Predicting brand share after LOE in chronic disease market using machine learning.

Loss of Exclusivity (LOE) marks a turning point for brand-name drugs, triggering rapid share erosion as generics enter the market. Yet existing forecasting methods-often based on simple parametric decay curves-struggle with data scarcity, suffer from low predictive accuracy, and fail to capture counter-trend share rebounds. In this study, we assemble a 20-year panel of chronic-disease LOE events in South Korea (Hypertension, Lipidemia, Diabetes) and evaluate a range of forecasting algorithms, from classical machine learning to cutting-edge neural networks. Our best-in-class N-BEATS model predicts absolute brand share with an RMSE of .034 and MAPE of .073, while a Random Forest achieves an RMSE of .014 and MAPE of .147 for quarter-to-quarter changes. Notably, our approach successfully captures rare post-LOE share recoveries that traditional parametric benchmarks miss. SHAP analysis reveals that the number of generics is the dominant driver of absolute share levels, whereas time since LOE most strongly influences short-term fluctuations; additionally, brand-holder partnerships with local distributors meaningfully affect both outcomes. These findings lay the groundwork for a data-driven, practical forecasting framework-enabling patent holders and generic manufacturers to align lifecycle and market-access strategies with the key levers of post-LOE dynamics-and point toward even more robust insights as larger LOE datasets become available in future research.