Conference Coverage

LLM Pipeline Flags 218 Untreated Patients With Suspected ATTR-CA

Edited by:
Brian McCurdy

Key Highlights

  • The pipeline identified 218 untreated candidates, representing 0.9% of the screened cohort.
  • Among these patients, 165 (76%) had no pyrophosphate scintigraphy order, and only 5 had a scheduled study.
  • Light-chain amyloidosis testing was absent or incomplete in 193 candidates; 19 had positive or equivocal findings.
  • The pipeline identified all 28 positive-control patients with established ATTR-CA, including nine who had discontinued therapy.

A multi-agent large language model (LLM) pipeline identified 218 untreated patients with suspected transthyretin cardiac amyloidosis (ATTR-CA) whose diagnostic evaluations had stalled at the University of Texas Medical Branch. Incomplete pyrophosphate (PYP) testing was the principal barrier reported. The team presented the findings at the 2026 American Heart Association meeting.

Investigators examined whether an LLM system with citations supporting its findings, audit trails, and human oversight could identify and risk-stratify patients with documented suspicion of amyloidosis but incomplete evaluation. The study focused on gaps between clinical suspicion and definitive assessment with PYP scintigraphy and exclusion of light-chain (AL) amyloidosis.

The pipeline was applied to 24,080 adults with heart failure, restrictive cardiomyopathy, or severe aortic stenosis at UTMB between 2024 and 2026. Hospice patients and those with known AL amyloidosis were excluded. The system operated within a HIPAA-compliant enterprise AI platform.

Sequential agents extracted cardiology-documented amyloid suspicion, PYP results, and AL test interpretations from unstructured notes and reports. The system classified untreated patients with active suspicion and incomplete diagnostic evaluations as care-gap candidates and stratified them by readiness for ATTR confirmation.

Study Findings

The pipeline identified an amyloid signal in 648 patients. Cardiology had ruled out or declined evaluation in 411, leaving 237 with active suspicion. After excluding 19 receiving therapy, 218 untreated candidates remained. Of these, 165 (76%) had no PYP order, and 5 had a scheduled study. Clinical phenotypes overlapped: 216 had heart failure, 61 had restrictive cardiomyopathy, and 49 had severe aortic stenosis. Within the aortic stenosis subgroup, 39 lacked a PYP order.

Six candidates had negative AL testing and were classified as immediately eligible for scintigraphy. Nineteen (9%) had positive or equivocal findings requiring hematology evaluation before ATTR-directed therapy, while 193 had absent or incomplete AL testing. Thirty-eight had documented suspicion dating to 2024 or earlier without subsequent PYP evaluation. As a positive control, the pipeline identified all 28 patients with established ATTR-CA, including nine who had discontinued treatment.

Clinical Implications

This approach may help close diagnostic gaps by stratifying patients and identifying those requiring AL exclusion, according to the authors. Systematic patient review and outreach are underway.

Expert Commentary

“By risk-stratifying patients and identifying those requiring exclusion of AL amyloidosis, this scalable approach may help close diagnostic care gaps and accelerate treatment initiation,” the researchers concluded.

Reference

Qadeer A, Wardhere A, McCrumb T, et al. Closing the ATTR-CM diagnostic care gap: a governed multi-agent LLM pipeline for identification and risk stratification of untreated candidates. Abstract scheduled for presentation at: American Heart Association 2026; November 9, 2026. Abstract MDP2177.