TruthCert > The Warning
==================== 第 1 单元：警告 ====================
幻灯片 1.2：CAST 故事
幻灯片 1.3：死亡人数
幻灯片 1.4：副歌
幻灯片1.5：问题
======================模块2：土地====================
幻灯片 3.5：四个等级
幻灯片 3.7：禁止
==================== 模块 4：方法====================
幻灯片 5.2：危机
幻灯片 5.3：种子
幻灯片5.5：副歌
==================== 模块 6：决策者 ====================
幻灯片 6.2：决策矩阵
幻灯片 6.4：一页摘要
==================== 第 7 单元：实践教程 ====================
幻灯片 7.2：您要做什么需要
幻灯片 7.3：步骤
幻灯片 7.4：数据源
======================模块 8：可持续性与限制 ====================
幻灯片 8.3：诚实的真相
==================== 第 9 单元：参考资料和测验 ====================
==================== 第 10 单元：真实故事 ====================
你没听过医生的故事
who, certain of their cure,
killed more patients than the disease itself?
who, certain of their cure,
killed more patients than the disease itself?
那一年1989
UNITED STATES
美国的每位心脏病专家都知道这个逻辑：心脏病发作后心律不齐是危险的。抑制这些心律失常的药物肯定会挽救生命。
推理是合理的。机制很清楚。这些药物是为 200,000 Americans every year.
开的，没有人要求提供证据。逻辑就足够了。
推理是合理的。机制很清楚。这些药物是为 200,000 Americans every year.
开的，没有人要求提供证据。逻辑就足够了。
Echt DS et al. NEJM 1991; Moore TJ. Deadly Medicine, 1995
然后进行了名为 CAST...
50,000
Americans killed by "the cure" each year
IN SIMPLE WORDS
医生开出的药物 made sense but had never been tested properly。这种药杀死的人比越南战争还要多。每年。
“数字赤裸裸 —
without witness, without proof, without certification.
人们都死了。”
without witness, without proof, without certification.
人们都死了。”
这就是 TruthCert 存在的原因。
Do you think such catastrophes happen only in America?
Consider Africa.
Consider Africa.
借用证据的负担
90%
of clinical trials
outside Africa
outside Africa
13%
of global disease
burden in Africa
burden in Africa
<2%
of global health
research funding
research funding
IN SIMPLE WORDS
Africa has many sick people but 很少有研究 about how to treat them.
Doctors must use treatments tested on people in other countries.
But what works in London may not work in Lagos.
幻灯片2.1：十三个
幻灯片2.2：国家
幻灯片 2.4：七种痛苦
幻灯片 2.5：副歌
====================== 第 3 单元：解决方案====================
你没看到十三个国家
where mothers die giving life,
where children fall to fever,
where 在黑暗中做出的决定?
where mothers die giving life,
where children fall to fever,
where 在黑暗中做出的决定?
Thirteen Nations
Maternal deaths per 100,000 live births (WHO 2020)
NGA
Nigeria
917
GHA
Ghana
308
KEN
Kenya
342
TZA
Tanzania
524
UGA
Uganda
284
ZAF
South Africa
127
SLE
Sierra Leone
1,120
LBR
Liberia
652
CIV
Cote d'Ivoire
617
BEN
Benin
523
BFA
Burkina Faso
320
GMB
The Gambia
458
SWZ
Eswatini
240
Compare: UK = 10, USA = 21, Sweden = 4
SIERRA LEONE
In Freetown, a woman giving birth is 112 times more likely to die
than a woman in London.
卫生部长必须决定：哪些干预措施值得有限的预算？哪种治疗方法最能挽救母亲的生命？
她有来自欧洲的研究。她有来自美国的模特。
But does she have proof that applies to her people?
卫生部长必须决定：哪些干预措施值得有限的预算？哪种治疗方法最能挽救母亲的生命？
她有来自欧洲的研究。她有来自美国的模特。
But does she have proof that applies to her people?
WHO Maternal Mortality Report, 2023
Seven Disease Groups
HIV
25.6M in Africa
MAL
95% of deaths
MCH
Mother & Child
NCD
Chronic Disease
CVD
Heart Disease
NTD
Tropical Disease
HSP
Health Systems
13 countries x 7 disease groups = 91 configuration packs
“他们从远处获取了证据，
without testing if it fit their land,
without certifying if it fit their people.
