Air-Moving Device

一致性评价是确保仿制药有效、安全的关键监管措施。然而，若仿制药在通过一致性评价后，生产环节发生变化，是否依旧符合与参比试剂的一致性？实际上，一款仿制药在通过一致性评价后，可对原材料供应商、生产工艺、生产厂址等多项生产环节进行变更，而无需重新进行一致性评价，多数情况只需在省级药监部门进行备案。我分析了国家药监局公布的2019年至今的16万余条药物补充备案，发现通过一致性评价的仿制药、进入集采的药品中，广泛存在过评后生产环节的变更。并且，进入集采的药品，相对于同成分但未进入集采的药品，进行了更多此类变更。这些变更并非一定会影响药效、安全性，但仍需解决如何对此进行有效监管的问题。

定于一尊, one supreme authority, is finally quantified. I analyzed all People's Daily articles mentioning 讲话 (speech) since 1989, and found when Xi speaks, it's usually important (重要讲话). And he is increasingly the only important voice on the Politburo Standing Committee.

A note on the colors: each leader’s line is colored only while they hold the top post (General Secretary or Premier). When they’re just another PSC member, the line is gray.

Almost a decade ago (2017), I started my Twitter account analyzing mentions of Xi in the People's Daily. I looked into 讲话/重要讲话 back then but the data was pretty noisy early in Xi's first term. I was inspired by a tweet (h/t @jingyn1891) to revisit this, with interesting results.

Method: I pulled every People’s Daily article mentioning 讲话 from 1989/1/1 to 2026/10/23 (~70,000 mentions), worked out which PSC member each one refers to, and counted how often the phrase is 重要讲话.

I am only counting instances where PSC members made a speech, not when others refer to the speech. For example, there are many instances where others studied Xi's important speeches, and those were not counted. When I do count them, I find essential the same result.

Before Xi, premiers and other PSC members also gave important speeches. Under Xi, that’s rare. Li Keqiang gave far fewer speeches than Xi (bottom panel), and a lower share were important. His second-term spike is small-sample noise. Li Qiang: even fewer, almost none important.

20 Feb

Younger Chinese are marrying later than previous generations. Here I plotted when different aged cohorts of Chinese first married and saw clear trend of younger generations delaying marriage. Source: Seventh National Population Census, National Bureau of Statistics (2020)

20 Feb

The same trend holds for both urban (城市) and rural (乡村) areas.

22 Sept 2025

Update: 在最近公布的《全国药品集中采购文件》中，增加了两项关于集采药品生产环节变更的规定：变更时需公开变更内容、未发生变更的企业在同等价位时优先入选。这两项规定明显弱于之前征求意见稿中的硬性规定：首个中选周期内不得进行重要生产环节的变更，否则取消中选资格。大概是多方博弈的结果，baby steps也好吧。

Air-Moving Device@airmovingdevice.bsky.social

DOGE/Musk preferentially cancelled grants and contracts to recipients in counties that voted for Harris in 2024. Among cancellations with election data available, 92.9% and 86.1% cancelled grants and contracts went to Harris counties, representing 96.6% and 92.4% of total dollar amounts.

Potential caveat: DOGE doesn't specify how it chose certain contract/grant cancellations to disclose. They claim the ones disclosed represent "~30% of total savings".

It is therefore possible that they made cancellations unbiasedly across the Trump-Harris political spectrum but preferentially disclosed ones to Harris counties for publicity purposes.

To answer this, I need a good background/control set. I compiled all contracts/grants from FY2021-2025 on USAspending, totaling ~19M/24M. ~99% of all cancelled contracts/grants were from this period. Similar results were seen with more restricted time periods, e.g. only FY2024.

Clearly, the background/control sets are distributed across the Trump-Harris spectrum, but the cancellations are biased towards Harris counties. Statistically significant differences shown with Mann-Whitney and Kolmogorov-Smirnov tests (p < 1e-100). Large cluster on very left is DC.

I used election data scraped from Fox News (www.foxnews.com/elections/20...) by github.com/tonmcg/US_Co... For each contract/grant, I found Trump's popular vote margin over Harris in the recipient county. Similar results were seen with NYT's election data (github.com/nytimes/pres...).

I plotted every cancellation, with total dollar amount obligated on the y axis against Trump-over-Harris margin on x. Clearly, there's a bias for more cancellations in Harris counties. But does this reflect true bias or simply more contracts/grants awarded to Harris counties?

