AUSTIN, TEX. – The findings could point to new therapeutic strategies, according to study author Ali Ezzati, MD.
“A lot of diagnostic criteria that we have in the migraine world come from consensus groups of experts, and based on their experience and available data. They classify different types of headache and then on top of that different types of migraine. Unfortunately, this type of classification does not necessarily lead to having very homogeneous groups,” said Dr. Ezzati, who presented the study at the annual meeting of the American Headache Society.
Migraines are generally categorized as episodic (0-14 headache days per month) or chronic (15 or more per month), or as with or without aura. But these broad categories fail to capture the true diversity of migraine, according to Dr. Ezzati, and this may contribute to the fact that response to migraine therapy hovers around 60%. “We feel that the key to improving therapeutic efficacy is to identify individuals who are more homogeneous, more similar to each other, so that when we give a treatment, it is specifically targeting the underlying pathophysiology that those people have,” said Dr. Ezzati, who is an associate professor of neurology and director of the neuroinformatics program at University of California, Irvine.
The analysis revealed some clinically interesting results, said Dr. Ezzati. “For example, allodynia is a symptom that is not particularly used for classification of different types of migraine. There was a specific group that was very high in allodynia, and they were not very responsive to treatments, so that might be a [group] that people have to focus on. Also, we talk a lot about comorbidities in migraine, but we don’t talk about how these comorbidities affect the therapeutic strategies and treatment response to specific medications. We showed that people who have depression are actually less responsive than other groups to treatments, especially prescription medications,” he said.
Machine learning reveals clusters
The researchers analyzed data from 4,423 patients drawn from the American Migraine Prevalence and Prevention Study, which was conducted every year between 2005 and 2009. They included adult patients who filled out surveys in both 2006 and 2007. The study population was 83.7% female and had a mean age of 46.8 years, and 6.4% had chronic migraine. The researchers then used a machine-learning based self-organizing map to group patients into similar clusters.
The algorithm produced five such groups: Cluster 1 had the lowest symptom severity, and 0.6% had chronic migraine. Cluster 2 had mild symptom severity with no chronic migraine. Cluster 3 had moderate symptom severity and a high prevalence of allodynia (88.5%, vs. 63.4% overall, P < .001) and no chronic migraine. Cluster 4 had a high frequency of depressive symptoms (63.1% vs. 19.8% overall, P < .001) and 5.2% had chronic migraine. Cluster 5 had frequent and severe migraines, and most (83.0%) had chronic migraine (P < .001).
There were some other broader trends. Triptans were more commonly used in clusters 2 (25.6%), 3 (27.9%), and 5 (28.0%), but less so in cluster 4 (17.1%; P < .001). Pain freedom at 2 hours was most common in cluster 1 (53.1%), followed by cluster 2 (46.4%), but was significantly less frequent in clusters 3 (32.2%), 4 (32.2%), and 5 (34.7%; P < .001).