The Impact of Changes to the Payment Restrictions of the National Reimbursement Drug List in China: An Empirical Study Employing a Stacked Difference-in-Differences Approach

Background: Since 2004, China’s National Reimbursement Drug List has implemented a payment restriction management system for certain drugs, to safeguard the fund and avoid unreasonable medication. With the development of medical technology and changes in clinical needs, the fairness and rationality of payment restrictions have come under scrutiny. In recent years, the National Healthcare Security Administration has gradually lifted the payment restrictions for some drugs during dynamic adjustments to the National Reimbursement Drug List; however, the net effect of this policy adjustment still lacks rigorous causal evidence. Methods: This study used panel data of 328 drugs experiencing the lifting of payment restrictions from 2018 to 2024 and employed a stacked difference-in-differences method to assess causal effects. The logarithm of annual usage and sales were used as outcome variables. Parallel trend tests were conducted using the event study method, and robustness tests were performed using the Callaway and Sant’Anna estimator. Results: Baseline regression showed that after the lifting of payment restrictions, the average annual usage of drugs increased by 60.2% and the average annual sales increased by 180.1%. However, the alternative CS estimator yields large, statistically significant negative estimates, with all estimable cohorts negative (2020: −2.519; 2023: −0.805). The CS estimator automatically excludes the 2024 cohort due to data limitations, while Stacked DID includes all 328 drugs. Conclusions: While the evidence suggests that lifting payment restrictions did have an influence on drug utilization and sales, the data do not support a single causal conclusion about whether this influence increases or decreases these outcomes. The divergence between the two estimators reflects differences in sample composition and comparison-group construction: the CS estimator excludes the 2024 cohort, whereas the Stacked DID estimator includes it. Because the two estimates are not computed on a common sample, the divergence cannot be resolved by reweighting; this indicates that the estimated average effect is sensitive to these methodological choices. Policy implications must be conditional on this uncertainty. Cohort-specific CS decomposition reveals that while all cohorts exhibit negative effects, the magnitude decreases a convergence toward positive territory over time.