WHY ADVANCED PSORIATIC ARTHRITIS THERAPIES ARE STOPPED AND SWITCHED, BY DRUG CLASS: EVIDENCE FROM US REAL-WORLD CLINICAL NOTES
Author(s)
Karan Bhanot, PhD, Raghu R, PhD, Shrinal Patel, MD, Rahul K. Das, PhD.
Norstella, New York, NY, USA.
Norstella, New York, NY, USA.
OBJECTIVES: Advanced psoriatic arthritis (PsA) therapy is frequently stopped and switched, but the rationale resides in clinical notes. We characterized these reasons by drug class and tested their construct (class pharmacology and patient-reported disease activity) and convergent (pharmacy-fill corroboration) validity.
METHODS: In NorstellaLinQ — a US real-world dataset linking open claims, structured EHR, clinical notes, and pharmacy fills; data through December 2025 — human-in-the-loop large language model extraction of notes identified 83,176 patients with PsA (19.5% with no PsA code in claims or EHR). Among 53,356 (64.1%) advanced-therapy users (TNF, IL-17, IL-23, IL-12/23, JAK, PDE4 inhibitors, abatacept), 28,999 (54.4%) discontinued; reasons were classified as inefficacy, safety, access/cost, preference, or other. RAPID3 scores were extracted. Core fields were clinical-expert-validated at mean F1 ≥ 95%.
RESULTS: Across 42,120 documented discontinuations (one per patient per class), inefficacy predominated (~34%), followed by safety (~29%). Safety led only for apremilast (~42% vs ~26% inefficacy), intermediate for other classes (~23-32%), and lowest for IL-23 (~19%) and IL-12/23 (~16%) inhibitors, where inefficacy dominated (~39%, ~48%). Access/cost was heaviest for IL-17 inhibitors (~11%) versus orals (~6%). This by-class safety ordering recurred among 23,242 switchers (43.6%), apremilast highest (~29%), IL-23-pathway agents lowest. Most switches left a TNF inhibitor (~38%), restarting on TNF (~37%) or IL-17 (~28%); IL-23 leavers restarted on IL-17 (~35%) or TNF (~26%); IL-17 leavers on IL-17 (~28%) or TNF (~26%). Pharmacy fills corroborated this TNF-dominant, IL-pathway-second direction. RAPID3 (N=4,220) differed by stop reason: highest for inefficacy (median 9.8), intermediate for safety (9.0), lowest for preference or access (6.7, 7.0; Kruskal-Wallis p<0.001).
CONCLUSIONS: LLM-based extraction of clinical notes identified why advanced PsA therapies are stopped and switched, consistent with class pharmacology and patient-reported disease activity. Class-level patterns let payers quantify inefficacy- versus safety-led switching, informing formulary decisions. Reported shares are documented, approximate, cross-sectional lower bounds — not adverse-event incidence.
METHODS: In NorstellaLinQ — a US real-world dataset linking open claims, structured EHR, clinical notes, and pharmacy fills; data through December 2025 — human-in-the-loop large language model extraction of notes identified 83,176 patients with PsA (19.5% with no PsA code in claims or EHR). Among 53,356 (64.1%) advanced-therapy users (TNF, IL-17, IL-23, IL-12/23, JAK, PDE4 inhibitors, abatacept), 28,999 (54.4%) discontinued; reasons were classified as inefficacy, safety, access/cost, preference, or other. RAPID3 scores were extracted. Core fields were clinical-expert-validated at mean F1 ≥ 95%.
RESULTS: Across 42,120 documented discontinuations (one per patient per class), inefficacy predominated (~34%), followed by safety (~29%). Safety led only for apremilast (~42% vs ~26% inefficacy), intermediate for other classes (~23-32%), and lowest for IL-23 (~19%) and IL-12/23 (~16%) inhibitors, where inefficacy dominated (~39%, ~48%). Access/cost was heaviest for IL-17 inhibitors (~11%) versus orals (~6%). This by-class safety ordering recurred among 23,242 switchers (43.6%), apremilast highest (~29%), IL-23-pathway agents lowest. Most switches left a TNF inhibitor (~38%), restarting on TNF (~37%) or IL-17 (~28%); IL-23 leavers restarted on IL-17 (~35%) or TNF (~26%); IL-17 leavers on IL-17 (~28%) or TNF (~26%). Pharmacy fills corroborated this TNF-dominant, IL-pathway-second direction. RAPID3 (N=4,220) differed by stop reason: highest for inefficacy (median 9.8), intermediate for safety (9.0), lowest for preference or access (6.7, 7.0; Kruskal-Wallis p<0.001).
CONCLUSIONS: LLM-based extraction of clinical notes identified why advanced PsA therapies are stopped and switched, consistent with class pharmacology and patient-reported disease activity. Class-level patterns let payers quantify inefficacy- versus safety-led switching, informing formulary decisions. Reported shares are documented, approximate, cross-sectional lower bounds — not adverse-event incidence.
Conference/Value in Health Info
2026-11, ISPOR Europe 2026, Vienna, Austria
Value in Health, Volume 29, Issue 12S
Code
MSR35
Topic
Health Service Delivery & Process of Care, Methodological & Statistical Research
Topic Subcategory
Artificial Intelligence, Machine Learning, Predictive Analytics
Disease
Musculoskeletal Disorders (Arthritis, Bone Disorders, Osteoporosis, Other Musculoskeletal), Systemic Disorders/Conditions (Anesthesia, Auto-Immune Disorders (n.e.c.), Hematological Disorders (non-oncologic), Pain)