CLINICAL VALIDATION OF A MACHINE-LEARNING-ENRICHED IMMUNE-MEDIATED NECROTIZING MYOPATHY COHORT USING LINKED LABORATORY AND CLAIMS DATA
Author(s)
Parisa F. Asad, Sr., PhD1, Charlotte E. Ward, PhD2, Shreyas Jarmale, BS3, Andre Gladiator, PhD4;
1argenx BV, Basel (BS), Switzerland, 2ZS Associates, Boston, MA, USA, 3ZS Associates, Bangalore, India, 4argenx, Ghent, Belgium
1argenx BV, Basel (BS), Switzerland, 2ZS Associates, Boston, MA, USA, 3ZS Associates, Bangalore, India, 4argenx, Ghent, Belgium
OBJECTIVES: Real-world evidence generation in immune-mediated necrotizing myopathy (IMNM) is constrained by disease rarity and limited numbers of patients with laboratory confirmation and a specific diagnosis code. Machine-learning (ML) approaches may enable cohort enrichment; however, credibility of ML-derived populations must be demonstrated. This study assessed concordance of an ML-enriched IMNM cohort relative to a laboratory-confirmed reference cohort with longitudinal enrollment.
METHODS: A reference cohort of IMNM patients (N=36) was identified based on documented anti-HMGCR and/or anti-SRP antibody positivity and required ≥12 months of continuous enrollment before and after index. Laboratory data were derived from Quest Diagnostics, while diagnoses, procedures, treatments, and healthcare utilization were obtained from Komodo Health’s US claims database. A gradient-boosted decision tree model (LightGBM) was trained using claims-based features capturing demographics, diagnostic evaluation (e.g., electromyography, muscle biopsy), laboratory testing patterns (creatine kinase and aldolase test presence and frequency), treatment exposure (corticosteroids, immunosuppressive therapies, IVIG/SCIG), and healthcare utilization, emphasizing longitudinal care patterns. The model was applied to a large claims population; patients with predicted probability ≥0.99 were classified as ML-enriched IMNM (N=5,523). Analyses were descriptive and intended for cohort enrichment rather than diagnosis or prevalence estimation.
RESULTS: Model performance was strong (AUROC=0.914). The ML-enriched cohort demonstrated high concordance with the reference cohort. Muscle weakness was common (78% reference; 99% ML-enriched), as was muscle enzyme testing (89-100%). Neuromuscular diagnostic procedures were frequently observed, including electromyography (22% vs. 51%) and imaging (75% vs. 98%). Specialist involvement and physical therapy use were prevalent in both cohorts. Corticosteroid exposure occurred in 83% of reference patients and 95% of ML-enriched patients; IVIG use was observed in 11% and 20%, respectively.
CONCLUSIONS: A conservative ML-based approach identified an IMNM-enriched cohort mirroring a laboratory-confirmed population across clinical features, diagnostic evaluation, specialist care, and treatment patterns, supporting ML-based cohort enrichment for real-world research in rare diseases.
METHODS: A reference cohort of IMNM patients (N=36) was identified based on documented anti-HMGCR and/or anti-SRP antibody positivity and required ≥12 months of continuous enrollment before and after index. Laboratory data were derived from Quest Diagnostics, while diagnoses, procedures, treatments, and healthcare utilization were obtained from Komodo Health’s US claims database. A gradient-boosted decision tree model (LightGBM) was trained using claims-based features capturing demographics, diagnostic evaluation (e.g., electromyography, muscle biopsy), laboratory testing patterns (creatine kinase and aldolase test presence and frequency), treatment exposure (corticosteroids, immunosuppressive therapies, IVIG/SCIG), and healthcare utilization, emphasizing longitudinal care patterns. The model was applied to a large claims population; patients with predicted probability ≥0.99 were classified as ML-enriched IMNM (N=5,523). Analyses were descriptive and intended for cohort enrichment rather than diagnosis or prevalence estimation.
RESULTS: Model performance was strong (AUROC=0.914). The ML-enriched cohort demonstrated high concordance with the reference cohort. Muscle weakness was common (78% reference; 99% ML-enriched), as was muscle enzyme testing (89-100%). Neuromuscular diagnostic procedures were frequently observed, including electromyography (22% vs. 51%) and imaging (75% vs. 98%). Specialist involvement and physical therapy use were prevalent in both cohorts. Corticosteroid exposure occurred in 83% of reference patients and 95% of ML-enriched patients; IVIG use was observed in 11% and 20%, respectively.
CONCLUSIONS: A conservative ML-based approach identified an IMNM-enriched cohort mirroring a laboratory-confirmed population across clinical features, diagnostic evaluation, specialist care, and treatment patterns, supporting ML-based cohort enrichment for real-world research in rare diseases.
Conference/Value in Health Info
2026-05, ISPOR 2026, Philadelphia, PA, USA
Value in Health, Volume 29, Issue S6
Code
MSR23
Topic
Methodological & Statistical Research
Topic Subcategory
Artificial Intelligence, Machine Learning, Predictive Analytics
Disease
SDC: Musculoskeletal Disorders (Arthritis, Bone Disorders, Osteoporosis, Other Musculoskeletal), SDC: Rare & Orphan Diseases, SDC: Systemic Disorders/Conditions (Anesthesia, Auto-Immune Disorders (n.e.c.), Hematological Disorders (non-oncologic), Pain)