The Effect of Weather Conditions on Demand Forecasting of Mi | 99629

Health Economics & Outcome Research: Open Access

ISSN - 2471-268X


The Effect of Weather Conditions on Demand Forecasting of Migraine Medications in a Specialty Pharmacy Applying ARIMA and VARMA Models

Lucas Scharf da Costa*, Alex C. Lin, Justin Campbell, Mina K. Gabriel and Andrew M. Ferguson

Background: Demand forecasting in a specialty pharmacy is a challenge that requires a data-driven decision-making process to improve its procedures. Patients with migraine are sensitive to weather conditions, but it is still unclear how it can affect the demand for migraine medications in pharmacies. The objective of the study is to apply ARIMA (Autoregressive Integrated Moving Average) and VARMA (Vector Autoregressive Moving Average) analytical methods to the demand forecasting of four most-prescribed migraine medications in a specialty pharmacy and to assess the impact of weather conditions in the demand forecasting of those medications.
Methods and Findings: This study was a collaboration between the University Of Cincinnati Medical Center, LLC (UC Health Specialty Pharmacy) and the James L. Winkle College of Pharmacy, both located in Cincinnati, Ohio. The pharmacy provided 26 months of pre-recorded actual sales data of Aimovig, Ajovy, Emgality, and Nurtec ODT, representing a total of 1,043 patients ordering migraine medications from the UC Health Specialty Pharmacy. After preprocessing the data, two variables were considered for the forecasting model: demand for each day and the date of each purchase by the pharmacy. For each medication, an approximate total of 800 data points were used to develop the forecast model. Weather conditions variables were added to each medication model to assess its effect on their demand. Then, a comparison was performed between the forecasting models with and without weather conditions as potential factors that could influence the demand for each medication. For the ARIMA models, only the demand was considered as the response variable while for the VARMA models, both demand and weather conditions variables were considered as factors. For MAPE and RMSE accuracy metrics, the lower the value, the more accurate the model. The ARIMA was determined to be the best forecast model for Emgality, Aimovig, and Nurtec ODT. However, VARMA was the best forecast model for Ajovy, indicating that weather conditions may affect the demand forecasting of this medication. Both models (ARIMA and VARMA) for all four medications in this study were considered either very accurate (MAPE<10) or a good predictor (MAPE<20). This is especially true for the ARIMA models for Aimovig, Emgality, and Nurtec ODT and VARMA models for Emgality and Nurtec ODT (MAPE less than 6).
Conclusions: The study showed that weather conditions had a significant effect on Ajovy’s demand during the 26-month period. When forecasting its demand, the model had higher accuracy when compared to forecasting its demand without weather conditions as a factor. On the other hand, weather conditions had no significant effect on the demand forecasting models for Aimovig, Emgality, and Nurtec ODT. Those models had higher accuracy when their demands were forecasted without weather conditions as factors.