I retrieved all publicly available cancellations from DOGE on 3/22, which according to DOGE is a subset of all cancellations. I then cross-referenced them to official spending data on USAspending using links provided by DOGE and ended up with 5,137 and 4,679 contracts and grants with rich metadata.

These metadata include total dollar amounts obligated, dates, and information on contract/grant recipients (address, county, congressional district, etc). I extracted county info (FIPS code) and cross-referenced them to county-level presidential election data from 2024.

Data source: doge.gov/savings — cancelled federal grants and contracts USAspending.gov — contract/grant recipient info github.com/tonmcg/US_Co... & github.com/nytimes/pres... — county-level election data

Thread: prevalence of post-approval changes in generic drugs and jicai (集采) drugs. I analyzed >160k supplemental filings and found widespread post-approval changes in generics and jicai drugs. Importantly, jicai drugs underwent more changes than non-jicai counterparts. bsky.app/profile/airm...

Air-Moving Device@airmovingdevice.bsky.social

While these production-related changes do not necessarily affect drug efficacy or safety, NMPA does not disclose the regulatory tests and inspections done (if any) that addresses whether new suppliers, processes, or sites materially impact drug composition or performance.

Again, I’m in no way against centralized volume-based procurement and generics. In fact, I think they’re fundamentally a great idea for patients, given that their safety and effectiveness are demonstrated with rigorous testing and proper regulatory oversight.

Air-Moving Device@airmovingdevice.bsky.social

Here I am plotting the cumulative number of changes, averaged by the number of drugs. Clearly, Telmisartan generics that entered jicai underwent more production-related changes than non-jicai generics. Same trends were also seen for metformin hydrochloride generics.

Expanding this to the entire dataset, I found 121 drugs that had both jicai and non-jicai generics, n = 352 and 768 respectively. While the number of total filings were similar between the two groups, production-related changes in jicai drugs were ~2-fold that of non-jicai.

Air-Moving Device@airmovingdevice.bsky.social

To test this more rigorously, I compared jicai drugs with drugs that (1) share the same active ingredient and (2) passed BE tests, but (3) did not enter jicai.

For example, there are a total of 28 Telmisartan generics that passed BE, and 7 of these entered jicai on 2021/2/8. I tabulated the number of production-related changes (supplier, process, or site) that happened for jicai vs non-jicai drugs starting on 2021/2/8.

Air-Moving Device@airmovingdevice.bsky.social

Drugs that entered jicai: Here I plotted for each drug the date it entered jicai and dates of all supplemental filings. * 45.7% of jicai drugs changed suppliers post-approval * 16.4% changed production processes * 15.3% changed manufacturing sites

An interesting finding here is that percentages of drugs that underwent postapproval changes are higher for jicai drugs than generics. A hypothesis is that jicai drugs undergo more postapproval changes due to cost pressures associated with low bids.

Air-Moving Device@airmovingdevice.bsky.social

I focused on 3 types of changes: supplier, production process, and manufacturing site. These are more likely to impact drug efficacy than other changes. Using permit no (国药准字), drug name and manufacturer name, I matched filings to approved generics and jicai drugs.

Generics that passed BE: Here I plotted for each drug the date it passed BE and dates of all supplemental filings. * 28.2% of generics changed suppliers post-approval * 9.6% changed production processes * 14.1% changed manufacturing sites

Air-Moving Device@airmovingdevice.bsky.social

I analyzed supplemental filings disclosed by NMPA during 2019/1/1-2025/2/5 and found a total of >160k filings. I cross referenced these with generics that passed BE disclosed by NMPA (n = 1,988) and drugs that entered centralized procurement jicai disclosed by NHSA (n = 1,933).

Data source-- Supplemental filings: www.nmpa.gov.cn/datasearch/s... Generics that passed BE: www.cde.org.cn Jicai drugs: Shanghai division of NHSA www.smpaa.cn

Air-Moving Device@airmovingdevice.bsky.social

After passing bioequivalence tests, generics can undergo changes in supplier, manufacturing process or site. These are often submitted as supplemental filings to province-level drug admins, and no further BE tests are required. Here I quantified how prevalent these changes are.

