“The reality is that you have to work with the data you have and find the best models that fit that type of data.” Each model will not suit every data set. However, this also raises the issue of model dependability and validation.”
According to studies, the clinical symptoms of migraine vary significantly across and among persons, from attack to attack, and with time. Between 2005 and 2009, the American Migraine Prevalence and Prevention research (AMPP), a longitudinal, population-based research of persons with self-reported headache, collected data on demographics, migraine characteristics, disability, and depression. Investigators presented findings from the study, detecting naturally occurring clusters utilizing a data-driven method, at the recently ended American Headache Society (AHS) Annual Meeting, held June 15-18 in Austin, Texas.

The study, led by Ali Ezzati, MD, employed a self-organizing map (SOM), an unsupervised machine learning approach that maintains topological data properties and is based on competitive learning of a 2-layer artificial neural network. The sample included 4423 people, the majority of whom were female (83.5%) and had an average age of 46.8 years. The research found five major migraine clusters with varied degrees of severity based on migraine symptom severity score, cutaneous allodynia, monthly headache days, Migraine Disability Assessment Scale, and Patient Health Questionnaire-9 results.

Following his presentation, Ezzati, head of the Neuroinformatics Program at the University of California, Irvine, spoke with NeurologyLive® about the many big data techniques that are being used in clinical trials and if some give larger benefits than others. He also dissected the study’s distinct clusters and the criteria that helped define patient groups.