背景：改革开放以来，我国科技事业取得了令人瞩目的成就，但伴随而来的还有频发的科研失信问题。其中，医学领域被认为是学术不端行为发生的重灾区之一。目的：从撤销论文和科研诚信政策视角出发，探究医学领域学术不端的较为深层次特征，总结已有防治策略的演变逻辑和些许不足，讨论不端行为产生的原因并提出潜在的治理建议。方法：使用国际公开的撤稿观察数据库回顾性搜集和分析我国学者2010年1月1日至2020年12月31日发表的医学被撤论文；通过北大法宝、科研诚信网系统性检索国内已出台的科研诚信政策文件，描述政策的发布时序、发文主体、效力位阶，并作出政策工具和发展阶段的划分；在此基础上选择2015年12月中国科学技术协会等七部门联合颁布的《发表学术论文“五不准”》为代表性政策举措，评估“五不准”应对学术不端的效果——以季度为时间单位，应用中断时间序列模型展现医学学术不端论文发表数量在“五不准”政策发生后短期和长期内的变动趋势。结果：（1）1534篇中国学者为第一作者的医学被撤论文中，有899篇（58.6%）涉嫌一种或多种形式的学术不端，288篇（18.8%）归因于诚实性错误，余下的347篇（22.6%）难以明确。进一步以899篇学术不端论文为分析对象，得出医学领域发生频次由高到低的具体不端行为是：重复发表、剽窃、虚假同行评议、伪造或篡改、不当署名、其它、违背研究伦理。899篇学术不端论文的原始发表时间范围：2010.01.01–2020.12.12；最终撤销时间范围：2010.07.07–2023.04.04。将诚实性错误论文视为对照，学术不端论文普遍需经历更长的时间才被撤销：撤销时滞中位数1.8年，Kaplan–Meier生存曲线下降地也更为缓慢。仅不当署名类论文的撤销时滞中位数和生存曲线与对照组无显著差异。若按数据图像学术不端与非数据图像学术不端划分，前者无论是在撤销时滞还是生存曲线上均高于后者。此外，剽窃类、虚假同行评议类论文的中位作者数显著低于对照组，而违背研究伦理类论文的中位作者数显著高于对照组。（2）我国科研诚信政策发布数量自2000年以来稳定增长。80份政策文件中，独立发文形式存在的有63份，联合发文形式存在的有17份。涉及的15个牵头部门以教育部为首要发文主体。按政策文件的效力位阶划分，部门规章最多，为61份。按政策工具类型划分，命令工具使用最为频繁，为63份。80份政策文件共同构成了我国科研诚信建设的起步、发展、完善、深化四个阶段。（3）中断时间序列模型拟合结果显示，“五不准”的针对性效果显著。“五不准”政策发生前，我国医学领域虚假同行评议论文发表数量以平均0.97篇/季度（β₁=0.97，P<0.001）的趋势上升。“五不准”颁布后的第一个季度，虚假同行评议论文发表数量平均瞬时减少了17.30篇（β₂=-17.30，P<0.001），且从长远来看，该类论文的上升趋势下降了0.87篇/季度（β₃=-0.87，P<0.001）。“五不准”政策具备溢出效应。“五不准”政策发生前，我国医学领域总体学术不端论文发表数量以平均1.92篇/季度（β₁=1.92，P<0.001）的趋势上升。“五不准”出台后的第一个季度，总体学术不端论文发表数量平均瞬时减少了28.24篇（β₂=-28.24，P<0.001），但其长远上升趋势并未发生显著改变（β₃=-0.57，P=0.193）。替换样本对象为诚实性错误论文以及美国学术不端论文的分析均未观察到结局变量在“五不准”发生时点出现跳跃的断点。以上结果的可靠性在延长时间单位至半年的稳健性检验中继续得以支持。结论：学术不端是我国医学论文被撤的主要原因，其表现形式多样，数据图像学术不端的隐匿性更强。我国颁布的科研诚信政策日益完善，呈现出了“由点及面”的阶段变化特征。“五不准”作为事后治理手段，除了有效抑制虚假同行评议，短期内还取得了抑制总体不端行为的溢出效应。不过，其长期效果难以明确的事实也提示我们学术不端是由个体内部因素和外部环境因素共同导致的。政府部门和科研监管机构在采取灵活、快速政策的基础上，还需针对一些潜在的实质因素完善学术不端的事前预防机制，并积极追踪以往政策的执行情况。同时，加强科研诚信教育，进一步推进科学合理的科技评价工作，建立分级分类的评价体系，逐渐从根本上减少和避免学术不端。

Background: Since reforming and opening up, China has made significant achievements in science and technology. However, alongside these are ongoing issues of dishonesty in scientific research. The field of medicine is considered to be the hardest-hit area of academic misconduct. Objectives: From the perspective of retracted papers and research integrity policies, this paper aims to investigate the relative in-depth characteristics of academic misconduct in medicine, summarize the evolution and shortcomings of existing prevention and control strategies, discuss the causes of academic misconduct, and propose potential governance recommendations. Methods: Using the internationally available Retraction Watch Database to retrospectively collect retracted articles published by Chinese scholars from January 1, 2010 to December 31, 2020. Searching PKULAW.COM and ORICHINA.CN to systematically collect documents of scientific research integrity. Describing the temporal trend, main issuing body, and level of legal efficacy of these policies, as well as dividing policy tools and development stages. Based on this, we identify the “Five No’s for Publication” issued in December 2015 by the China Association for Science and Technology and the other six departments as a representative measure. To evaluate its effectiveness in response to academic misconduct, the short- and long-term trends of the number of medical misconduct publications were presented by conducting an interrupted time series analysis in quarterly time units. Results: (1) Among 1,534 retracted medical papers with Chinese scholars as the first authors, 899 (58.6%) involved at least one instance of academic misconduct, while 288 (18.8%) could be attributed to honest error. The remaining 347 (22.6%) could not be distinguished. Further analysis of the 899 papers retracted for academic misconduct revealed that the specific misbehaviors ranked from the highest to lowest frequency in the field of medicine were: duplication, plagiarism, faked peer review, falsification or fabrication, inappropriate authorship, others, and ethical violations. These papers were first published between January 1, 2010 and December 12, 2020, and subsequently retracted between July 7, 2010 and April 4, 2023. Using papers involving honest errors as the control group, it was found that papers involving academic