Machine learning identifies factors most associated with seeking medical care for migraine: Results of the OVERCOME (US) study

Abstract

Objective: Utilize machine learning models to identify factors associated with seeking medical care for migraine.

Background: Migraine is a leading cause of disability worldwide, yet many people with migraine do not seek medical care.

Methods: The web-based survey, ObserVational survey of the Epidemiology, tReatment and Care Of MigrainE (US), annually recruited demographically representative samples of the US adult population (2018-2020). Respondents with active migraine were identified via a validated diagnostic questionnaire and/or a self-reported medical diagnosis of migraine, and were then asked if they had consulted a healthcare professional for their headaches in the previous 12 months (i.e., "seeking care"). This included in-person/telephone/or e-visit at Primary Care, Specialty Care, or Emergency/Urgent Care locations. Supervised machine learning (Random Forest) and Least Absolute Shrinkage and Selection Operator (LASSO) algorithms identified 13/54 sociodemographic and clinical factors most associated with seeking medical care for migraine. Random Forest models complex relationships (including interactions) between predictor variables and a response. LASSO is also an efficient feature selection algorithm. Linear models were used to determine the multivariable association of those factors with seeking care.

Results: Among 61,826 persons with migraine, the mean age was 41.7 years (±14.8) and 31,529/61,826 (51.0%) sought medical care for migraine in the previous 12 months. Of those seeking care for migraine, 23,106/31,529 (73.3%) were female, 21,320/31,529 (67.6%) were White, and 28,030/31,529 (88.9%) had health insurance. Severe interictal burden (assessed via the Migraine Interictal Burden Scale-4, MIBS-4) occurred in 52.8% (16,657/31,529) of those seeking care and in 23.1% (6991/30,297) of those not seeking care; similar patterns were observed for severe migraine-related disability (assessed via the Migraine Disability Assessment Scale, MIDAS) (36.7% [11,561/31,529] vs. 14.6% [4434/30,297]) and severe ictal cutaneous allodynia (assessed via the Allodynia Symptom Checklist, ASC-12) (21.0% [6614/31,529] vs. 7.4% [2230/30,297]). Severe interictal burden (vs. none, OR 2.64, 95% CI [2.5, 2.8]); severe migraine-related disability (vs. little/none, OR 2.2, 95% CI [2.0, 2.3]); and severe ictal allodynia (vs. none, OR 1.7, 95% CI [1.6, 1.8]) were strongly associated with seeking care for migraine.

Conclusions: Seeking medical care for migraine is associated with higher interictal burden, disability, and allodynia. These findings could support interventions to promote care-seeking among people with migraine, encourage assessment of these factors during consultation, and prioritize these domains in selecting treatments and measuring their benefits.

Keywords: OVERCOME; burden; disability; impact; medical care; migraine.