Improving the classification of migraine subtypes: An empirical approach based on factor mixture models in the american migraine prevalence and prevention (AMPP) study

Objectives Refine the classification of migraine subtypes by applying factor mixture models (FMM) to a large population sample of people with headache. Background Current classification of primary headache disorders is symptom-based and uses somewhat arbitrary boundaries developed by expert consensus. Symptom profiles and headache frequency are used to distinguish among probable migraine (PM), episodic migraine (EM), high-frequency episodic migraine (HFEM), and chronic migraine (CM). Herein, we used statistical approaches to parse the heterogeneity in the broad group of persons with migraine and test the hypothesis that the groups that emerge differ in prognosis. Methods The American Migraine Prevalence and Prevention study mailed surveys to a sample of 120,000 US households selected to represent the US population in 2004. Follow-up surveys were sent to a random sample of 24,000 respondents with "severe headache" on an annual basis from 2005 to 2009. People meeting International Classification of Headache Disorders, Second Edition, criteria for migraine were classified as EM (<15 headache days/month) and CM (≥15 headache days/month) based on modified Silberstein-Lipton criteria. The EM group was subdivided into HFEM (10 to 14 headache days/month) and low-frequency episodic migraine (LFEM; <10 headache days/month). Factor mixture models (FMM) identified 5 subgroups of migraine (taxa) using data from the 2005 survey on the severity of migraine symptoms, average migraine pain intensity, headache-related disability, cutaneous allodynia and depression, as well as monthly headache and migraine frequency as determinants of class membership. We assessed the validity of these taxa by examining the distribution of clinical diagnoses at cross-section and the rate of CM onset within these groups. Results Data from the 2005 American Migraine Prevalence and Prevention survey were used for the FMM and data from the 2006-2009 surveys were used to assess prognosis of groups defined based on FMM. In total, 12,860 participants were eligible for classification analysis, including 10,162 with LFEM and 601 with HFEM, 1302 with probable migraine, and 795 with CM. Of these, 3152 (24.5%), 1076 (8.4%), 3896 (30.3%), 2251 (17.5%), and 2485 (19.3%) were assigned to Taxons 1, 2, 3, 4, and 5, respectively. Overall, there was a strong association between taxon assignment and clinical diagnosis. As the most prevalent disorder in the sample, EM was the largest contributor to each of the 5 taxa, constituting more than 80% of each group other than Taxon 2. Taxon 2 was enriched with the most severe spectrum of migraine including the highest concentrations of CM (28.4%) and HFEM (22.6%), whereas Taxon 5 represented the least severe end of the migraine spectrum including the lowest concentrations of CM (0%) and HFEM (0.08%). Validity of taxon assignment was tested by the ability of taxon membership to predict clinical course. For Taxon 2, 22% of those free of CM at baseline developed it. For Taxon 5, less than 2% of CM-free Taxon 5 members developed it. Conclusion Statistically based classification using FMM extends traditional clinical syndrome-based diagnosis. FMM can serve as an important tool to parse phenotypic heterogeneity and identify natural migraine subgroups. This approach may improve our ability to diagnosis migraine, to select initial therapy, to predict prognosis, and to discover biomarkers and genes.