CHARACTERIZING FGFR BIOMARKER TESTING IN METASTATIC UROTHELIAL CARCINOMA THROUGH NORSTELLALINQ LINKED CLAIMS AND ELECTRONIC HEALTH RECORDS, INCLUDING CLINICAL NOTES WITH HUMAN-IN-THE-LOOP LLM EXTRACTION
Author(s)
Shefali Patel, MS1, ilan behm, MPH2, Atharva Manjrekar, MS3, Rahul Das, PhD4, Allison Perry, PhD, MHS4.
1Norstella, Ferndale, MI, USA, 2Norstella, Englewood, CO, USA, 3Norstella, West Hartford, CT, USA, 4Norstella, New York, NY, USA.
1Norstella, Ferndale, MI, USA, 2Norstella, Englewood, CO, USA, 3Norstella, West Hartford, CT, USA, 4Norstella, New York, NY, USA.
OBJECTIVES: To characterize real-world FGFR biomarker testing rates among patients with metastatic urothelial carcinoma (mUC) and assess variation by provider specialty, practice setting, geography, and line of therapy using linked claims and electronic health record (EHR) data.
METHODS: A retrospective descriptive analysis was conducted using NorstellaLinQ US real-world linked open claims, structured EHR, and clinical notes (January 2021-December 2025). mUC patients were identified via ICD-10 codes (C67x, C65x, C66x, C68x) with ≥2 diagnoses ≥30 days apart, anchored to receipt of systemic therapy (intravesical-only agents excluded) or NLP-confirmed muscle-invasive or metastatic staging. FGFR biomarker documentation was ascertained across three complementary capture arms: human-in-the-loop LLM extraction of biomarker entities from clinical notes (primary), structured reference laboratory records, and claims-based molecular panel codes. Line-of-therapy (LOT) was assigned using a 90-day gap rule. Testing patterns were characterized by LOT, HCP specialty, practice setting, and state.
RESULTS: Among 44,707 patients with mUC, FGFR biomarker documentation was identified in 1,556 (3.5%), with clinical notes representing the primary source of ascertainment. Documented testing increased from 396 patients in 2021 to 605 in 2025. Among patients with LOT-linked FGFR documentation, FGFR-positive rates increased from 24.7% in first-line therapy (N=1,524) to 39.1% in second line (N=1,032) and 53.5% in third line or later (N=608). Documented testing was concentrated in medical oncology and community oncology settings.
CONCLUSIONS: FGFR biomarker documentation remained low overall but increased over time, suggesting growing adoption of biomarker testing in mUC. Progressive enrichment of FGFR-positive disease across treatment lines was consistent with biomarker-guided treatment selection. Persistent variation by care setting suggests opportunities to improve equitable access to molecular testing. Integrating human-in-the-loop LLM extraction with structured EHR and claims data substantially enhances real-world ascertainment of FGFR biomarker testing beyond structured data alone.
METHODS: A retrospective descriptive analysis was conducted using NorstellaLinQ US real-world linked open claims, structured EHR, and clinical notes (January 2021-December 2025). mUC patients were identified via ICD-10 codes (C67x, C65x, C66x, C68x) with ≥2 diagnoses ≥30 days apart, anchored to receipt of systemic therapy (intravesical-only agents excluded) or NLP-confirmed muscle-invasive or metastatic staging. FGFR biomarker documentation was ascertained across three complementary capture arms: human-in-the-loop LLM extraction of biomarker entities from clinical notes (primary), structured reference laboratory records, and claims-based molecular panel codes. Line-of-therapy (LOT) was assigned using a 90-day gap rule. Testing patterns were characterized by LOT, HCP specialty, practice setting, and state.
RESULTS: Among 44,707 patients with mUC, FGFR biomarker documentation was identified in 1,556 (3.5%), with clinical notes representing the primary source of ascertainment. Documented testing increased from 396 patients in 2021 to 605 in 2025. Among patients with LOT-linked FGFR documentation, FGFR-positive rates increased from 24.7% in first-line therapy (N=1,524) to 39.1% in second line (N=1,032) and 53.5% in third line or later (N=608). Documented testing was concentrated in medical oncology and community oncology settings.
CONCLUSIONS: FGFR biomarker documentation remained low overall but increased over time, suggesting growing adoption of biomarker testing in mUC. Progressive enrichment of FGFR-positive disease across treatment lines was consistent with biomarker-guided treatment selection. Persistent variation by care setting suggests opportunities to improve equitable access to molecular testing. Integrating human-in-the-loop LLM extraction with structured EHR and claims data substantially enhances real-world ascertainment of FGFR biomarker testing beyond structured data alone.
Conference/Value in Health Info
2026-11, ISPOR Europe 2026, Vienna, Austria
Value in Health, Volume 29, Issue 12S
Code
RWD23
Topic
Clinical Outcomes, Epidemiology & Public Health, Real World Data & Information Systems
Topic Subcategory
Distributed Data & Research Networks, Health & Insurance Records Systems
Disease
Oncology, Urinary/Kidney Disorders