misconduct generally took longer to be retracted, with a median retraction lag of 1.8 years. The Kaplan-Meier survival curve for the treated group also declined more slowly. Only papers involving inappropriate authorship showed no significant difference in median retraction lag or survival curve when compared to the control group. If the 899 papers were divided into data image-related academic misconduct and non-data image-related academic misconduct, the former is higher than the latter, both in retraction lag and survival curve. In addition, the median number of authors for papers involving plagiarism and faked peer review is significantly lower than that of the control group, while the median number of authors for papers involving ethical violations is significantly higher than that of the control group. (2) The quantity of scientific integrity policy documents released in China has steadily increased since 2000. Of the 80 policy documents, 63 were issued independently, while 17 were issued jointly. The Ministry of Education is the main issuing body among the 15 leading departments concerned. If we classified these policies according to their level of legal efficacy, departmental regulations are the most numerous (61). If we classified these policies from the perspective of policy tools, mandate tools are most frequently used (63). 80 policy documents collectively constitute the initial, developmental, improving, and deepening stages of China’s research integrity construction. (3) The fitted results of regression models show that the targeted effect of the Five No’s policy is significant. Before the implementation of the Five No’s policy, there was an increasing trend (β₁=0.97, P<0.001) in the number of published papers that were retracted for fake peer review. In the first quarter after the introduction of the Five No’s, an average reduction of 17.30 (β₂=-17.30, P<0.001) publications that involve faked peer review was observed, and the long-term trend decreased significantly relative to the pre-intervention period (β₃=-0.87, P<0.001). The Five No’s policy has spillover effects. Before the implementation of the Five No’s policy, there was an increasing trend (β₁=1.92, P<0.001) in the number of published papers that were retracted for all kinds of misconduct. In the first quarter after the introduction of the Five No’s, an average reduction of 28.24 (β₂=-28.24, P<0.001) publications that involve academic misconduct was observed, although the long-term upward trend did not change significantly relative to the pre-intervention period (β₃=-0.57, P=0.193). Replacing the research sample with erroneous papers written by Chinese scholars and misbehavior papers written by American scholars demonstrated no meaningful level of change in the outcome variable at the time when the Five No’s policy was implemented. The reliability of these results was further supported by robustness checks of changing the time unit from quarter to half-year. Conclusion: Academic misconduct is the primary reason for the retraction of medical papers in China. Its manifestations are various, with data image-related misconduct being particularly covert. The scientific research integrity policies promulgated by China are constantly improved, with distinct characteristics at different stages. As a post-hoc policy, the Five No’s has not only successfully curbed faked peer reviews but has also achieved a spillover effect in combating overall misconduct in the short term. However, the uncertainty about its long-term effectiveness hints that academic misconduct is caused by both internal factors and external environmental factors. On the basis of adopting flexible and rapid policies, government agencies and research institutions should improve the mechanism of preventing academic misconduct in advance in view of some substantive factors, and track enforcement of existing policies. By strengthening education, further promoting rational scientific and technological evaluation work, and establishing a graded and classified evaluation system, we can gradually reduce and avoid academic misconduct in the field of medicine.