Skip to main content
ENDO 2026 Main banner

Pediatric and Adolescent Endocrinology

SUN-191 - Race and Gender Differences in Large Language Model Responses for Adolescents with Newly Diagnosed Type 1 Diabetes

Sunday, June 14, 2026
9:00 AM - 4:00 PM CT
Location: ENDOExpo; Poster Floor
Disclosure(s):
Gloria Wu, MD, MBA: No financial relationships to disclose
Background
Large language models (LLMs) are used by adolescents and families seeking health information when given a new diagnosis such as type 1 diabetes mellitus (T1DM). Pediatric T1DM management requires age-appropriate education, insulin safety, psychosocial support, and coordinated specialty care.

Objective
To evaluate whether LLM responses to a pediatric T1DM query differ by patient race and gender with respect to readability, response length, and inclusion of clinically and psychosocially relevant content.

Methods
Query was asked “I am a recently diagnosed 13-year-old [race] [gender]. I was just diagnosed with type 1 diabetes and need some help managing everything.” to three LLMs (ChatGPT 5.2, Gemini 3.0, and Grok 4.1). The races (White, Black, Hispanic, Asian) and gender (male/female) were observed. Responses were analyzed for word count (WC), Flesch–Kincaid Grade Level (FK), Gunning Fog Index (GF), SMOG Index (SI) and keyword inclusion.

Results
Variation was observed across both race and gender. In ChatGPT responses, Black male received a higher grade level score than a White male (FK 11.7 vs 6.8, respectively) with similar response length (WC 564 vs 576, respectively). In Grok, responses for Black female and Asian female were longer (WC 686 and 651, respectively) than those for a Latina (WC 476), and Grok produced some of the highest grade level scores overall (FK ≈ 11 for Asian female and Black male). Gemini responses generally demonstrated a lower grade level range (White male FK 6.2; Black male and female FK ≈ 6.7–6.8). Across all models, insulin therapy and glucose monitoring were universally mentioned; however, hypoglycemia management, endocrinology follow-up, and diabetes self-management education were inconsistently included. Cultural framing around diet and family context appeared more frequently in responses to Hispanic and Asian, while White more often received generic, task-oriented guidance.

Conclusion
LLM responses for adolescents with newly diagnosed T1DM varied by race and gender, showing a need for pediatric-focused, bias-aware safeguards in patient-facing AI tools.

*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.*