IPD Course > The Vanishing
====================== 模块 1：消失（为什么 IPD 很重要） ====================
幻灯片1.2：汇总错觉
幻灯片 1.3：隐藏的异质性
幻灯片 1.5：发现
幻灯片1.6：副歌
==================== 模块 2：汇总 VS IPD ====================
幻灯片 2.2：两种方法
幻灯片 2.3：生态谬误
=====================模块3：CTT启示（他汀类药物） ====================
幻灯片 3.2：CTT
幻灯片 3.3：数字
幻灯片 3.4：发现
幻灯片3.5：为什么AD无法回答
幻灯片 4.2：决策框架
====================== 模块 5：一级 VS 两级 ====================
==================== 模块 6：皮质类固醇时间故事====================
幻灯片 6.3：IPD 解决方案
====================== 第 7 单元：获取 IPD（任务）====================
幻灯片 7.2：探索
幻灯片7.3：数据源
幻灯片7.4：协作信
======================模块8：数据协调===================
幻灯片 8.2：协调挑战
幻灯片 8.3：过程
幻灯片 8.5：验证步骤
====================== 模块 9：测试交互 ====================
==================== 模块 10：预后和预测====================
幻灯片 10.2：影响故事
===================== 模块 11：阿司匹林悖论 (ATT) ====================
幻灯片 11.2：Netflix 奖的故事
幻灯片 11.3：ATT 协作
幻灯片 11.4：启示
幻灯片11.5：决策树
==================== 第 12 单元：血压显示 (BPLTTC) ====================
幻灯片 12.2：辩论
幻灯片 12.3：协作
幻灯片 12.4：答案
幻灯片 12.5：J 曲线神话
===================== 模块13：中风时间窗口====================
幻灯片 13.2：风险
幻灯片 13.3：IPD 解决方案
幻灯片 13.4：时间曲线
幻灯片13.6：为什么 AD 无法回答
===================== 模块 14：IPD 中丢失数据 ====================
幻灯片 14.2：三种类型
幻灯片 14.3：决策树
==================== 第 15 单元：可用性陷阱 ====================
幻灯片15.2：达菲的故事
幻灯片 15.3：证据
幻灯片 15.4：比较
幻灯片 15.5：检查表
幻灯片 15.7：组合方法
====================== 第 16 单元：测验和参考资料 ====================
你没看到一千名患者
can be reduced to a single number,
在这个数字中， 生命被抹去?
can be reduced to a single number,
在这个数字中， 生命被抹去?
聚合错觉
EVERY META-ANALYSIS YOU'VE EVER READ
A meta-analysis reports: "Treatment reduces mortality by 15%."
但是哪些患者受益？年轻人还是老年人？是轻症还是重症？男性还是女性？
总体无法回答。
在平均水平内，一些患者获救，而另一些则受到伤害。
但是哪些患者受益？年轻人还是老年人？是轻症还是重症？男性还是女性？
总体无法回答。
在平均水平内，一些患者获救，而另一些则受到伤害。
What Aggregates Hide
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● Responders ● Non-responders
THE TRAGEDY
"Overall benefit: 30%"
But who 是 30%？如果没有个人数据，我们就无法识别响应者和非响应者。我们不能练习 precision medicine.
But who 是 30%？如果没有个人数据，我们就无法识别响应者和非响应者。我们不能练习 precision medicine.
Oxford, 1985
EARLY BREAST CANCER TRIALISTS' COLLABORATIVE GROUP
Richard Peto gathered something unprecedented: the raw data
on every woman in every tamoxifen trial.
不总结。不是平均值。 Each patient, each tumor, each outcome.
What he discovered changed breast cancer treatment forever.
不总结。不是平均值。 Each patient, each tumor, each outcome.
What he discovered changed breast cancer treatment forever.
Early Breast Cancer Trialists' Collaborative Group. Lancet 1988;2:61-72
已发表的试验表示：
"Tamoxifen works."
But the individual data revealed:
"Tamoxifen works."
