodology to larger datasets and refine emotion and risk detection using more advanced techniques, such as valence-arousal modeling or acoustic emotion analysis.

Funding disclosures:
This study was sponsored by Merck (CrossRef Funder ID: 10.13039/100009945).

*Unless otherwise noted, all abstracts presented at ENDO must not be released to the press or the public until the date and time of presentation. For oral presentations, the abstracts are embargoed until the session begins. The Endocrine Society reserves the right to lift the embargo on specific abstracts that are selected for promotion prior to or during ENDO.*