需要指出的是，进行了供应商、生产工艺、生产厂址变更，并非一定会对药物成分、药效造成影响。但绝大多数非业内人士对于过评药品、集采药品可能会在过评后进行生产变更并不知情。药监部门也并未披露关于这些变更的详细信息——如厂家对新供应商的原料进行杂质分析的情况、药监部门是否对新增的生产厂址进行检查、采用新工艺生产的药品是否进行了体外生物等效性分析等等。

推广仿制药、集采制度是解决医保支付问题、为广大患者提供低价药品的必要的、有效的方法。为应对、缓解患者和广大民众对仿制药和集采制度的质疑，药厂、药监部门、医保部门应扩大信息公开的力度并确保公开的信息及时准确。药监部门也应对于可能影响药效、安全性的过评后变更进行更加有效的审核与监管。

这里引出了一个重要的问题：对于进入集采的药品，厂商是否会因为降低成本的压力，而进行更多的生产环节变更？为分析这一问题，我对比了进入集采的仿制药和未进入集采（但也通过了一致性测评）的同种药品。比如，CDE数据库中的仿制替米沙坦片共有28种，其中7种在2021-1批进入集采（2/8日公布）。我统计了自当日起，进入集采和未进入集采的替米沙坦片进行生产环节变更（供应商、工艺、厂址）的次数，以及所有备案的次数。我将这些次数平均到了每个药品上（集采7种，非集采21种），以便比较。进入集采的替米沙坦片相比未进入集采的，进行了更多的生产环节变更，整体备案数也更多。盐酸二甲双胍片也存在类似的现象。

为确保这两个例子并非孤例，我将此分析推广到了CDE数据库、集采数据库中的所有药品，发现共有121类药品可供对比（同一个有效成分下，既有进入集采的，也有未进入集采的药品）。合并后共有352种进入集采的药品，768种未进入集采的药品。统计生产环节变更和所有备案次数后发现，进入集采的药品和未进入集采的药品，总备案数类似。然而，在关键的、可能会影响药物成分、药效的变更上，进入集采的药品发生了更多的生产环节的变更，约2倍于未进入集采的药品。再次强调，这里的分析并非简单地对比所有进入集采的药品和未进入集采的药品。而是找到了对应的同成分的集采、非集采仿制药，进行对比。

我首先分析了通过一致性评价的仿制药，是否会在过评后进行生产变更。这里，每一行代表了一种过评仿制药（共1988种），横轴代表事件发生的时间。我标注了几种事件：过评的时间、发生各类变更的时间、进行其它备案的时间。可见，在过评后（即斜黑线右侧），出现许多例生产环节上的变更。我对此进行了分析，只统计了过评之后的生产变更，发现28.2%、9.6%、14.1%的药品在过评后分别进行了供应商变更、生产工艺变更、生产厂址变更。

进入集采的药品出现了类似的情况，甚至变更的比例更高。这里我标注了每个药品进入国家集采的时间（共8批）和进行变更备案的时间。 45.7%、16.4%、15.3%的集采药品在进入集采之后，分别进行了供应商变更、生产工艺变更、生产厂址变更。三个比例都比通过一致性评价的药品更高（进入集采的仿制药需经过一致性评价，但不是所有过评的仿制药都能够进入集采）。

为了分析通过一致性评价的仿制药和进入集采的药品，是否存在过评后、进入集采后进行变更的情况，我抓取了从2019-1-1到2025-2-5的16万余条药物补充备案信息。我将此信息与CDE公开的1988条过评药品信息和医保局公布的1933条集采药品信息进行了对比。备案信息来源： www.nmpa.gov.cn/datasearch/s... （经业内权威数据库整理）一致性评价药品信息： www.cde.org.cn 集采药品信息：上海医保局阳光采购网 www.smpaa.cn

我分析了三类*可能*会影响药效、安全的生产变更：供应商变更（绝大多数为原料供应商）、生产工艺变更、生产厂址变更。具体做法是在备案信息中寻找相应字段：供应商，生产工艺/工艺参数/工艺变更，生产场地/生产地址/生产企业地址。为了将备案信息与通过一致性评价药物、集采药物对应，我使用了国药准字或者药品名称+生产厂商全名进行匹配。

国家药监局网站定期公开“境内生产药品备案信息”（https://www.nmpa.gov.cn/datasearch/search-result.html）。经查询可见，许多药品会进行原料供应商、生产工艺、生产厂址的多项变更，并只需在省级药监部分进行备案。即使是进行了一致性评价的仿制药，进行此类变更后，也无需重新进行一致性评价。

13 Feb 2025

It’s weird seeing my data contributing to public discussions in China — something I’ve wanted to do from the start, but for obvious reasons I cannot post inside the GFW. A small part of me wants some credit, but it’s more the powerlessness that I can’t explain and defend my data on my own terms.