But the individual data revealed:
ER+ vs ER−
Tamoxifen only worked in estrogen receptor positive tumors
隐藏了什么
没有一项试验足够大来检测这种相互作用。只有结合所有试验中的 个体患者数据 才能得出真相：
Giving tamoxifen to ER− patients was useless.
Giving tamoxifen to ER− patients was useless.
“总的来说，个体消失了。
在个体中，真相出现了。
这就是我们寻找隐藏患者的原因。“
在个体中，真相出现了。
这就是我们寻找隐藏患者的原因。“
这是个体参与者数据荟萃分析。
What is lost when we summarize?
What is found when we look closer?
What is found when we look closer?
Two Ways to Synthesize
Aggregate Data (AD)
- Study-level summaries
- 出版物中的影响大小
- Mean age, % male, etc.
- 快速且可访问
- Cannot see within-study variation
个人参与者数据(IPD)
- 患者层面的原始数据
- Every participant's characteristics
- Actual ages, actual outcomes
- Time-intensive to obtain
- Can see who responds and who doesn't
生态谬误
综合思维的危险
Imagine two trials of a drug for heart failure.
Trial A: Mean age 55 years, Effect size 0.70 (benefit)
Trial B: Mean age 75 years, Effect size 0.90 (less benefit)
Tempting conclusion: "The drug works better in younger patients."
But this could be completely wrong.
Trial A: Mean age 55 years, Effect size 0.70 (benefit)
Trial B: Mean age 75 years, Effect size 0.90 (less benefit)
Tempting conclusion: "The drug works better in younger patients."
But this could be completely wrong.
The Trap
ECOLOGICAL FALLACY
在每个试验中，年龄和治疗效果之间的关系可能是 completely different 来自试验间的关系。
Perhaps in Trial A, older patients within that trial responded better. Perhaps in Trial B, younger patients within that trial responded better.
如果没有个体，你就无法知道数据。
Perhaps in Trial A, older patients within that trial responded better. Perhaps in Trial B, younger patients within that trial responded better.
如果没有个体，你就无法知道数据。
THE RULE
Between-study relationships ≠ Within-study relationships
Aggregate data can only show between-study patterns
What IPD Enables
1
True Effect Modification
Test whether treatment effect varies by patient characteristics (age, biomarkers, disease severity)
2
Time-to-Event Analysis
Use actual survival curves, not just hazard ratios. Handle censoring properly.
3
Consistent Definitions
Standardize outcome definitions, exposure timing, covariate categories across studies
4
Subgroup Credibility
测试研究中的相互作用，避免生态谬误
“汇总数据显示森林。
Individual data shows each tree.
如果您需要知道哪些树生病了—
you must walk among them."
Individual data shows each tree.
如果您需要知道哪些树生病了—
you must walk among them."
你没看到 one collaboration
gathered data on 170,000 patients
and answered questions no single trial could ask?
gathered data on 170,000 patients
and answered questions no single trial could ask?
Oxford, 1994-Present
CHOLESTEROL TREATMENT TRIALISTS' COLLABORATION
CTT 从 every major statin trial.
27 trials. 174,149 patients. Every baseline characteristic. Every cardiovascular event. Every death.
The published trials asked: "Do statins work?"
收集个人数据 CTT 问道： “为谁做他汀类药物有效吗？”
27 trials. 174,149 patients. Every baseline characteristic. Every cardiovascular event. Every death.
The published trials asked: "Do statins work?"
收集个人数据 CTT 问道： “为谁做他汀类药物有效吗？”
CTT Collaboration. Lancet 2010;376:1670-81
IPD 的规模
27
Trials Combined
174,149
Individual Patients
5
多年的数据收集
THE INVESTMENT
收集 IPD 需要多年的协商、数据传输、清理和收集
但是您可以回答的问题 值得投资.
但是您可以回答的问题 值得投资.
What CTT Discovered
✓
Benefit Proportional to LDL Reduction
Every 1 mmol/L LDL reduction = 22% lower CV events. True across all subgroups.
