Constructing a Dynamic Budget Impact Model: An AI Pretraining Approach for a New Parkinson's Disease Drug
Author(s)
Nicholas Y. Chen, MS, Kae-Kuen Hu, PhD.
Professional Master's Program of Biotechnology Management, School of Professional Education and Continuing Studies, National Taiwan University, Taipei, Taiwan.
Professional Master's Program of Biotechnology Management, School of Professional Education and Continuing Studies, National Taiwan University, Taipei, Taiwan.
OBJECTIVES: Facing the dual challenges of Taiwan’s aging population and the financial strain of high-priced innovative drugs on its single-payer National Health Insurance (NHI) system—with over 99% population coverage and joint funding—this study aims to improve budget impact assessments for new drug reimbursement. We developed a dynamic Budget Impact Analysis (BIA) model tailored to Taiwan’s context, using a new Parkinson’s disease drug (“Drug N”) as a case.
METHODS: The BIA model dynamically integrates real-world NHI data, market analytics, and expert input. It employs machine learning to estimate penetration and substitution rates, incorporates both direct and indirect costs (including productivity loss, caregiving, and healthcare workforce), and supports multi-factor sensitivity and marginal budget impact analyses. The simulation results generated by this Taiwan-specific BIA model for Drug N were then used as a pretraining dataset for an AI module, enabling future AI-driven BIA predictions for other high-cost drugs.
RESULTS: A five-year empirical simulation shows that the dynamic BIA model reduces forecast errors by over 40% compared to current CDE approaches. Model predictions were highly consistent with original simulations (R²=0.99985; MAE=NT$42,932), confirming strong structural consistency and high learnability. Marginal budget impact analysis revealed moderate budget changes with incremental market share or price increases, while multi-scenario sensitivity analyses demonstrated robustness under varied assumptions. Incorporating indirect costs, Drug N’s annual net budget impact grows moderately (NT$8.7M-NT$20.2M), with societal savings providing financial buffers.
CONCLUSIONS: This dynamic BIA model provides a robust and adaptable foundation for evaluating new drugs’ financial and societal impacts in Taiwan’s NHI. By serving as a pretraining dataset for AI models, it enables more efficient and scalable budget impact predictions for future high-cost drugs, supporting value-based payment and sustainable health policy development.
METHODS: The BIA model dynamically integrates real-world NHI data, market analytics, and expert input. It employs machine learning to estimate penetration and substitution rates, incorporates both direct and indirect costs (including productivity loss, caregiving, and healthcare workforce), and supports multi-factor sensitivity and marginal budget impact analyses. The simulation results generated by this Taiwan-specific BIA model for Drug N were then used as a pretraining dataset for an AI module, enabling future AI-driven BIA predictions for other high-cost drugs.
RESULTS: A five-year empirical simulation shows that the dynamic BIA model reduces forecast errors by over 40% compared to current CDE approaches. Model predictions were highly consistent with original simulations (R²=0.99985; MAE=NT$42,932), confirming strong structural consistency and high learnability. Marginal budget impact analysis revealed moderate budget changes with incremental market share or price increases, while multi-scenario sensitivity analyses demonstrated robustness under varied assumptions. Incorporating indirect costs, Drug N’s annual net budget impact grows moderately (NT$8.7M-NT$20.2M), with societal savings providing financial buffers.
CONCLUSIONS: This dynamic BIA model provides a robust and adaptable foundation for evaluating new drugs’ financial and societal impacts in Taiwan’s NHI. By serving as a pretraining dataset for AI models, it enables more efficient and scalable budget impact predictions for future high-cost drugs, supporting value-based payment and sustainable health policy development.
Conference/Value in Health Info
2025-11, ISPOR Europe 2025, Glasgow, Scotland
Value in Health, Volume 28, Issue S2
Code
EE153
Topic
Economic Evaluation, Methodological & Statistical Research, Real World Data & Information Systems
Topic Subcategory
Budget Impact Analysis
Disease
Geriatrics, Neurological Disorders