而且数字是赤裸裸的。”
without testing if it fit their land,
without certifying if it fit their people.
而且数字是赤裸裸的。”
But what if every number had to show its face?
What if every statistic had to name its witness?
What if naked numbers were forbidden?
What if every statistic had to name its witness?
What if naked numbers were forbidden?
TruthCert
为裸体数字穿衣
🔒
核心规则
No number shall be shown to decision-makers
unless it carries proof of its origin,
proof of its transformation,
及其在此上下文中的有效性证明。
IN SIMPLE WORDS
每个数字都必须回答： “你从哪里来来自？” and “我可以相信你吗？”
What Is Proof?
1
Evidence Locator
该数字的确切研究、数据库或来源
2
Content Hash
证明数据未被更改的数字指纹
3
Transformation Trail
从原始数据到最终数据的每个计算步骤number
4
Validation Status
自动检查是否通过？警告是什么？
What TruthCert Output Looks Like
TRUTHCERT CERTIFIED CLAIM
claim: "Misoprostol reduces PPH by 24%"
grade: STABLE
context: NGA (Nigeria)
african_studies: k = 6 # meets GO threshold
effect_size: RR 0.76 [95% CI: 0.68-0.84]
evidence_locator: doi:10.1016/S0140-6736(10)60348-7
hash: sha256:a3f2c9...
cost_per_dose: NGN 450 # local currency
icer: NGN 12,400 per DALY averted
validated: PASS # all checks passed
Four Grades of Certainty
STABLE
有力的证据。充满信心地继续。
MODERATE
有差距的良好证据。状态不确定性。
EXPOSED
Structural uncertainty. Scenarios only.
UNCERTAIN
Refuse false precision. Bounds only.
如何分配等级
Grade
African Studies (k)
您可以做什么
STABLE
k >= 4，低异质性 (I² < 50%)
Full HTA, PSA, VOI, point estimates
MODERATE
k >= 4，高异质性，OR k = 2-3
Full HTA with widened CIs, emphasize uncertainty
EXPOSED
k = 1, OR significant transportability concerns
Scenario analysis only, no point estimates
UNCERTAIN
k = 0，或来自记忆/未经验证的证据
Bounds only, worst-case, value-of-research
What Is Forbidden
MEMORY-LEAK = BLOCK
如果有人说“我记得统计数据约为 30%”—
that cannot be certified.
记忆不是证据。回忆不是证据。
that cannot be certified.
记忆不是证据。回忆不是证据。
IN SIMPLE WORDS
您不能说 "I think I heard..." or "Someone told me..."
您必须显示 exactly where 号码来自。
幻灯片 4.1：价值问题
幻灯片 4.3：GDP 陷阱
幻灯片 4.4：更好的方式
幻灯片 4.6：副歌
====================== 第 5 单元：纪律 ====================
When a government has only $50 per person per year for health,
how do you decide which treatments to fund?
how do you decide which treatments to fund?
Health Technology Assessment
核心问题
For every Naira, Cedi, or Shilling spent,
how much health 我们获益了吗？还有什么 could we have done 用这笔钱？
阈值陷阱
历史警告
For years, economists said: "Cost-effective if it costs less than 3x GDP per capita."
通过这种逻辑， 3,000 美元的治疗在美国（GDP 76,000 美元）“值得”，但在塞拉利昂（GDP 500 美元）“不值得”。
同样的治疗。同样的好处。根据您的出生地点，得出不同的结论。
通过这种逻辑， 3,000 美元的治疗在美国（GDP 76,000 美元）“值得”，但在塞拉利昂（GDP 500 美元）“不值得”。
同样的治疗。同样的好处。根据您的出生地点，得出不同的结论。
Woods B et al. Lancet Global Health 2016; Revill P et al. Health Policy Plan 2018
A Better Way
A
Affordability Analysis
Can the health system actually pay? What else gets cut?
N
Net Health Benefit
Health gained minus health lost by diverting resources
V
Value of Information
Is it worth doing more research before deciding?
