Procurement data were obtained from the CMEI Network, one of the most extensive and authoritative pharmaceutical information platforms in China, which covers more than 1,500 member hospitals across 31 provinces, municipalities, or autonomous regions, primarily consisting of secondary and tertiary public hospitals. These facilities include approximately 55% of all tertiary hospitals nationwide (25), which are the principal providers of oncology diagnosis and treatment. In China, specialized cancer treatment services are restricted to secondary and tertiary hospitals (26). Consequently, despite the underrepresentation of primary and private hospitals, the data remain highly representative of the principal settings for cancer care delivery. A potential selection bias should be acknowledged, as patients treated in lower-level hospitals may not have been adequately represented. Consequently, the generalizability of findings may be primarily confined to the public tertiary system rather than the entire healthcare system. The credibility of this dataset is substantiated by its widespread adoption in peer-reviewed publications within leading international journals (19, 27, 28). For this study, we utilized monthly averages of drug expenditure and DDDs for each medication from January 2015 to December 2019, forming a continuous time series dataset.

Two primary outcome measures were analyzed to assess drug utilization:

Defined Daily Doses (DDDs): As recommended by the World Health Organization (WHO), DDDs constitutes a standardized unit of measurement for drug utilization (29), independent of pricing, packaging, or dosage variations. This metric thereby provides an objective reflection of clinical prescription patterns and therapeutic trends. Higher DDDs indicate increased medication consumption within the studied population.

Expenditure: Monthly drug expenditures (in Chinese Yuan, CNY) were recorded at the hospital procurement level and were not adjusted for inflation to reflect real-world purchasing behavior during the policy intervention. The study period (2015–2019) was of limited duration, during which national Consumer Price Index (CPI) fluctuations remained marginal. Consequently, the utilization of nominal values is unlikely to materially bias the interpretation of expenditure trends. Furthermore, given that the substantial growth in DDDs occurred alongside only moderate increases in nominal costs, the conclusion regarding improved affordability remains robust without formal inflation adjustment.

The procurement data obtained from CMEI undergo multi-stage validation by pharmacists, regulatory experts, and technical staff prior to release. The database aligns all records with a standardized national drug information reference library to ensure uniformity in drug nomenclature, strength specifications, dosage formulations, and packaging sizes.

In this study, we included 887 public hospitals (241 secondary and 646 tertiary) with continuous monthly procurement records for 18 negotiated anticancer drugs and six comparison drugs from January 2015 to December 2019. DDDs were obtained directly from the CMEI database and calculated according to WHO ATC/DDD adult maintenance doses, which inherently accounted for variations in dosage strengths and package sizes. Expenditures were defined as the total monthly procurement cost in nominal CNY; no recalculation of per-DDD pricing was required.

The data exhibited a high level of completeness, with no missing monthly records for the majority of drugs. The sole exception was Chidamide, which presented three non-consecutive months of missing data during the 12-month pre-policy period; these instances were coded as zero procurement quantities and included in the analysis.

As the data represent aggregated procurement volumes across 887 hospitals, as opposed to sampled estimates, descriptive figures (Figures 1–2) are presented without confidence intervals. Statistical inference and confidence intervals are derived from DID and ITS regression models.

The NRDLN policy was officially implemented in July 2017 (month 31 relative to the study’s initiation). Accordingly, we defined the following groups: (a) Intervention group: 15 Western anticancer drugs included in the 2017 NRDLN. (b) Comparison group: Six similar-targeted anticancer drugs excluded from the negotiation (e.g., Cetuximab, Sunitinib). Furthermore, we performed pairwise comparisons between specifically matched drug pairs, such as Bevacizumab (negotiated) versus Cetuximab (non-negotiated), and Erlotinib versus Crizotinib (Table 1).

We utilized two complementary quasi-experimental designs to evaluate the policy impact. The results derived from the DID model constitute the principal findings. Findings from the ITS analysis were reported specifically in instances where trends or effects diverged.

The DID approach estimated the average treatment effect of the NRDLN policy by comparing the pre-post changes in outcomes between the intervention and comparison groups. The regression specification was as follows:

where Y_it is the outcome of interest, P is a dummy variable for the second time period and T is a dummy variable for the treatment group. The interaction term, P*T, is equivalent to a dummy variable equals 1 for observations in the treatment group, in the second period.

We inspected the parallel trends assumption using pre-policy trajectory plots (monthly averages by group) and a diagnostic pre-trend regression. Consistent with recent guidance that pre-trend tests are low-power and diagnostic rather than definitive, we treat visualization as the primary check of magnitude and practical relevance of trend differences (30, 31). To triangulate identification when group comparability is imperfect, we additionally use ITS analyses, which do not require cross-group parallelism (32, 33).