✓
No Age Threshold
Benefit continues even in patients >75 years (contradicting earlier AD analyses)
✓
Primary Prevention Works
Patients without prior CVD benefit proportionally to their baseline risk
✗
No Cancer Signal
关于他汀类药物致癌的担忧已被 IPD 明确驳斥后续
为什么聚合数据失败
THE LIMITATION
Published trials categorized patients differently.
Trial A: "High risk" = 10-year CVD risk >20%
Trial B: "High risk" = Prior MI
Trial C: "High risk" = Diabetes
You cannot compare or combine what is defined differently.
IPD allowed the CTT to redefine everyone consistently.
Trial A: "High risk" = 10-year CVD risk >20%
Trial B: "High risk" = Prior MI
Trial C: "High risk" = Diabetes
You cannot compare or combine what is defined differently.
IPD allowed the CTT to redefine everyone consistently.
"When trials speak different languages,
IPD提供翻译。
What seemed contradictory becomes clear:
the same truth, measured differently."
IPD提供翻译。
What seemed contradictory becomes clear:
the same truth, measured differently."
IPD takes years to gather.
聚合数据需要 weeks.
什么时候值得投资
聚合数据需要 weeks.
什么时候值得投资
决策树
您应该追求 IPD 吗？
您的问题是关于治疗效果修改吗？
↓
YES
"Who benefits most?"
NO
"Does it work overall?"
↓
IPD likely needed
AD may suffice
When IPD Is Essential
1
Time-to-Event Outcomes
当生存曲线重要而不仅仅是最终风险比时当您需要处理审查时。
2
Continuous Effect Modifiers
Testing whether treatment effect varies by age, BMI, biomarker level (not just "high" vs "low")
3
Outcome Definition Problems
当试验以不同的方式定义结果并且您需要标准化
4
Longer Follow-Up Available
当试验者有未发表的后续数据时，您希望包括
🔀
研究策略决策
您正在计划对皮质类固醇治疗 COVID-19 肺炎进行荟萃分析。一位同事问：“我们应该尝试获得 IPD，还是只进行汇总分析？”
您的关键问题是：“益处是否因疾病严重程度和治疗时间而异？”
您的关键问题是：“益处是否因疾病严重程度和治疗时间而异？”
您有什么建议？
A
只需进行 AD - 我们需要快速获得结果，IPD 也需要时间长
B
追求 IPD - 我们的问题是关于患者特征的效果修改
C
根据出版物中的试验级别严重性类别进行亚组分析
当聚合数据足够时
✓
Overall Treatment Effect
当您唯一的问题是“它有效吗？”不是“为谁？”
✓
Homogeneous Population
When trials enrolled similar patients and effect modification is unlikely
✓
Binary Outcomes, Short Follow-up
当审查不是问题并且结果很简单是/否时
✓
IPD Unobtainable
当审判者不共享、数据丢失或资源不可用时
“并非每个问题都需要个人。
但每个问题关于 which individuals
要求它们存在于您的数据中。”
但每个问题关于 which individuals
要求它们存在于您的数据中。”
您已收集了各个数据。
现在：您是否将其分析为 one combined dataset
or trial by trial, then combine?
现在：您是否将其分析为 one combined dataset
or trial by trial, then combine?
Two Analytical Approaches
Two-Stage Approach
- Stage 1: Analyze each trial separately
- 第2阶段：对结果进行元分析
- Preserves trial structure
- Familiar (like standard MA)
- Cannot handle sparse data well
One-Stage Approach
- 同时分析所有数据
- Mixed-effects regression model
- 聚类的随机效应
- 更适合稀疏数据
- More flexible modeling
两阶段：熟悉的路径
STAGE 1 (Within Each Trial)
Estimate treatment effect for trial k: θk
使用来自该试验的个体患者数据
STAGE 2 (Across Trials)
Meta-analyze θ1, θ2, ... θK
使用标准随机效应方法
ADVANTAGE
Easy to understand. Easy to explain.
Each trial's estimate is transparent.
熟悉的森林图和我2 statistics.
Each trial's estimate is transparent.
熟悉的森林图和我2 statistics.