Local Currency. Always.
NGN
Nigerian Naira
GHS
Ghana Cedi
KES
Kenyan Shilling
XOF
CFA Franc
为什么这很重要
A minister in Accra budgets in Cedis.
She shouldn't have to convert from dollars.
Her budget, her currency, her decision.
“当数字被掩盖时证明，
when every claim showed its source,
当货币与土地匹配时 —
then the decision-makers could see clearly."
when every claim showed its source,
当货币与土地匹配时 —
then the decision-makers could see clearly."
In 2012, scientists at Amgen tried to reproduce
53 项里程碑式的癌症研究.
How many could they replicate?
53 项里程碑式的癌症研究.
How many could they replicate?
6
Only 6 of 53 landmark studies could be reproduced.
IN SIMPLE WORDS
89% 的“突破性”研究是错误的 or couldn't be repeated.
Treatments given to patients. Based on findings that failed.
Begley CG & Ellis LM. Nature 2012;483:531-533
种子的纪律
THE REQUIREMENT
每项 TruthCert 分析都必须 exactly reproducible.
An analyst in Nairobi and an analyst in Geneva must get identical results.
786888
Master Seed
Fixed
All Random Processes
Logged
Every Step
专为现实
THE CONSTRAINT
A health economist in Freetown has an ordinary laptop, maybe unreliable internet.
TruthCert must run on what she actually has.
60s
Core Analysis
45s
Uncertainty (5K draws)
Offline
Works Without Internet
“同样的种子在阿克拉和日内瓦种下，
两个地方都结出了同样的果实。
任何人都可以再次种下种子，
并收获相同的真理。”
两个地方都结出了同样的果实。
任何人都可以再次种下种子，
并收获相同的真理。”
部长和董事
What you need to know in 3 minutes
每个等级对您意味着什么
When TruthCert Says... You Should...
STABLE
充满信心地前进。 Present point estimates. Defend the numbers.
MODERATE
谨慎地前进。 Show ranges. Say "between X and Y."
EXPOSED
Show scenarios. "If A, then X. If B, then Y." No single number.
UNCERTAIN
Say "we don't know." Request more research before committing.
当政治与证据发生冲突时
困难情况
选民要求它做什么？
TruthCert 为您提供盾牌：
“我想帮助我们的人民。但证据表明我们不知道这在这里是否有效。让我资助一张 pilot study first—so we can be sure we're helping, not harming."
You're not saying no. You're saying "let's be sure."
TruthCert 为您提供盾牌：
“我想帮助我们的人民。但证据表明我们不知道这在这里是否有效。让我资助一张 pilot study first—so we can be sure we're helping, not harming."
You're not saying no. You're saying "let's be sure."
您的一页摘要
1
Numbers must have sources
If staff can't show where a number came from, don't use it
2
寻找非洲证据
询问“在非洲进行了多少项研究？” （需要 4+ 才能获得信心）
3
Demand local currency
Costs should be in Naira/Cedi/Shilling, not USD
4
Embrace uncertainty
"We don't know" is more honest than a fake precise number
运行您的第一个分析
A practical walkthrough
您需要什么
Browser
Chrome/Firefox/Edge
Data
您输入的 YAML
~2 hrs
Training time
NO SPECIAL SOFTWARE NEEDED
TruthCert runs in your browser. No installation. Works offline after first load.
Five Steps
1
Select Country + Disease
Choose from 13 countries, 7 disease groups. Example: NGA + MCH
2
输入本地数据
Unit costs, target population, budget. In NGN (local currency).
3
Select Evidence
研究链接。系统自动统计非洲研究 (k)。
4
Run Analysis
Click "Certify." Wait ~60 seconds. System assigns grade.
