Prediction of chronic migraine attacks using machine learning based on clinical and behavioral data

Abstract

Background: Chronic migraine is a disabling neurological condition whose unpredictable attacks compromise clinical management and patients’ quality of life. Evidence suggests that behavioral and clinical factors exhibit detectable patterns prior to the onset of attacks, enabling the application of artificial intelligence–based models for early prediction. Objective: To develop and evaluate a machine learning model capable of identifying patterns associated with the occurrence of migraine attacks within the subsequent 24 hours. Methods: A prospective observational study was conducted with 85 patients diagnosed according to current international criteria. For 90 days, participants recorded daily pain intensity, sleep duration, stress level, caffeine intake, physical activity, and the presence of prodromal symptoms using a digital application. Random forest and regularized logistic regression models were trained. Model performance was evaluated using cross-validation, accuracy, sensitivity, and specificity metrics. Results: The random forest model demonstrated the best performance, achieving an accuracy of 82%, sensitivity of 79%, and specificity of 84% in predicting attacks within the following 24 hours. Sleep deprivation, increased stress levels, and prodromal symptoms were the most relevant variables, and their combined presence increased the probability of an attack on the subsequent day by 3.4 times. Conclusion: Machine learning–based models were able to identify clinical and behavioral patterns associated with the onset of chronic migraine attacks, demonstrating potential for the development of early warning systems and personalized preventive strategies.