PO-70 | A nomogram for the prediction of response to anti-CGRP mAbs: the CGRP score

Marina Romozzi,¹ Ammar Lokhandwala,² Catello Vollono,¹ David García-Azorín,³ Giulia Vigani,⁴ Francesco De Cesaris,⁴ Claudia Altamura,⁵ Fabrizio Vernieri,⁵ Paolo Calabresi,¹ Sonia Di Tella,⁶ Luigi Francesco Iannone⁷ 1Dipartimento Universitario di Neuroscienze, Università Cattolica del Sacro Cuore, Fondazione Policlinico Universitario Agostino Gemelli IRCCS, Rome, Italy; 2Drexel University, Philadelphia, PA, USA; 3Hospital Universitario del Río Hortega, Valladolid, Headache Unit, Department of Neurology, Valladolid, Spain; 4Section of Clinical Pharmacology and Oncology, Department of Health Sciences, University of Florence, Florence, Italy; 5 Neurologia, Dipartimento di Medicina e Chirurgia, Università Campus Bio-Medico di Roma, Rome, Italy; 6Dipartimento di Psicologia, Università Cattolica del Sacro Cuore, Milan, Italy; 7Department of Biomedical, Metabolic and Neural Sciences, University of Modena and Reggio Emilia, Modena, Italy

Background: Real-world studies have explored potential predictors of response to anti-CGRP monoclonal antibodies, though results have remained inconsistent. Machine learning algorithms are becoming increasingly relevant in migraine research, offering a data-driven approach to identifying predictors of response to preventive treatments. To maximize their potential, a practical approach is needed to promote the use of these algorithms in research and, eventually, as supportive tools in clinical practice.

Methods: This prospective cohort study included adults with migraine treated with anti-CGRP mAbs (anti-ligand and receptor) at two headache centers. Responders were defined as patients achieving ≥50% reduction in monthly headache days (MHDs) at 12 months. A logistic regression model was trained (80%) and tested (20%) using 11 baseline variables, including age, sex, migraine subtype, medication overuse, MHDs, and disability scores. Model performance was evaluated using accuracy, precision, recall, and F1-score. A nomogram was created for future research and clinical application. The model was then validated against an external test cohort treated with anti-CGRP mAbs.

Results: Among 430 patients, 311 completed 12 months of treatment, with 236 (55.1%) classified as responders. The external cohort included 109 patients. The ML model achieved an overall accuracy of 70%, with strong performance in identifying "responders" (precision: 0.66, recall: 0.84, F1-score: 0.74). The model yielded predictions with an overall accuracy of 74% when tested against an external test cohort. Chronic migraine status, older age, and lower baseline MHDs were associated with higher response likelihood. Medication overuse and frequent analgesic use were negatively associated with response. The nomogram provided a clinically interpretable tool to estimate response probability, providing a total score named "CGRP Score" (CGRP mAbs, Global Response Prediction).

Conclusion: This ML-based predictive score achieved strong performance in identifying responders to anti-CGRP mAbs. The nomogram has the potential to be a practical, user-friendly tool for supporting clinical decision-making after validation.