5
Export Results
下载 PDF 报告。所有声明都有证据定位器 + 哈希值。
从哪里获取数据
1
Unit Costs
WHO-CHOICE database, iDSI costing studies, local MOH price lists
2
Disease Burden
GBD (Global Burden of Disease), DHS surveys, local HMIS
3
Effect Sizes
Cochrane reviews, published meta-analyses, WHO guidelines
4
Budget Data
National health accounts, MOF budget documents, donor reports
当非洲证据稀疏时
IF k < 4（少于 4 个非洲）研究）
TruthCert 不会给您虚假的自信答案。相反：
1
Pivot to broader group
If "CVD in Ghana" has k=1, try "NCD in West Africa"
2
使用情景分析
Show best-case, worst-case, and middle scenarios
3
Calculate Value of Information
Would a new local study be worth funding?
Sustainability
How TruthCert survives long-term
📁
Open Source
Code is public. Anyone can run, modify, improve.
🎓
Train Locals
Each country builds own HTA capacity. Not dependent on outsiders.
💻
Offline-First
Works without internet. No cloud subscription needed.
What TruthCert Cannot Do
⚠
无法创建证据。 If no studies exist, TruthCert can't invent them.
⚠
Cannot guarantee political acceptance. 证据是必要的，但还不足以政策变更。
⚠
Cannot replace judgment. It informs decisions, doesn't make them.
⚠
无法修复错误数据。 如果您的输入成本错误，输出也会错误。
⚠
无法解释实施情况。 A cost-effective intervention badly implemented may fail.
"TruthCert does not promise certainty.
It promises 关于诚实不确定性.
诚实可以拯救生命。”
It promises 关于诚实不确定性.
诚实可以拯救生命。”
References
本课程引用的主要来源
- Echt DS、Liebson PR、Mitchell LB 等。接受恩卡尼、氟卡尼或安慰剂的患者的死亡率和发病率：CAST。 NEJM 1991;324:781-788.
- Moore TJ. Deadly Medicine: Why Tens of Thousands of Heart Patients Died in America's Worst Drug Disaster. Simon & Schuster, 1995.
- Begley CG, Ellis LM. Drug development: Raise standards for preclinical cancer research. Nature 2012;483:531-533.
- Woods B、Reville P、Sculpher M、Claxton K。国家级成本效益阈值：初步估计和进一步研究的必要性。 Lancet Global Health 2016;4:e594-e601.
- WHO. Trends in maternal mortality 2000-2020. Geneva: World Health Organization, 2023.
- WHO. World Malaria Report 2022. Geneva: World Health Organization, 2022.
- UNAIDS. Global HIV Statistics 2023. Geneva: UNAIDS, 2023.
- Revill P, Ochalek J, Lomas J, et al. Cost-effectiveness thresholds: guiding health care spending for population health improvement. Health Policy Plan 2018;33:707-716.
在“知道”其治疗方法有效的医生的故事中，发生了什么？ CAST 试验显示？
药物按预期发挥作用
The drugs killed more people than they saved
药物没有效果
试验没有结论
当数字为“裸”时，这意味着什么？
They are displayed without formatting
它们是近似值
They have no proof of where they came from
它们的货币是错误的
A minister receives a TruthCert report marked "EXPOSED." What should she do?
完全拒绝该计划
就好像证据充足一样继续
Present scenarios ("if A then X, if B then Y") and consider a pilot
等待更多的国际研究
Why does TruthCert insist on local currency?
To make calculations easier
Because ministers budget in local currency and shouldn't have to convert
To avoid exchange rate fluctuations
It's an arbitrary requirement
✔
Course Complete
“当数字是穿上衣服，
when every claim carried its proof,
当证据在使用它的土地上进行测试时—
然后母亲们生活了，
孩子们茁壮成长，
决策者可以睡在”
when every claim carried its proof,
当证据在使用它的土地上进行测试时—
然后母亲们生活了，
孩子们茁壮成长，
决策者可以睡在”
This is TruthCert.
幻灯片 10.1：模块开场
幻灯片 10.3：莱因哈特-罗格夫的故事
幻灯片 10.7：杜克癌症故事
幻灯片 10.11：RIAT 故事
幻灯片 10.15：复制危机故事
幻灯片 10.19：模块摘要
What if history had given us warnings—
用电子表格编写，隐藏在数据中，
buried in files no one thought to check?
用电子表格编写，隐藏在数据中，
buried in files no one thought to check?
