Personalized machine learning prediction of next-day migraine persistence using digital headache diary data

Area of Science:

• Neurology
• Data Science
• Computational Medicine

Background:

• Migraine is a disabling neurological disorder with significant variability in attack duration.
• Predicting the persistence of a migraine attack into the next day is clinically challenging.

Purpose of the Study:

• To evaluate the feasibility and predictive performance of personalized machine learning models.
• To predict next-day migraine persistence using longitudinal digital headache diary data.

Main Methods:

• Prospective longitudinal cohort study involving 25 patients with 5-14 monthly headache days.
• Development of personalized and generalized models using k-nearest neighbors (KNN), support vector machine, random forest, and eXtreme Gradient Boosting (XGBoost).
• Model performance assessed using discrimination (AUC), calibration (Brier score), and sensitivity analyses.

Main Results:

• The personalized KNN model demonstrated superior predictive performance (AUC 0.83 ± 0.09) compared to the generalized KNN model (AUC 0.63 ± 0.04).
• Personalized KNN outperformed other personalized models (SVM, Random Forest, XGBoost) with significant ΔAUC values.
• The personalized KNN model showed good calibration (Brier score 0.13 ± 0.04) and was most sensitive to pain intensity, menstrual cycle, medication effectiveness, and pain location.

Conclusions:

• Personalized machine learning models utilizing digital headache diary data show feasibility in predicting next-day migraine persistence.
• Further external validation studies are required to confirm the generalizability and clinical applicability of these predictive models.