China’s Drug Price Cuts Under Scrutiny: Massive Review Finds No Clear Evidence Procured Generics Perform Worse

When China launched its National Volume-Based Procurement (NVBP) policy in late 2018, the promise was simple: guaranteed hospital purchase volumes in exchange for dramatically lower drug prices. By the end of 2025, eleven rounds of competitive bidding had covered 491 products, with average price reductions exceeding 50 percent and, in some cases, cuts surpassing 90 percent. But the policy’s aggressive cost savings have long been shadowed by a persistent clinical question: do generics procured at such steep discounts actually work as well as the originator drugs they replaced? A new systematic review published in Frontiers in Public Health offers the most comprehensive answer yet—and the answer is more nuanced than either critics or champions of the policy might expect.

Researchers Lele Zhang and Youjie Zhang from China Pharmaceutical University in Nanjing systematically searched six databases—PubMed, Web of Science, Cochrane CENTRAL, CNKI, Wanfang Data, and VIP—for studies published between January 2018 and January 2026 that compared NVBP-selected medicines against their comparators. After screening 2,903 bibliographic records and 862 full-text reports, they identified 81 eligible studies covering more than 40 active ingredients across cardiovascular, anti-infective, psychiatric, endocrine, oncology, gastrointestinal, respiratory, and antiviral therapeutic areas. The review, registered with PROSPERO (CRD420261303351), was conducted according to PRISMA 2020 guidelines and represents the first systematic synthesis of real-world clinical evidence on China’s flagship drug procurement policy.

The central finding is striking in its consistency: across the 69 comparisons in the main analysis—where a procurement-selected generic was compared directly against the originator product—60 of 62 counted efficacy comparisons found no statistically significant difference, and 55 of 65 safety comparisons likewise showed no difference. No efficacy comparison carried an assigned direction favoring either product. This pattern held across drug classes, from atorvastatin and clopidogrel to ceftriaxone and meropenem, from metformin and empagliflozin to bortezomib and pemetrexed. In the largest single study, a 16-hospital matched cohort of over 39,000 atorvastatin users, lipid outcomes were comparable between generic and originator users.

Yet the picture is not uniformly reassuring. Seven safety comparisons did show statistically significant differences: four favored the originator (moxifloxacin tablets, olanzapine, pediatric ceftriaxone, and atorvastatin), while three favored the selected product (linezolid tablets, ceftriaxone, and bortezomib). In three further comparisons, no direction could be assigned because significant findings within a single cohort pointed in opposite directions. Notably, when the analysis was restricted to the 19 multicenter studies, the balance of safety signals reversed—three of the four comparisons retaining a direction favored the selected product rather than the originator, suggesting that the unfavorable signals arose disproportionately from smaller single-center studies.

The review’s methodological rigor is itself a finding. Risk of bias was assessed across eight domains, three of which were developed specifically for this policy setting because no existing instrument captured them. Exposure misclassification emerged as the dominant threat: 21 of the 81 studies began recruitment more than 12 months before the relevant procurement round was announced, meaning patients recorded in the selected-product arm during those periods could not have received a product holding selected status. In one extreme case, recruitment opened nearly nine years before the medicine entered procurement. This dilution of the exposure contrast means the observed concordance of null findings is, to an unquantifiable degree, a comparison of like with like.

Perhaps most sobering is the review’s assessment of what the existing evidence can actually detect. Using the event counts and group sizes reported in the included studies, the authors calculated the smallest difference each comparison could have detected with 80 percent statistical power. The median smallest detectable fall in response rate was 16.5 percentage points, and the median smallest detectable relative risk for harm was 2.67. In practical terms, 17 of 22 efficacy comparisons could not have detected a drop in response rate of less than 10 percentage points, and 19 of 27 safety comparisons could not have detected even a doubling of risk. The concordance of null findings is therefore largely a property of sample sizes rather than a demonstration of equivalence.

The certainty of the evidence, rated using the GRADE framework, was very low for efficacy, safety, and pharmacokinetics, and low for economic outcomes. This is not a bureaucratic formality. An effect estimate was unavailable for 42 of the 81 studies, which reported only a statement of significance or no difference without the underlying data. Sixty-two of the 81 studies were conducted at a single center, and 23 originated from just five research groups or multicenter platforms, meaning the evidence base is less independent than the raw count suggests. Most strikingly, when the authors re-examined 24 included studies against their source articles, six reported test statistics that could not be reproduced from their own summary statistics or counts. In one case, every reported pairing of t and P was arithmetically impossible, and the abstract contradicted the results section.

The review also highlights a distinctive feature of China’s procurement model that sets it apart from generic substitution in the United States, Europe, or Australia. In those settings, substitution typically occurs at the pharmacy level, and patients generally retain the right to request the originator brand. Under NVBP, winning manufacturers receive guaranteed procurement volumes, and non-selected products—including originator brands—may be effectively unavailable within the hospital formulary. Patients and prescribers face a near-complete loss of choice. This mandatory conversion means that any quality or safety signal, even a rare one, reaches a much larger share of the patient population than in voluntary substitution systems, raising the stakes for post-market surveillance.

One pharmacokinetic observation deserves particular attention. In a within-patient study, one procured olanzapine formulation produced a significantly lower dose-corrected steady-state concentration than another procured formulation, even though each had demonstrated bioequivalence to the reference product. Both products were winning suppliers—one from the pilot stage and one from the expansion stage. This finding underscores a fundamental limitation of bioequivalence testing: two products each bioequivalent to a common reference are not thereby bioequivalent to one another, and the intervals permitted by bioequivalence criteria allow a wider separation between two generics than between either and the originator. The authors note that China currently lacks a unified, dynamically updated national narrow therapeutic index catalogue, and they argue that differentiated procurement and monitoring protocols for narrow therapeutic index medicines should be considered a policy priority.

The review’s conclusions are measured. The available evidence does not show that NVBP-selected generics perform worse than their comparators, and this provides cautious reassurance that the policy’s dramatic price reductions have not come at an evident cost to clinical quality. But the authors are equally clear that the concordance of null findings cannot be read as demonstrated equivalence. They call for prospective, adequately powered studies reporting effect estimates with confidence intervals rather than significance tests alone, targeted real-world safety studies for the specific medicines where signals were detected, and product-specific rather than class-wide post-procurement monitoring. They also recommend that future evaluations record the manufacturer and, where feasible, the batch of the dispensed product, since the evidence characterizes individual products rather than classes of procured generics. As the policy enters its renewal phase—with contracts for the first eight rounds renewed nationwide in February 2026—these findings establish both a baseline and a roadmap for the evidence that will ultimately determine whether the world’s largest experiment in centralized drug procurement delivers on its clinical as well as its economic promises.

Subject of Research: Clinical efficacy and safety of generics procured through China's National Volume-Based Procurement policy

Article Title: Clinical efficacy and safety of medicines procured through China’s national volume-based procurement policy: a systematic review of 81 studies

Article References: Clinical efficacy and safety of medicines procured through China’s national volume-based procurement policy: a systematic review of 81 studies. (n.d.). https://doi.org/10.3389/fpubh.2026.1820505

Image Credits: AI Generated

DOI: 10.3389/fpubh.2026.1820505

Keywords: China, National Volume-Based Procurement, generic drugs, drug pricing, systematic review, bioequivalence, drug safety, health policy, pharmacovigilance, therapeutic equivalence, narrow therapeutic index, GRADE certainty