这些是确定性、错误和不验证的代价的故事。
What if a spreadsheet error
塑造国家的命运？
塑造国家的命运？
改变历史的电子表格
真实的DATA
In 2010, Harvard economists Reinhart and Rogoff published "Growth in a Time of Debt".
Their claim: when a country's debt exceeds 90% of GDP, economic growth collapses.
Politicians worldwide cited this to justify austerity measures—cutting public spending, reducing services, freezing wages.
Then in 2013, a graduate student named Thomas Herndon asked for their spreadsheet. He found: an Excel error that excluded 5 countries，选择性数据加权，非常规方法。真正的关系是 much weaker.
Politicians worldwide cited this to justify austerity measures—cutting public spending, reducing services, freezing wages.
Then in 2013, a graduate student named Thomas Herndon asked for their spreadsheet. He found: an Excel error that excluded 5 countries，选择性数据加权，非常规方法。真正的关系是 much weaker.
Herndon, Ash & Pollin (2014). Cambridge Journal of Economics.
两条路径
您是一名政策顾问。莱因哈特-罗格夫就在您的办公桌上。你会做什么？
路径 A：相信结果
接受 90% 阈值作为事实
↓
Advocate austerity policies
↓
Millions suffer from unnecessary budget cuts, unemployment rises, recovery stalls
路径 B：请求数据
索取电子表格和方法
↓
发现 Excel错误
↓
Make evidence-based policy 实际为公民服务
THE REVELATION
Excel电子表格错误是我们需要再现性的原因。如果莱因哈特和罗格夫被要求分享他们的电子表格，错误就会在 2010 年被发现，而不是 2013 年。
Three years of policy built on a formula that missed five cells.
Three years of policy built on a formula that missed five cells.
THE LESSON
可重复性不是官僚主义。 这是各国采取行动之前和之后发现错误之间的区别。
What happens when genomic signatures
建立在沙子？
建立在沙子？
建立在欺诈基础上的试验
真实的DATA
Dr. Anil Potti at Duke University published genomic predictors claiming to match
cancer patients to the chemotherapy most likely to help them.
临床试验招募了患者。根据他的算法选择治疗方法。承诺： personalized medicine that could revolutionize cancer care.
Then investigations revealed: 捏造数据，不可能的结果，操纵数字。试验被停止。患者接受了基于欺诈的治疗。
临床试验招募了患者。根据他的算法选择治疗方法。承诺： personalized medicine that could revolutionize cancer care.
Then investigations revealed: 捏造数据，不可能的结果，操纵数字。试验被停止。患者接受了基于欺诈的治疗。
Baggerly & Coombes (2009). Annals of Applied Statistics. IOM Report (2012).
两条路径
您是 2010 年的癌症患者。您接受了 Potti 指导的治疗。你做什么？
路径A：信任出版物
Accept the prestigious journal's findings
↓
参加临床试验
↓
Receive treatment based on fraud—potentially harmful, definitely not personalized
路径B：询问验证
请求独立复制的证据
↓
Discover none exists
↓
Avoid harmful treatment—wait for properly validated approaches
THE REVELATION
Reproducibility is not bureaucracy. It is patient protection.
杜克大学丑闻导致医学研究所推荐 omics-based tests require independent validation before clinical use.
对于这些试验中的患者来说，这项政策来得太迟了。
杜克大学丑闻导致医学研究所推荐 omics-based tests require independent validation before clinical use.
对于这些试验中的患者来说，这项政策来得太迟了。
THE LESSON
Independent validation saves lives. 著名机构、著名研究人员、顶级期刊——这些都不能取代让其他人检查工作的简单行为。
Can dead trials
be brought back to life?
be brought back to life?
