Patient classification in episodic migraine based on data-driven pain curve dynamics

Abstract

Purpose Migraine is characterized by heterogeneous manifestations, with pain being its most disabling feature. Although the International Classification of Headache Disorders, 3rd edition (ICHD-3), defines diagnostic criteria for migraine, it does not incorporate temporal dynamics of pain. While studies have described attack-level pain trajectory patterns, the translation of these dynamic features into patient-level stratification remains unexplored. This study aimed to classify individuals with episodic migraine according to pain trajectories and to characterise the clinical features of the resulting groups. Patients and methods Patients fulfilling ICHD-3 criteria for episodic migraine participated in a prospective real-time study using a smartphone application to record migraine attacks. Pain onset, duration, intensity, and associated premonitory, concomitant, and postdromic symptoms were documented. A total of 344 migraine attacks from 51 patients were analyzed. Attack-level pain trajectories identified in earlier work were used to construct patient-level feature vectors representing the proportional distribution of attack types. These vectors were clustered using a K-means algorithm to identify patient subgroups. Results Fifty-one patients (mean age: 38.8 ± 10.9 years; 90.2% women) recorded 344 migraine attacks. Five patient profiles were identified from the combination of four previously established pain trajectories: high-intensity, mixed pattern, acute-onset, prolonged-intense, and low-intensity. The mixed profile was the most frequent (39.2%), while the remaining profiles showed a clearer predominance of a specific attack type. Most conventional demographic and clinical variables, including age, sex, baseline diagnosis, and monthly migraine frequency, did not differ across profiles. Concomitant symptom burden differed across profiles, with significant differences in nausea, photophobia, and osmophobia after accounting for within-patient clustering and multiple comparisons. The low-intensity profile showed a lower premonitory symptom burden than the high-intensity, mixed, and prolonged-intense profiles, whereas the prolonged-intense profile showed a greater postdromic symptom burden than the low-intensity profile. Conclusion Our findings suggest that the way migraine pain evolves over time may help identify clinically distinct patient profiles. Considering pain trajectory adds a dynamic dimension to migraine phenotyping that is not captured by traditional diagnostic criteria and may contribute to improved phenotyping and hypothesis generation for personalized therapeutic strategies. Further studies in larger cohorts are needed to clarify its potential clinical relevance.