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

Area of Science:

• Pharmaceutical Market Analysis
• Computational Epidemiology
• Health Economics

Background:

• Loss of Exclusivity (LOE) for branded drugs leads to rapid generic competition and market share decline.
• Existing forecasting models often use simple decay curves, facing challenges with data scarcity, low accuracy, and inability to predict share rebounds.
• Accurate post-LOE forecasting is crucial for strategic planning by both patent holders and generic manufacturers.

Purpose of the Study:

• To develop and evaluate advanced machine learning algorithms for forecasting brand share after Loss of Exclusivity (LOE).
• To compare the performance of various models, including neural networks and Random Forests, against traditional methods.
• To identify key drivers influencing post-LOE market dynamics and share fluctuations.

Main Methods:

• Assembled a 20-year panel dataset of chronic-disease LOE events in South Korea (Hypertension, Lipidemia, Diabetes).
• Evaluated a range of forecasting algorithms, from classical machine learning to neural networks (e.g., N-BEATS, Random Forest).
• Utilized SHAP analysis to interpret model predictions and identify significant influencing factors.

Main Results:

• The N-BEATS model achieved high accuracy in predicting absolute brand share (RMSE .034, MAPE .073).
• A Random Forest model demonstrated strong performance in predicting quarter-to-quarter share changes (RMSE .014, MAPE .147).
• The models successfully captured rare post-LOE share recovery events, outperforming traditional benchmarks.

Conclusions:

• Advanced machine learning models offer a superior, data-driven framework for forecasting post-LOE drug market share.
• The number of generics, time since LOE, and brand-holder partnerships with local distributors are key drivers of market dynamics.
• These findings enable more effective strategic alignment for lifecycle and market-access planning in the pharmaceutical industry.