复活埋藏的证据
真实的DATA
RIAT 倡议 (Restoring Invisible and Abandoned Trials) 确定了未发表或误报结果的临床试验。
Study 329 （帕罗西汀治疗青少年抑郁症）：2001 年的原始出版物声称该药物 “安全且可靠”有效。” 14年来，医生们都是根据这一说法开药。
RIAT研究人员获得了完整数据。他们 2015 年的重新分析发现：该药物 neither safe nor effective 适合该人群。原作者选择性地报告了结果。
Study 329 （帕罗西汀治疗青少年抑郁症）：2001 年的原始出版物声称该药物 “安全且可靠”有效。” 14年来，医生们都是根据这一说法开药。
RIAT研究人员获得了完整数据。他们 2015 年的重新分析发现：该药物 neither safe nor effective 适合该人群。原作者选择性地报告了结果。
Le Noury et al. (2015). BMJ. RIAT Project: riat-support.org
两条路径
您发现 15 年前的一项试验被误报。伤害可能仍在发生。你会做什么？
Path A: Accept It Is Too Late
Assume the moment has passed
↓
有缺陷的研究仍然记录
↓
Old harms continue—doctors keep prescribing based on false evidence
路径B：寻求原始数据
Obtain raw data through legal/regulatory means
↓
重新分析并发布更正
↓
Prevent future harm—update guidelines, change practice
THE REVELATION
It is never too late to correct the record.
RIAT证明，通过决心和数据访问， even decades-old deceptions can be overturned.
事实并非如此过期。寻求它的义务也没有。
RIAT证明，通过决心和数据访问， even decades-old deceptions can be overturned.
事实并非如此过期。寻求它的义务也没有。
THE LESSON
数据访问实现正义。 当原始数据被锁定时，错误和欺诈将成为永久性的。当数据可访问时，科学记录始终可以更正。
What fraction of published findings
实际上是真的吗？
实际上是真的吗？
复制危机
真实的DATA
The Open Science Collaboration (2015) attempted to replicate
100 篇已发表的心理学研究.
Original studies: 97% 报告了具有统计学意义的结果.
Replications: Only 36% achieved significance. Effect sizes dropped by half on average.
“复制危机”成为主流。我们自以为了解的——关于启动、自我损耗、权力摆姿势——远没有公布的那么确定。
Original studies: 97% 报告了具有统计学意义的结果.
Replications: Only 36% achieved significance. Effect sizes dropped by half on average.
“复制危机”成为主流。我们自以为了解的——关于启动、自我损耗、权力摆姿势——远没有公布的那么确定。
Open Science Collaboration (2015). Science. doi:10.1126/science.aac4716
The Numbers
97%
Original studies
significant
significant
36%
Replications
significant
significant
50%
Average effect
size drop
size drop
这意味着什么
对于你读到的每 3 个心理学发现， 2 可能无法复制。不是因为欺诈，而是因为噪音、分析的灵活性以及发布积极结果的压力。
两条路径
您正在根据已发表的心理学研究设计一个临床计划。你做什么工作？
Path A: Trust Original Publications
Build your program on published findings
↓
Invest resources in implementation
↓
发现基础不可靠—program fails to produce expected outcomes
路径B：检查复制
Search for independent replications first
↓
找到36%的复制率
↓
需要更有力的证据—pilot test, require multi-site replication before scale
THE REVELATION
"Published" does not mean "true."
复制是不容置疑的——这是科学方法正如预期的.
危机并不是科学失败。危机在于我们发现我们很少进行检查。
复制是不容置疑的——这是科学方法正如预期的.
危机并不是科学失败。危机在于我们发现我们很少进行检查。
THE LESSON
单一研究只是假设，而不是事实。 在制定研究政策之前，问问：这是否已被复制？由谁来？结果如何？
Four Stories, One Truth
1
Reinhart-Rogoff
分享您的数据。多年来，Excel 错误塑造了紧缩政策。
2
Duke Cancer Scandal
Require independent validation. Prestige does not equal truth.
3
RIAT Initiative
It is never too late. Buried evidence can be resurrected.
4
Replication Crisis
Published does not mean true. Replication is how science works.
“当他们没有验证时，
when they trusted without checking,
当他们认为享有盛誉的人是绝对正确的时 -
数字仍然是赤裸裸的，
人们支付了“
when they trusted without checking,
当他们认为享有盛誉的人是绝对正确的时 -
数字仍然是赤裸裸的，
人们支付了“
"But when they demanded proof,
when they shared their data,
when they replicated before they trusted—
真相浮出水面，
危害得到预防，
知识进步。”
when they shared their data,
when they replicated before they trusted—
真相浮出水面，
危害得到预防，
知识进步。”
1 / 6