Forecasting pharmaceutical life cycles

This paper discusses the modelling and forecasting of pharmaceutical life cycles. Three different scenarios were found to exist when exploring the difference between the branded and generic life cycles. First after patent expiry, we examine the case where branded sales decline and the generic sales increase (branded then generic), once the patent associated with the branded drug has expired. Then irrespective of patent expiration we examine two further cases. The first is where branded sales are high and generic sales are low (high branded, low generic) and the second is where branded sales are low and generic sales are high (high generic, low branded). Understanding the patterns of brand decline (and the associated generic growth) is increasingly important because in a market worth over £7bn in the UK, the number of new ‘blockbuster’ drugs continues to decline. As a result pharmaceutical companies make efforts to extend the commercial life of their brands, and the ability to forecast is important in this regard. Second, this paper provides insights for effective governance because the use of a branded drug (when a generic is available) results in wasted resources. The pharmaceutical prescription data comes from a database known as JIGSAW. The prescription drugs that were modelled were those that had the highest number of prescriptions within the database. Six methods were then used to model and forecast the life cycles of these drugs. The models used were: Bass Diffusion Model, Repeat Purchase Diffusion Model (RPDM), and Naïve with and without drift, Exponential Smoothing and Moving Average models. Based on previous research it was expected that the more complex models would produce more accurate forecasts for the branded and generic life cycles than the simple benchmark models. The empirical evidence presented here suggests that the use of the Naïve model incorporating drift provided the most accurate and robust method of modelling both types of prescribing, with the more advanced models being less accurate for all three scenarios examined.