一步：强大的路径
MIXED-EFFECTS MODEL
Yij = β0 + β1Treatmentij + uj + εij
i = patient, j = trial, uj = random trial effect
ADVANTAGES
1. Handles trials with zero events (no continuity corrections)
2. More powerful for detecting interactions
3. Can model complex covariate relationships
4. Exact likelihood (no normal approximations)
2. More powerful for detecting interactions
3. Can model complex covariate relationships
4. Exact likelihood (no normal approximations)
选择你的方法
One-Stage or Two-Stage?
Do trials have sparse events (rare outcomes)?
↓
YES (sparse)
↓
One-stage preferred
Avoids zero-cell problems
NO (common events)
↓
Either works
结果应该类似
"The two-stage preserves each trial's voice.
The one-stage hears all voices at once.
当数据稀疏且事件罕见时 -
一级捕获两级错过的内容。”
The one-stage hears all voices at once.
当数据稀疏且事件罕见时 -
一级捕获两级错过的内容。”
您没有听说过拯救婴儿生命的药物
的故事吗——
but only if given at the right time?
的故事吗——
but only if given at the right time?
早产儿悖论
NEONATAL INTENSIVE CARE UNITS WORLDWIDE
产前皮质类固醇可以降低早产儿的死亡率。 1970 年代。
But a puzzle remained: When should they be given?
24 hours before birth? 48 hours? A week?
Published trials couldn't answer—they didn't report timing consistently.
But a puzzle remained: When should they be given?
24 hours before birth? 48 hours? A week?
Published trials couldn't answer—they didn't report timing consistently.
Roberts D, et al. Cochrane Database Syst Rev 2017
IPD 揭示了窗口
24h-7d
Optimal Window
>7 days
Benefit Wanes
THE DISCOVERY
By examining each baby's exact time from steroid to delivery,
IPD meta-analysis showed:
Maximum benefit: 24 hours to 7 days before birth
Reduced benefit: >7 days (lung maturity effect fades)
No benefit: <24 hours (not enough time to work)
Maximum benefit: 24 hours to 7 days before birth
Reduced benefit: >7 days (lung maturity effect fades)
No benefit: <24 hours (not enough time to work)
Lives Saved, Practice Changed
临床影响
在 IPD 之前：临床医生给予类固醇并希望最好。
After IPD: Guidelines now recommend repeat dosing 如果在第一个疗程后 7 天内未发生分娩。
没有个体数据就不可能实现这种精确度 saved thousands of premature babies.
After IPD: Guidelines now recommend repeat dosing 如果在第一个疗程后 7 天内未发生分娩。
没有个体数据就不可能实现这种精确度 saved thousands of premature babies.
Timing is Treatment
在错误的时间给予相同的药物，也可能是安慰剂
为什么汇总数据无法显示这个
THE PROBLEM
Trial A reported: "Steroid given within 48 hours"
Trial B reported: "Steroid given antenatally"
Trial C reported: "Steroid-to-delivery interval: median 3 days"
Different categories. Different definitions. Incompatible summaries.
Only by examining each baby's actual steroid-to-delivery time could the optimal window be identified.
Trial B reported: "Steroid given antenatally"
Trial C reported: "Steroid-to-delivery interval: median 3 days"
Different categories. Different definitions. Incompatible summaries.
Only by examining each baby's actual steroid-to-delivery time could the optimal window be identified.
“已知该药物有效。
But when 如何给予它是未知的。
IPD 将“某个时间”变成“正确的时间”——
在这种精确度下，孩子们”
But when 如何给予它是未知的。
IPD 将“某个时间”变成“正确的时间”——
在这种精确度下，孩子们”
数据是存在的。
在文件和数据库的某个地方，
每个患者的故事都被记录.
问题是：他们会分享它吗？
在文件和数据库的某个地方，
每个患者的故事都被记录.
问题是：他们会分享它吗？
寻找数据
60-80%
Typical IPD Retrieval Rate
6-24
Months to Gather
REALITY CHECK
您可能 not get 100% of trials.
Some investigators will refuse. Some data is lost.
Some companies won't share.
This is expected. Plan for it.
This is expected. Plan for it.
