OBJECTIVES: Current measures of clinical effectiveness of migraine prophylaxis are predominantly based on the reduction in frequency of migraine days (MDs) per 28 days. It is therefore important to accurately link economic and quality of life outcomes to reductions in MD frequency to model the cost-effectiveness of migraine prophylactics. Traditionally, headache day frequency has been characterized by Poisson, binomial and negative binomial distributions, and zero-inflated variants of these. All of these distributions are associated with theoretical and practical limitations. The beta-binomial distribution has been applied in other fields to model this type of count data. This analysis aims to compare the beta-binomial distribution to Poisson, binomial, negative binomial and zero-inflated negative binomial distributions in modeling MD frequency, using data from a phase II study of erenumab (NCT01952574). METHODS: The study compared erenumab 70mg against placebo in patients with episodic migraine. Data on the frequency of MDs per 28 days were used to fit Poisson, binomial, negative binomial, zero-inflated negative binomial and beta-binomial distributions. The average root mean squared errors (RMSE) across all 28-day observation periods was used to quantify the deviation of each distribution from the trial observations. RESULTS: Data from 257 patients were available from the study, with a maximum follow-up of 64 weeks. The average RMSEs were 0.239, 0.257, 0.104, 0.081 and 0.098 for the Poisson, binomial, negative binomial, zero-inflated negative binomial and beta-binomial distributions, respectively, for the erenumab group, and 0.201, 0.221, 0.089, 0.081 and 0.089 for the placebo group. CONCLUSIONS: Based on the RMSE estimates, the negative binomial, zero-inflated negative binomial and beta-binomial provide comparable fits to the trial observations, and fit better than the Poisson and binomial fits. This was observed in both the erenumab and placebo groups. The results suggest that the beta-binomial may be an alternative choice of distribution for modeling MD frequency in migraine populations.