What question did this study set out to answer?

This research aims to understand the documentation and impact of GLP-1 medication use among breast cancer patients.

February 19, 2026

Abstract PD8-07: Real-world glucagon-like peptide-1 use and association with clinical characteristics, social determinants, and circulating tumor DNA positivity in patients with breast cancer

Key Points

This research aims to understand the documentation and impact of GLP-1 medication use among breast cancer patients.
Extracted EHR data using large language models (LLMs) from Flatiron Health Research Database.
Assessed GLP-1 use in relation to clinical and sociodemographic factors.
Applied multivariable logistic regression models to evaluate associations between characteristics and GLP-1 use.
Conducted exploratory analyses on GLP-1 use in relation to ctDNA testing and results.
6.65% of breast cancer patients had documented evidence of GLP-1 use.
GLP-1 users were more likely to be obese, non-Latinx Black, rural residents, and telemedicine users.
Higher rates of ctDNA testing were observed in GLP-1 users (OR=1.84).
GLP-1 users had a lower likelihood of ctDNA positivity (OR=0.76).

Abstract

Abstract Background: Obesity is a known risk factor for breast cancer (BC) incidence and recurrence. Glucagon-like peptide-1 (GLP-1) medications have emerged as effective treatments for weight management, with potential implications for oncology supportive care. However, their real-world use and impact among patients with BC remain poorly characterized. We extracted information from the electronic health record (EHR) using large language models (LLMs) to assess (1) documentation of GLP-1 use among patients with BC; (2) clinical, sociodemographic, and social determinants of health (SDOH) factors associated with GLP-1 use; and (3) the association between GLP-1 use and ctDNA testing and test results among a US-based cohort of patients with BC. Methods: We leveraged the US-based Flatiron Health Research Database, focusing on patients diagnosed with BC from January 2011 to February 2025. We used an LLM to extract GLP-1 medication details from unstructured EHR documents. We tested several prompt engineering strategies to optimize LLM extraction performance, including chain-of-thought and text curation techniques. We estimated a series of multivariable logistic regression models assessing associations between patient characteristics and GLP-1 use (i.e., documentation of GLP-1 use in EHR), sequentially adjusting for clinical factors (e.g., age, stage, year of diagnosis), sociodemographics (e.g., race/ethnicity), and SDOH (e.g., telemedicine use, practice setting, urban/rural). In exploratory analyses, we also estimated the unadjusted association between GLP-1 use and ctDNA testing (yes vs. no) and result (positive vs. negative) at any time. Results: Among our cohort of 967,968 patients diagnosed with BC, 64,400 (6.65%) patients had documented evidence of GLP-1 use and 16,689 (1.72%) patients underwent ctDNA testing. The LLM extracted GLP-1 use with high performance (F1 score = 0.94). Adjusted analyses revealed a higher likelihood of GLP-1 use among the following patient groups relative to their counterparts: patients with obesity (BMI≥30 vs. BMI30; Odds Ratio OR=2.38; 95%CI:2.31-2.45), HR+/HER2- patients (vs. HR-/HER2-; OR=1.05; 95%CI:1.01-1.10), non-Latinx (NL) Black patients (vs. NL-White; OR=1.37; 95%CI:1.31-1.44), patients residing in rural areas (vs. urban; OR=1.13; 95%CI:1.08-1.18), and telemedicine users (vs. non-users; OR=1.39; 95%CI:1.35-1.43) were more likely to have evidence of GLP-1 use. Older patients (aged 75 vs aged 18-49; OR=0.54; 95%CI:0.46-0.64), advanced stage patients (stage IV vs. stage I OR=0.64; 95%CI:0.58-0.71), Latinx patients (vs. NL-White; OR=0.94; 95%CI:0.89-0.99), Asian patients (NL-White; OR=0.67; 95%CI:0.60-0.75), patients receiving care at community practices (vs. academic; OR=0.81; 95%CI:0.77-0.85), and those residing in neighborhoods with higher levels of limited English proficiency (high LEP vs. low LEP; OR=0.83; 95%CI:0.78-0.89) were less likely to have evidence of GLP-1 use. Compared with patients with no evidence of GLP-1 use, GLP-1 users had higher rates of ctDNA testing (OR=1.84; 95%CI:1.75-1.93) and a lower likelihood of ctDNA positivity (OR=0.76; 95%CI:0.67-0.85). Conclusions: Clinical, sociodemographic, and SDOH factors influence GLP-1 use among patients with BC. In exploratory analyses, we found that GLP-1 use was also associated with lower rates of ctDNA positivity, a biomarker linked to minimal residual disease and increased risk of recurrence. This study demonstrates the feasibility of using LLMs to extract GLP-1 medication details from the EHR, and helps lay the groundwork for future research evaluating the clinical impact of GLP-1 use on BC outcomes and recurrence risk, as well as equitable access to these therapies. Citation Format: C. A. Ryals, A. Blarre, B. Adamson, D. Bower, G. G. Ho, O. Mbah, S. Radlein, E. Fidyk, F. Cody Stanford, A. B. Cohen. Real-world glucagon-like peptide-1 use and association with clinical characteristics, social determinants, and circulating tumor DNA positivity in patients with breast cancer abstract. In: Proceedings of the San Antonio Breast Cancer Symposium 2025; 2025 Dec 9-12; San Antonio, TX. Philadelphia (PA): AACR; Clin Cancer Res 2026;32(4 Suppl):Abstract nr PD8-07.

Mark Helpful

Bookmark

Relay