Multidimensional Bayesian Classifier for Predicting the Multi-stage Patient's Response to the BoNT-A Treatment for Migraine

Abstract

Since other treatments often do not work for treating migraine headaches, Onabotulinumtoxin-A, or BoNT-A, has gained a lot of popularity, especially in chronic migraines. The treatment consists of multiple sessions of drug injections. At the moment, it's unclear why BoNT-A therapy produces a beneficial reaction. In order to address this issue, the present work explores the use of Multidimensional Bayesian Classifiers (MBC) for training a multi-stage prediction model. It is carried out in a realistic setting by considering retrospective data from migraine patients receiving treatment with BoNT-A. As far as we are aware, there are no known studies using MBC in this domain. The model has achieved an average accuracy of 79.45%, 82.57%, and 77.35 % when predicting responses to the <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$1^{st}, 2^{nd}$</tex>, and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$3^{rd}$</tex> stages of treatment, respectively. It has also achieved similar values of sensitivity and specificity, enabling medical professionals to get a panoramic prediction of a patient's reaction to therapy and base their decisions accordingly. When looking at the prediction models, some clinical features have been identified as important, such as the number of days with a headache, the anesthetic blockade of the greater occipital nerve (GON), and others. Doctors have also described these features as significant aspects.