Where IPD Lives
1
Trialist Collaboration
Direct contact with trial investigators. Build relationships. Offer co-authorship.
2
数据共享平台
YODA Project, ClinicalStudyDataRequest.com, Vivli, ICPSR
3
Regulatory Agencies
EMA Policy 0070, FDA (limited), Health Canada
4
Journal Requirements
Many journals now require data sharing; check supplementary materials
The Approach
HOW TO ASK
1。提出您的问题—explain why IPD is essential
2. Offer co-authorship—make sharing worthwhile
3。描述数据安全—how you'll protect their patients
4.提供数据字典—specify exactly what you need
5. Set clear timelines—respect their time
2. Offer co-authorship—make sharing worthwhile
3。描述数据安全—how you'll protect their patients
4.提供数据字典—specify exactly what you need
5. Set clear timelines—respect their time
可用性偏差陷阱
THE DANGER
如果 share data 的试验与 don't share?
Industry trials: less likely to share
Negative trials: less likely to share
Older trials: data may be lost
Your IPD sample may be biased.
Industry trials: less likely to share
Negative trials: less likely to share
Older trials: data may be lost
Your IPD sample may be biased.
THE RULE
Always compare: IPD trials vs. non-IPD trials
Do they differ in effect size, sample size, funding source?
的试验系统性不同怎么办？“数据存在。问题在于信任。
Will they share what they have guarded?
搭建桥梁小心——
因为在那座桥上，患者的未来交叉。“
Will they share what they have guarded?
搭建桥梁小心——
因为在那座桥上，患者的未来交叉。“
您已经收集了数据
from twelve trials, five countries, three decades.
But Trial A calls it "cardiovascular death"
和试验B称之为 "cardiac mortality".
它们相同吗？
from twelve trials, five countries, three decades.
But Trial A calls it "cardiovascular death"
和试验B称之为 "cardiac mortality".
它们相同吗？
巴别塔问题
EVERY IPD META-ANALYSIS
Trial from Japan: Age in years since 1900 (e.g., "54" = 1954 birth)
Trial from USA: Age in decimal years (e.g., 65.7)
来自英国的审判：年龄范围("65-74")
Diabetes: HbA1c ≥ 6.5% vs. fasting glucose ≥ 126 vs. "physician diagnosis"
Outcome: "Major adverse cardiac event" (one trial includes stroke, another doesn't)
分析之前：协调一切。
Trial from USA: Age in decimal years (e.g., 65.7)
来自英国的审判：年龄范围("65-74")
Diabetes: HbA1c ≥ 6.5% vs. fasting glucose ≥ 126 vs. "physician diagnosis"
Outcome: "Major adverse cardiac event" (one trial includes stroke, another doesn't)
分析之前：协调一切。
协调过程
1
创建主数据字典
Define every variable you need: name, type, permitted values, derivation rules
2
Map Each Trial's Variables
Document how each trial's coding maps to your standardized definitions
3
检查试用者
验证您的解释。他们比您更了解自己的数据。
4
Validate Transformations
重现 IPD 发布的结果。如果它们不匹配，请进行调查。
🔀
定义困境
您正在协调 8 项糖尿病试验的 IPD。您的主要结局是“心血管死亡”。
六项试验使用了标准定义（ICD 代码）。 20 世纪 90 年代的两项试验使用了“研究者评估的心源性死亡”，但没有标准化标准。这两项试验显示出更大的治疗效果。
六项试验使用了标准定义（ICD 代码）。 20 世纪 90 年代的两项试验使用了“研究者评估的心源性死亡”，但没有标准化标准。这两项试验显示出更大的治疗效果。
您如何处理这个问题？
A
包括所有 8 项试验 - 数据越多越好，定义“足够接近”
B
Exclude the 2 non-standardized trials to maintain outcome consistency
C
6 项一致试验的初步分析；添加另外 2 个内容的敏感性分析
关键验证
THE TEST
在任何新分析之前：
从 IPD 重现每个试验已发布的结果。
如果您的分析给出 RR = 0.78，但出版物说 RR = 0.85，
something is wrong.
Find the discrepancy. Fix it. Then proceed.
