EXTRACTING PATIENT-CENTRIC DISEASE PROGRESSION INSIGHTS FROM NORSTELLALINQ CLINICAL NOTES: MG-ADL SCORES IN MYASTHENIA GRAVIS USING UNSTRUCTURED REAL-WORLD DATA
Author(s)
Becky Hollenberg, MPH1, ilan behm, MPH2, Daniel Park, PhD1, Rahul Das, PhD1, Allison Perry, PhD1.
1Norstella, New York, NY, USA, 2Norstella, Englewood, CO, USA.
1Norstella, New York, NY, USA, 2Norstella, Englewood, CO, USA.
OBJECTIVES: Myasthenia gravis (MG) is a chronic neuromuscular disease with meaningful variability in functional burden. The Myasthenia Gravis Activities of Daily Living (MG-ADL) scale is a validated, patient-centric instrument for tracking disease severity and progression; however, MG-ADL data are rarely captured in structured fields within electronic health records (EHR). This study aimed to demonstrate that large language model (LLM)-based extraction, with human-in-the-loop validation, of MG-ADL scores from NorstellaLinQ Clinical Notes can systematically characterize longitudinal disease trajectories from routine clinical documentation, enabling quantitative real-world outcomes research.
METHODS: Patients with MG were identified from NorstellaLinQ Clinical Notes using an LLM with human-in-the-loop validation. Using the same approach, MG-ADL total and domain scores were then extracted from clinical notes, and validated following clinician review. For each patient, extracted MG-ADL total scores were ordered chronologically, and disease trajectories were classified by fitting the trend across all documented MG-ADL assessments over time.
RESULTS: Among 4,888 patients with LLM-identified MG, 4,864 (99.5%) had at least one MG-ADL score extracted from clinical notes, and 4,468 (91.4%) had a documented MG-ADL total score. Of patients with extractable MG-ADL total scores, 2,696 (60.3%) had two or more longitudinal timepoints, with a mean follow-up of 1.4 years. Based on the trend across all documented assessments, 504 (11.4%) demonstrated sustained improvement, 1,291 (29.0%) remained stable, 607 (13.4%) showed a relapsing-remitting pattern, and 292 (6.6%) demonstrated progressive worsening; 1,770 (39.8%) had only a single documented assessment. Median baseline MG-ADL total score was 4 (mean 5.0, SD 5.3).
CONCLUSIONS: LLM-assisted extraction with human-in-the-loop validation enables scalable capture of longitudinal MG-ADL assessments from unstructured clinical documentation. These findings demonstrate the feasibility of generating patient-reported outcome measures from clinical notes, expanding opportunities for longitudinal real-world studies in MG where structured outcome data are otherwise sparse. Similar approaches may be applicable to other neurological diseases with clinically documented functional assessments.
METHODS: Patients with MG were identified from NorstellaLinQ Clinical Notes using an LLM with human-in-the-loop validation. Using the same approach, MG-ADL total and domain scores were then extracted from clinical notes, and validated following clinician review. For each patient, extracted MG-ADL total scores were ordered chronologically, and disease trajectories were classified by fitting the trend across all documented MG-ADL assessments over time.
RESULTS: Among 4,888 patients with LLM-identified MG, 4,864 (99.5%) had at least one MG-ADL score extracted from clinical notes, and 4,468 (91.4%) had a documented MG-ADL total score. Of patients with extractable MG-ADL total scores, 2,696 (60.3%) had two or more longitudinal timepoints, with a mean follow-up of 1.4 years. Based on the trend across all documented assessments, 504 (11.4%) demonstrated sustained improvement, 1,291 (29.0%) remained stable, 607 (13.4%) showed a relapsing-remitting pattern, and 292 (6.6%) demonstrated progressive worsening; 1,770 (39.8%) had only a single documented assessment. Median baseline MG-ADL total score was 4 (mean 5.0, SD 5.3).
CONCLUSIONS: LLM-assisted extraction with human-in-the-loop validation enables scalable capture of longitudinal MG-ADL assessments from unstructured clinical documentation. These findings demonstrate the feasibility of generating patient-reported outcome measures from clinical notes, expanding opportunities for longitudinal real-world studies in MG where structured outcome data are otherwise sparse. Similar approaches may be applicable to other neurological diseases with clinically documented functional assessments.
Conference/Value in Health Info
2026-11, ISPOR Europe 2026, Vienna, Austria
Value in Health, Volume 29, Issue 12S
Code
RWD90
Topic
Clinical Outcomes, Patient-Centered Research, Real World Data & Information Systems
Topic Subcategory
Data Protection, Integrity, & Quality Assurance
Disease
Neurological Disorders, Rare & Orphan Diseases