UNSTRUCTURED CLINICAL NOTES AS A SOURCE OF NON-SMALL CELL LUNG CANCER (NSCLC) BIOMARKER INFORMATION BEYOND STRUCTURED DATA: EVIDENCE FROM NORSTELLALINQ REAL-WORLD DATA UNSTRUCTURED CLINICAL NOTES AS A SOURCE OF NON-SMALL CELL LUNG CANCER (NSCLC...
Author(s)
Isabella Even-Chen, BA1, ilan behm, MPH2, Woojun Daniel Park, PhD3, Atharva Manjrekar, MS4, Rahul Das, PhD5, Allison Perry, PhD1.
1Norstella, New York, NY, USA, 2Norstella, Englewood, CO, USA, 3Norstella, Houston, TX, USA, 4Norstella, West Hartford, CT, USA, 5Norstella, Yardley, PA, USA.
1Norstella, New York, NY, USA, 2Norstella, Englewood, CO, USA, 3Norstella, Houston, TX, USA, 4Norstella, West Hartford, CT, USA, 5Norstella, Yardley, PA, USA.
OBJECTIVES: To characterize the incremental biomarker information captured from human-in-the-loop large language model (LLM) extracted clinical notes beyond structured laboratory records in patients with confirmed NSCLC, and to compare treatment patterns between patients with and without note-derived biomarker ascertainment.
METHODS: Using NorstellaLinQ’s US real-world linked open claims, structured EHR, and clinical notes (January 2019-present), EGFR, KRAS, and PD-L1 biomarker status was ascertained among 76,237 confirmed NSCLC patients via structured laboratory records and LLM-extracted clinical notes. Biomarker results available only in notes were identified, and treatment regimen distribution was compared by ascertainment source.
RESULTS: Among 76,237 confirmed NSCLC patients, EGFR testing was documented for 16,093 (21%), KRAS for 9,092 (12%), and PD-L1 for 29,444 (39%) prior to line 1 via structured laboratory records; the majority lacked structured pre-treatment biomarker documentation. Among EGFR-positive patients, mutation subtype annotations (Exon 19, 21, 20) were available in clinical notes but not in claims data, which captures CPT codes for assay ordering but not biomarker results or mutation subtypes. Among KRAS-positive patients (3,499; 38% of tested), G12C was the predominant subtype; subtype classification was similarly unavailable in claims and required clinical note extraction.
CONCLUSIONS: LLM-extracted clinical notes provide meaningful incremental biomarker information beyond structured laboratory records, enabling mutation subtype stratification not available from coded fields alone. Patients with note-derived biomarker ascertainment showed higher rates of guideline-concordant targeted therapy than the broader ICD-10-identified population. Structured-only analyses risk under ascertainment of biomarker-positive patients and underrepresentation of those receiving precision therapy. Integrating clinical note data is a methodological requirement for biomarker-based RWE in NSCLC, enabling therapy access evaluation, care pathway characterization, and patient stratification not achievable from structured data alone.
METHODS: Using NorstellaLinQ’s US real-world linked open claims, structured EHR, and clinical notes (January 2019-present), EGFR, KRAS, and PD-L1 biomarker status was ascertained among 76,237 confirmed NSCLC patients via structured laboratory records and LLM-extracted clinical notes. Biomarker results available only in notes were identified, and treatment regimen distribution was compared by ascertainment source.
RESULTS: Among 76,237 confirmed NSCLC patients, EGFR testing was documented for 16,093 (21%), KRAS for 9,092 (12%), and PD-L1 for 29,444 (39%) prior to line 1 via structured laboratory records; the majority lacked structured pre-treatment biomarker documentation. Among EGFR-positive patients, mutation subtype annotations (Exon 19, 21, 20) were available in clinical notes but not in claims data, which captures CPT codes for assay ordering but not biomarker results or mutation subtypes. Among KRAS-positive patients (3,499; 38% of tested), G12C was the predominant subtype; subtype classification was similarly unavailable in claims and required clinical note extraction.
CONCLUSIONS: LLM-extracted clinical notes provide meaningful incremental biomarker information beyond structured laboratory records, enabling mutation subtype stratification not available from coded fields alone. Patients with note-derived biomarker ascertainment showed higher rates of guideline-concordant targeted therapy than the broader ICD-10-identified population. Structured-only analyses risk under ascertainment of biomarker-positive patients and underrepresentation of those receiving precision therapy. Integrating clinical note data is a methodological requirement for biomarker-based RWE in NSCLC, enabling therapy access evaluation, care pathway characterization, and patient stratification not achievable from structured data alone.
Conference/Value in Health Info
2026-11, ISPOR Europe 2026, Vienna, Austria
Value in Health, Volume 29, Issue 12S
Code
RWD55
Topic
Clinical Outcomes, Epidemiology & Public Health, Real World Data & Information Systems
Topic Subcategory
Health & Insurance Records Systems
Disease
Oncology