从 IPD 重现每个试验已发布的结果。
如果您的分析给出 RR = 0.78，但出版物说 RR = 0.85，
something is wrong.
Find the discrepancy. Fix it. Then proceed.
"Different languages, different rulers,
different ways to name the same disease.
在合并之前，您必须翻译。
在翻译之前，您必须了解。”
different ways to name the same disease.
在合并之前，您必须翻译。
在翻译之前，您必须了解。”
治疗有效 on average.
但是它对于以下情况是否同样有效年轻人和老年人？
对于轻度和重度？
对于有生物标志物的人和没有生物标志物的人？
但是它对于以下情况是否同样有效年轻人和老年人？
对于轻度和重度？
对于有生物标志物的人和没有生物标志物的人？
改变肿瘤学的相互作用
EBCTCG, 1985-1998
The Early Breast Cancer Trialists' Group didn't just ask "Does tamoxifen work?"
They asked: "Does the effect differ by estrogen receptor status?"
相互作用是巨大的：
ER-positive: 47% reduction in recurrence
ER-negative: No benefit at all
They asked: "Does the effect differ by estrogen receptor status?"
相互作用是巨大的：
ER-positive: 47% reduction in recurrence
ER-negative: No benefit at all
EBCTCG. Lancet 1998;351:1451-67
pinteraction < 0.00001
The most important interaction in cancer medicine
Testing Treatment-Covariate Interactions
THE MODEL
Outcome = β1Treatment + β2Covariate + β3(Treatment × Covariate)
β3 is the interaction term - does treatment effect vary by covariate?
INTERPRETATION
If β3 is statistically significant:
治疗效果 differs between subgroups defined by the covariate.
如果不是：治疗效果为 similar across subgroups (or you lack power to detect a difference).
治疗效果 differs between subgroups defined by the covariate.
如果不是：治疗效果为 similar across subgroups (or you lack power to detect a difference).
关键区别
Between-Study Interaction
- Compares trial-level averages
- Ecological fallacy risk
- Confounded by trial design
- Low statistical power
- 可以使用聚合数据
Within-Study Interaction
- Compares patients within each trial
- No ecological fallacy
- Randomization preserved
- Much higher power
- Requires IPD
关键见解
IPD允许您测试 within-study interactions,
where randomization ensures the comparison is fair.
这是效果的黄金标准修改。
这是效果的黄金标准修改。
🔀
年龄相互作用问题
您的房颤抗凝 IPD 荟萃分析包括 12,000 名患者。您测试了年龄相互作用，发现 p=0.03。≥75 岁的患者比 <75 岁的患者 (RR 0.55) 显示出更大的获益。 0.72)。
您在分析中测试了 8 种潜在的效果调节剂。
您在分析中测试了 8 种潜在的效果调节剂。
您如何解释这一发现？
A
报告抗凝治疗在老年患者中效果更好 - 相互作用显着
B
Apply multiple testing correction; with 8 tests, p=0.03 is likely a false positive
C
Report as hypothesis-generating, note the multiple comparisons, request external validation
“平均值隐藏了
The interaction reveals who benefits.
在研究中进行测试，而不是在—
之间进行测试，因为这才是发现真相的地方。”
The interaction reveals who benefits.
在研究中进行测试，而不是在—
之间进行测试，因为这才是发现真相的地方。”
IPD 不仅仅用于治疗。
当您寻求 predict who will die,
who will recover, who will relapse—
个人就是一切.
当您寻求 predict who will die,
who will recover, who will relapse—
个人就是一切.
Predicting Recovery After Brain Injury
IMPACT CONSORTIUM, 2005-PRESENT
创伤后脑损伤后，家属提出了毁灭性的问题：
"Will they ever wake up?"
Individual trials were too small to develop accurate prediction models.
IMPACT 从 11 项研究、9,205 名患者中收集了 IPD，并建立了一个模型，根据初始临床特征预测 6 个月的结果。
"Will they ever wake up?"
Individual trials were too small to develop accurate prediction models.
