OBJECTIVES: The study aimed to compare the data fit of different parametric distributions to simulate migraine headache days (MHD) in patients with chronic migraine, as observed in the REGAIN (NCT02614261) study, using the approach that Porter et al. (2016) proposed for episodic migraine.

METHODS: Patient level data from one Phase 3, randomized, double-blind, placebo-controlled study in chronic migraine (REGAIN) was used. For every month and for each treatment group, the raw MHD data was plotted as bar charts with the fitted Poisson, negative binomial and binomial distributions. The mean MHD per month and per treatment group was used to simulate a negative binomial distribution (NBD) using the function to estimate the dispersion parameter that was originally suggested by Porter et al. (2016) for episodic migraine datasets (dispersion=6.15943-2.59754*mean+0.34453*mean). Fitted distributions were compared with the raw data using visual inspection of the plots and goodness of fit parameters. Then, regression models were computed to predict the dispersion parameter from the NBD as a function of the MHD mean in chronic migraine.

RESULTS: The fitted NBD was consistently better than the fitted Poisson and the binomial distributions at each month and for each treatment group. The simulated NBD using Porters equation for the dispersion has a poor fit when considering chronic migraine patients. Regression models were undertaken to predict the dispersion from the MHD mean and findings will be presented at the congress.

CONCLUSIONS: A NBD is a good fit for the MHD distribution in patients with chronic migraine as observed in the REGAIN study but not as per Porter equation. A function to estimate the dispersion parameter from the MHD mean is provided for chronic migraine. When no patient level data is available, this approach allows to simulate MHD distribution in patient with chronic migraine based only on the MHD mean.