P093 | GAINER-AI: PREDICTING ACUTE RIMEGEPANT RESPONSE IN MIGRAINE USING MACHINE LEARNING IN A MULTICENTER REAL-WORLD COHORT

M. Romozzi^1|2, A. Lokhawandala³, G. Vaghi^4|5, G. Sebastianelli⁶, R. Ornello⁷, F. Casillo⁶, A. Russo⁸, M. Silvestro⁸, F. Pistoia⁹, G. Saporito⁹, G. Dalla Volta¹⁰, M. Bolchini¹⁰, A. Burgalassi¹¹, G. Vigani¹¹, D. Antonio Montisano¹², L. Grazzi¹², M. Pia Prudenzano¹³, D. Mascarella¹⁴, E. Mampreso¹⁵, G. Avino¹⁶, M. Valente¹⁷, G. Carlucci¹⁸, S. Guerzoni¹⁹, S. Battistini²⁰, A. Granato²¹, E. Maria Piella²², I. Rainero²², M. Albanese²³, M. Trimboli²⁴, A. Doretti²⁵, F. De Santis⁷, S. Sacco⁷, C. Tassorelli^4|5, S. Di Tella²⁶, C. Vollono^1|2, P. Calabresi^1|2, R. De Icco^4|5, L. Francesco Iannone²⁷ | ¹Dipartimento Universitario di Neuroscienze, Università Cattolica del Sacro Cuore, Rome, Italy; ²Neurologia, Dipartimento di neuroscienze, Organi di Senso e Torace, Fondazione Policlinico Universitario Agostino Gemelli IRCCS, Rome, Italy; ³Drexel University, Philadelphia, USA; ⁴Department of Brain and Behavioral Sciences, University of Pavia, Italy; ⁵Headache Science and Neurorehabilitation Unit, IRCCS Mondino Foundation, Pavia, Italy; ⁶Department of Medico-Surgical Sciences and Biotechnologies, Sapienza University of Rome Polo Pontino, Latina, Italy; ⁷Department of Biotechnological and Applied Clinical Sciences, University of L’Aquila, Italy; ⁸Headache Centre, Department of Advanced Medical and Surgical Sciences, University of Campania "Luigi Vanvitelli", Naples, Italy; ⁹Department of Biotechnological and Applied Clinical Sciences, University of L'Aquila, Italy; ¹⁰Headache Center of Clinical Neurology of Istituto Clinico Città Di Brescia, Italy; ¹¹Department of Health Sciences, University of Florence, Italy; ¹²Neuroalgology Unit and Headache Center, Fondazione IRCCS Istituto Neurologico Carlo Besta, Milan, Italy; ¹³Headache Center, Department of Basic Medical Sciences, Neurosciences and Sense Organs, University of Bari, Italy; ¹⁴IRCCS Istituto Di Scienze Neurologiche Di Bologna, Programma Cefalee e Algie Facciali, Italy; ¹⁵Headache Centre, Neurology - Euganea, Health Unit, Padua, Italy; ¹⁶Neurology Unit, Ospedale Santo Stefano, USL Toscana Centro, Prato, Italy; ¹⁷Clinical Neurology, Azienda Sanitaria Universitaria Friuli Centrale, Presidio Ospedaliero Santa Maria Della Misericordia, Udine, Italy; ¹⁸NEUROFARBA Department, University of Florence - AOU Careggi, Italy; ¹⁹Digital and Predictive Medicine, Pharmacology, Clinical Metabolic Toxicology-Headache Center and Drug Abuse, Laboratory of Clinical Pharmacology and Pharmacogenomics, AOU Policlinico, Modena, Italy; ²⁰Department of Medical, Surgical and Neurological Sciences, University of Siena, Italy; ²¹Headache Centre, Clinical Unit of Neurology, School of Neurology, Department of Medicine, Surgery and Health Sciences, University Hospital and Health Services of Trieste, ASUGI, University of Trieste, Italy; ²²Department of Neurosciences “Rita Levi Montalcini”, University of Turin, Italy; ²³Unit of Neurology, Department of Systems Medicine, Tor Vergata University, Rome, Italy; Unit of Neurology - IRCCS Neuromed, Pozzilli (IS), Italy; ²⁴Headache Centre, Neurology Unit, AOU Renato Dulbecco, Catanzaro, Italy; ²⁵Department of Neurology- Laboratory of Neuroscience, IRCCS Istituto Auxologico Italiano, Milan, Italy; ²⁶Dipartimento di Psicologia, Università Cattolica del Sacro Cuore, Milan, Italy; ²⁷Department of Biomedical, Metabolic and Neural Sciences, University of Modena and Reggio Emilia, Italy

Abstract

Background. Rimegepant is the first calcitonin gene-related peptide (CGRP) receptor antagonist approved for both acute and preventive migraine treatment. Despite demonstrated efficacy in randomized trials and real-world studies, predictors of individual response remain poorly understood. Machine learning (ML) approaches may facilitate personalized treatment selection by identifying clinical features associated with treatment outcomes.
Objectives. To develop and evaluate an ML-based model to predict pain freedom 2 hours after acute rimegepant treatment and to create a web-based prediction tool using routinely available clinical variables.
Methods. We conducted a post hoc analysis of the prospective, multicenter, real-world GAINER study. Adults with episodic or chronic migraine treated up to four migraine attacks with rimegepant 75 mg orally disintegrating tablets. The primary outcome was pain freedom at 2 hours after treatment during the first treated attack. Multiple ML algorithms were evaluated, and a Random Forest model was selected based on performance. Data were randomly split into training (80%) and test (20%) sets. Feature selection was performed using sequential forward selection. Model explainability was assessed through SHapley Additive exPlanations (SHAP) analysis.
Results. A total of 191 participants (80.1% female; mean age 44.8 years) treated 377 migraine attacks. Overall, 44.6% of attacks achieved pain freedom at 2 hours, with higher response rates in episodic than chronic migraine (50.5% vs. 28.4%, p=0.001). The final Random Forest model retained 11 predictors and achieved a weighted F1-score of 0.671 in the independent test set. Monthly headache days (MHD) emerged as the most influential predictor, followed by concomitant anti-CGRP monoclonal antibody treatment, age, age at migraine onset, medication overuse headache, and monthly acute medication use. Higher MHD and older age were associated with lower probabilities of pain freedom. Based on the final model, a freely accessible web-based prediction tool was developed.
Conclusions. This study represents the first application of ML methods to predict response to any acute migraine treatment. Using readily available clinical variables, the GAINER-AI model demonstrated moderate predictive performance and enabled the development of a user-friendly web application. Although external validation is required, these findings support the potential of ML-based tools to advance personalized acute migraine management.

Roberto De Icco, Luigi Francesco Iannone, Marina Romozzi on behalf of the Italian Headache Registry (RICe) Study Group.