The DID approach estimated the average treatment effect of the NRDLN policy by comparing pre-post outcome changes between the intervention and comparison groups. Our specification excluded hospital-level covariates as the CMEI procurement data are aggregated and lack detailed hospital characteristics (e.g., size or case mix). To validate identification, we employed pre-policy trajectory plots and a diagnostic regression test to assess the parallel trends assumption. Additionally, we conducted ITS analyses, which offer robustness under imperfect cross-group comparability.

For each drug, we applied segmented regression models to estimate immediate level changes and changes in slope after the policy. The ITS model specification was:

Where Time_t is a continuous variable indicating the number of months since the beginning of the study period (August 2016), capturing the baseline trend. Second, Intervention is a binary variable equal to 0 before the implementation of the NRDLN policy and 1 afterward; it measures the immediate level change associated with the policy intervention, reflected by the coefficient β 2 . Third, PostTime_t indicates the number of months since the policy was implemented and captures any change in the trend of the outcome, with the effect measured by β 3 Together, β 2 and β 3 allow us to evaluate both the short-term impact and the sustained effects of the NRDLN policy on drug utilization in public hospitals.

All statistical analyses were conducted using R version 4.1.2. Statistical significance was set at a two-tailed p-value of < 0.05.

This study utilized secondary anonymized data devoid of individual-level patient identifiers. Consequently, ethical approval was not required, and no patients or members of the public participated in the research design or execution.

This study provides compelling quasi-experimental evidence that the NRDLN policy significantly improved access to negotiated anticancer drugs in Chinese public hospitals. Both drug expenditure and DDDs increased substantially after policy implementation, with DDDs rising more rapidly than expenditures. This divergence signals a reduction in per-unit cost and indicates expanded treatment access within a cost-constrained institutional environment.

The increase in DDDs far outpaced the growth of new cancer cases during the same period, suggesting that the policy not only responded to existing demand but also unlocked previously unmet needs. This reinforces the value of price negotiation as a structural lever to mitigate financial barriers and improve equitable access—particularly in upper-middle-income countries where out-of-pocket drug costs remain high and centralized procurement systems play a dominant role in treatment delivery (43, 44).

Despite its effectiveness, the NRDLN policy’s impact was gradual rather than immediate. The absence of significant level changes in ITS analysis, combined with the progressive trend increases, reflects the decentralized and staggered rollout across provinces. In practice, the policy announcement in July 2017 was not uniformly translated into operational changes until several months later in most regions. This underscores a well-recognized gap between policy formulation and administrative execution in large, heterogeneous health systems.

To accelerate translation from policy intent to patient-level benefit, future reforms should prioritize enhanced alignment of national and provincial purchasing schedules, synchronized drug formulary updates, and targeted provider-level guidance. These institutional bottlenecks—rather than policy content itself—likely shaped the observed implementation lag.

The policy’s effect varied markedly by drug type and administration route. Utilization growth was more prominent among Western medicines than TCMs, and among injectable drugs compared to oral formulations. This divergence can be attributed to differences in evidence strength, prescriber confidence, and clinical pathway integration. Western medicines, particularly monoclonal antibodies and kinase inhibitors, are supported by standardized indications and quantifiable clinical endpoints, making them more compatible with formulary-based procurement and insurance reimbursement systems. For instance, trastuzumab has been established as a guideline-recommended therapy for HER2-positive breast and gastric cancers, where it substantially improves response rates and survival outcomes when combined with chemotherapy (45–47). In contrast, TCMs—often adjunctive in nature—lack consistent clinical metrics and thus are more likely to be deprioritized under performance-based or quota-sensitive purchasing regimes (48). A representative example is Shenyi Capsule, which is mainly used as an adjunct to chemotherapy rather than as a stand-alone anticancer treatment (49); when more effective curative drugs become affordable, hospitals and patients may naturally shift away from such adjuvant options.‌‌

Injectable drugs demonstrated sharper increases in DDDs than oral drugs. This may reflect both clinical appropriateness and systemic incentives: injectable agents are typically administered in hospital-based settings, which facilitates institutional monitoring, quality assurance, and integration into bundled payment schemes. Oral therapies, though potentially more convenient, may face adoption barriers due to lower insurance reimbursement, limited physician control, or weaker procurement incentives.

The case of Bevacizumab illustrates the policy’s dual effect of increasing access while containing costs. Despite a substantial rise in DDDs, total expenditures decreased, reflecting the magnitude of negotiated price reductions. Bevacizumab’s expanded indications—covering multiple high-incidence cancers in China—and strong clinical benefit profile (50–52) likely contributed to its rapid uptake. The volume–price decoupling observed in this case is a hallmark of efficient strategic purchasing and suggests that volume-based reimbursement can be achieved without undermining fiscal sustainability.

