UNLOCKING CHRONIC LYMPHOCYTIC LEUKEMIA (CLL) DISEASE STAGE IN REAL-WORLD DATA: CLINICAL NOTES AS THE PRIMARY SOURCE OF RAI AND BINET STAGING INFORMATION USING NORSTELLALINQ
Author(s)
Isabella Even-Chen, BA1, ilan behm, MPH2, Shefali Patel, MS3, Rahul Das, PhD4, Atharva Manjrekar, MS5, Juan Diego Irizarry-Cole, PhD1, Allison Perry, PhD1.
1Norstella, New York, NY, USA, 2Norstella, Englewood, CO, USA, 3Norstella, Ferndale, MI, USA, 4Norstella, Yardley, PA, USA, 5Norstella, West Hartford, CT, USA.
1Norstella, New York, NY, USA, 2Norstella, Englewood, CO, USA, 3Norstella, Ferndale, MI, USA, 4Norstella, Yardley, PA, USA, 5Norstella, West Hartford, CT, USA.
OBJECTIVES: To demonstrate that Rai and Binet staging in CLL are absent from claims and structured EHR fields and are accessible only through human-in-the-loop large language model (LLM) extracted clinical notes, and to characterize the treated CLL population for whom stage-stratified analyses are enabled by note-derived data.
METHODS: Using NorstellaLinQ’s US real-world linked open claims, structured EHR, and clinical notes (January 2019-September 2025), 204,756 incident CLL patients were identified. Structured claims and EHR fields were assessed for CLL staging data (Rai or Binet). LLM-extracted clinical notes were used to extract staging and mutation status for IGHV, TP53, and del(17p). Treatment patterns were characterized for the LLM-extracted-staged subpopulation and compared to the unstaged remainder. Staging is not a coded field; clinical notes represent the only available data source.
RESULTS: Rai and Binet classifications were absent from billng claims and structured EHR fields across the incident CLL population. LLM-extracted clinical notes identified staging documentation for 6,523 patients (4,833 early-stage; 1,690 late-stage) and biomarker status for 3,612, populations otherwise not assessable from structured data alone. Late-stage patients initiated treatment substantially earlier (51% within 90 days vs 24% of early-stage patients) and had longer average BTKi line 1 durations (484 days vs 381 days). LLM-extracted-staged patients showed higher rates of later-line therapy and BTKi rechallenging relative to the unstaged remainder.
CONCLUSIONS: CLL disease stage is not a coded variable in claims or structured EHR but can be recovered from unstructured clinical documentation. LLM-extracted staging unlocks analyses entirely inaccessible from structured data alone, establishing a methodological case for its integration as a prerequisite for rigorous CLL RWE supporting HTA and coverage decisions.
METHODS: Using NorstellaLinQ’s US real-world linked open claims, structured EHR, and clinical notes (January 2019-September 2025), 204,756 incident CLL patients were identified. Structured claims and EHR fields were assessed for CLL staging data (Rai or Binet). LLM-extracted clinical notes were used to extract staging and mutation status for IGHV, TP53, and del(17p). Treatment patterns were characterized for the LLM-extracted-staged subpopulation and compared to the unstaged remainder. Staging is not a coded field; clinical notes represent the only available data source.
RESULTS: Rai and Binet classifications were absent from billng claims and structured EHR fields across the incident CLL population. LLM-extracted clinical notes identified staging documentation for 6,523 patients (4,833 early-stage; 1,690 late-stage) and biomarker status for 3,612, populations otherwise not assessable from structured data alone. Late-stage patients initiated treatment substantially earlier (51% within 90 days vs 24% of early-stage patients) and had longer average BTKi line 1 durations (484 days vs 381 days). LLM-extracted-staged patients showed higher rates of later-line therapy and BTKi rechallenging relative to the unstaged remainder.
CONCLUSIONS: CLL disease stage is not a coded variable in claims or structured EHR but can be recovered from unstructured clinical documentation. LLM-extracted staging unlocks analyses entirely inaccessible from structured data alone, establishing a methodological case for its integration as a prerequisite for rigorous CLL RWE supporting HTA and coverage decisions.
Conference/Value in Health Info
2026-11, ISPOR Europe 2026, Vienna, Austria
Value in Health, Volume 29, Issue 12S
Code
CO67
Topic
Clinical Outcomes, Epidemiology & Public Health, Real World Data & Information Systems
Disease
Oncology