Hiring in life sciences? Share your open positions with our professional community. Read more Close

Advertisement

Leveraging Large Language Models to Understand Reasons for SGLT2 Nontreatment Among Patients With Heart Failure.

Created on 11 Nov 2026

Authors

Elena Vasti, Minh Nguyen, Emily Alsentzer, Emily Chu, Xichong Liu, Madie Chakoumakos, Christian Haudenschild, Benjamin Wang, Jimmy Zheng, Neil M Kalwani, Ankeet S Bhatt, Andrew P Ambrosy, Sneha S Jain, Fatima Rodriguez, Paul Heidenreich, Alexander T Sandhu

Published in

Journal of cardiac failure. Volume 32. Issue 11. Pages 1902-1911.

Abstract

Many patients with heart failure (HF) and a guideline indication for sodium-glucose cotransporter-2 inhibitors (SGLT2is) do not receive a prescription, but the reasons behind this are unclear. This study optimizes an large language model framework to identify reasons for the lack of SGLT2i prescriptions.
We extracted clinical notes for patients >18 years old with a first HF diagnosis identified by International Classification of Diseases, 10th edition, code seen in cardiology clinics between August 10, 2021, and March 1, 2025. Five annotators labeled reasons for SGLT2i nonprescription (the test set); the remaining unlabeled notes served as the inference set. Using both zero-shot (prompts do not include examples) and few-shot prompting techniques, we compared 2 large language models from OpenAI on their ability to classify the documented reason for nonprescription into 1 of 6 themes and 1 of 14 subthemes. We identified 599 HF patients (with 1123 notes), with an average age of 71 years and predominantly having HF with preserved ejection fraction. We used 313 labeled and 705 unlabeled notes as the test and inference sets, respectively, for a total set of 1018 notes. GPT-o1 outperformed GPT-4o and was 84.7% accurate (95% confidence interval 80.2%-88.5%) in identifying theme reasons for SGLT2i nonprescription. The most frequent theme was physician-attributed factors (n = 292 notes [28.7%]), among the total set. Physician deferral to a future visit (n = 157 notes [15.4%]) and the physician perceiving the patient to be at low risk or unlikely to benefit (n = 75 [7.4%]) were the most prevalent subthemes.
We optimized a large language model to accurately discriminate reasons for not prescribing SGLT2is from clinical notes. When a reason for SGLT2i nonprescription was documented, delayed initiation of SGLT2i and physician perception of patients as being at low risk were the most common reasons, offering valuable insight to improve guideline-indicated SGLT2i prescription.

PMID:
43009243
Bibliographic data and abstract were imported from PubMed on 11 Nov 2026.

Read full publication at:
Please sign in to see all details.

Advertisement

Stats

  • Community rating n/a 0 votes
  • Reviewers' rating n/a 0 votes
  • Your rating

1-terrible, 9-excellent. How would you rate this publication? Sign in in to submit your rating.

  • Recommendations n/a n/a positive of 0 vote(s)
  • Views 7
  • Comments 0

Recommended by

  • No recommendations yet.

Post a comment

You need to be signed in to post comments. You can sign in here.

Comments

There are no comments yet.

Advertisement