Such outcomes also highlight how centralized negotiation mechanisms, when well-targeted, can serve as both financial tools and equity-enhancing instruments. The expanded institutional use of previously unaffordable medicines may also reflect broader shifts in care-seeking behavior toward hospitals offering formulary-based coverage, integrated clinical pathways, and reduced out-of-pocket burden (53).

Although the NRDL negotiation has substantially reduced drug prices and expanded access, several unintended consequences warrant attention. Previous studies have highlighted concerns regarding potential inequities and budgetary pressures (48). Consistent with these observations, recent empirical evidence shows that disparities persist across provinces and between urban and rural populations, with patients in less developed regions often facing delayed formulary inclusion and higher out-of-pocket costs (12, 28, 54), while price reductions lowered unit costs, hospital expenditures on oncology drugs sometimes increased due to higher utilization, creating new challenges for institutional budgets and long-term financial sustainability (55–57). Prescribing patterns have also shifted toward innovative targeted therapies, consistent with clinical guideline updates, but this trend raises concerns about rational allocation of limited resources (58, 59). Finally, although direct evidence of “crowding out” of older or non-negotiated drugs is limited, hospital budget constraints may indirectly restrict access to other high-cost therapies, underscoring the need for careful monitoring (59, 60).

While the NRDLN policy has delivered measurable improvements, its long-term success depends on institutionalizing supporting mechanisms. A stable, transparent HTA system is critical to inform future drug inclusion, price reassessment, and prioritization (61, 62). Real-world evidence should be systematically integrated into post-listing evaluations, enabling dynamic adjustments to reimbursement terms and fostering adaptive benefit design (63).

In parallel, payment systems must evolve to reward value over volume. Metrics such as the utilization–expenditure ratio and population-level treatment coverage should be embedded in hospital and insurer performance assessments. Importantly, such measures must be supported by data infrastructure capable of capturing longitudinal treatment outcomes and utilization patterns.

Finally, while this study demonstrates strong population-level effects, further research is warranted to explore patient-level impacts—including treatment adherence, clinical outcomes, and financial protection. Linking procurement data with clinical registries could support more granular assessments of the policy’s effectiveness, equity, and efficiency.

This study has several limitations. First, the analysis relied on aggregated hospital procurement data rather than individual-level clinical or prescription records. As such, it was not possible to assess the appropriateness of prescribing or patient-level outcomes. For interpretation of affordability and equity, this means that increases in hospital-level DDDs and expenditures primarily reflect supply-side availability rather than confirmed patient-level access. We could not observe out-of-pocket spending, copayment levels, reimbursement rates, or financial toxicity for different patient groups, nor could we determine whether utilization gains accrued disproportionately to patients with better insurance, higher income, or those living in urban/tertiary settings. Similarly, we were unable to assess treatment adherence, discontinuation, dose intensity, or time-to-treatment—all of which shape realized affordability and distributional equity. Nevertheless, procurement volumes remain a valid proxy for institutional drug utilization and reflect real-world supply-side changes.

Second, the DID analysis did not account for hospital-specific characteristics such as size, clinical specialty, or local economic context. Although a matched comparison group was used, unobserved heterogeneity across facilities may have influenced estimates. In addition, although strict parallel trends were not observed for all outcomes, transparent visualization and design triangulation with ITS align with contemporary recommendations for health policy evaluations facing imperfect comparators. The consistency between DID and ITS reduces concern that modest pre-trend differences fully account for the observed policy effects. Future work could extend to formal sensitivity analyses that quantify the degree of parallel-trend violation required to reverse conclusions (31, 64).

Third, due to the limited availability of non-negotiated but clinically identical anticancer drugs, the comparison group was based on therapeutic similarity rather than strict indication-matching. This may introduce clinical context variation, although multiple sensitivity checks across drug classes yielded consistent results.

Fourth, expenditures were not adjusted for inflation. However, the substantial growth in DDDs and the moderate increase in costs suggest genuine gains in affordability rather than price distortion.

Finally, the use of aggregated procurement data limits our ability to assess treatment appropriateness, patient adherence, clinical outcomes, or financial toxicity. Although procurement volumes serve as a valid proxy for institutional utilization, they cannot capture individual-level experiences or health impacts. Future research should link procurement with claims, registries, or discharge abstracts to quantify patient out-of-pocket payments, catastrophic health expenditure, and uptake by insurance type, socioeconomic status, and urban–rural location (e.g., concentration indices or distribution-sensitive DID), and to evaluate adherence and outcomes alongside utilization.

Future research should consider linking procurement datasets with clinical registries, hospital discharge records, or insurance claims to evaluate real-world effectiveness, equity, and patient financial burden. Integrating such data would provide a more comprehensive understanding of the policy’s impact across the full care continuum.