IMPACT 从 11 项研究、9,205 名患者中收集了 IPD，并建立了一个模型，根据初始临床特征预测 6 个月的结果。
IMPACT Study Group. PLoS Med 2008;5:e165
Why Prognosis Demands IPD
1
External Validation Across Populations
Develop in some studies, validate in others. True test of generalizability.
2
Non-linear Relationships
Explore how predictors relate to outcome: linear? threshold? U-shaped?
3
Multiple Predictor Interactions
Age + GCS + pupil reactivity may interact in ways aggregate data cannot reveal
4
Proper Handling of Missing Predictors
患者层面而非研究层面的多重插补
Reporting Standards
TRIPOD
- Prediction model reporting
- 开发和验证
- 校准和辨别
- 22 item checklist
PRISMA-IPD
- IPD meta-analysis reporting
- 数据采集详细信息
- Harmonization process
- Integrity checking
用于IPD预测模型
You need both：TRIPOD用于模型本身，PRISMA-IPD用于数据合成过程。
When Prediction Saves Lives
临床影响
The IMPACT model is now embedded in clinical decision support systems worldwide.
当患有创伤性脑损伤的患者到达时，模型会提供 probability of survival and probability of favorable outcome.
这指导与家人的对话。这告知治疗强度。这有助于分配 ICU 资源。
根据个人数据构建。为个体患者服务。
当患有创伤性脑损伤的患者到达时，模型会提供 probability of survival and probability of favorable outcome.
这指导与家人的对话。这告知治疗强度。这有助于分配 ICU 资源。
根据个人数据构建。为个体患者服务。
"To predict one patient's future,
您必须从数千个过去中学习。
IPD 拥有这些故事 —
each one a teacher, if you will listen."
您必须从数千个过去中学习。
IPD 拥有这些故事 —
each one a teacher, if you will listen."
你有没有考虑过为
送给他们心脏的白色小药片 millions ？
他们被告知： “拿着这个，你就会受保护。”
But was every heart equally in need of protection?
送给他们心脏的白色小药片 millions ？
他们被告知： “拿着这个，你就会受保护。”
But was every heart equally in need of protection?
Netflix 奖的故事
A TRUE STORY
In 2006, Netflix released 100 million "anonymized" movie ratings for a $1 million prize.
Researchers quickly re-identified users by matching ratings with public IMDb reviews.
One lawsuit alleged a closeted lesbian was outed through her viewing patterns.
The lesson: removing names isn't anonymization.
IPD 包含年龄、诊断、治疗反应和日期的足够组合，以唯一地识别个人。隐私需要的不仅仅是删除姓名列，还需要了解数据组合如何成为指纹。
Researchers quickly re-identified users by matching ratings with public IMDb reviews.
One lawsuit alleged a closeted lesbian was outed through her viewing patterns.
The lesson: removing names isn't anonymization.
IPD 包含年龄、诊断、治疗反应和日期的足够组合，以唯一地识别个人。隐私需要的不仅仅是删除姓名列，还需要了解数据组合如何成为指纹。
Oxford, 2009
ANTITHROMBOTIC TRIALISTS' COLLABORATION
ATT 从 95,000 patients
across 6 primary prevention trials of aspirin.
The published trials said: "Aspirin prevents heart attacks."
收集个人数据，但 ATT 提出了禁止的问题：
“代价是什么？以及谁？”
The published trials said: "Aspirin prevents heart attacks."
收集个人数据，但 ATT 提出了禁止的问题：
“代价是什么？以及谁？”
ATT Collaboration. Lancet 2009;373:1849-60
揭示的数字
-12%
Heart Attacks Prevented
+32%
Major Bleeds Caused
THE BALANCE
For every 1,000 low-risk patients 服用阿司匹林 5 年：
2-3 heart attacks prevented
2-3 major bleeds caused
益处和危害相互抵消。
2-3 heart attacks prevented
2-3 major bleeds caused
益处和危害相互抵消。
基于风险的决策
该患者是否应该服用阿司匹林进行一级预防？
10-Year CVD Risk?
↓
>20%
High Risk
10-20%
Intermediate
<10%
Low Risk
↓
Consider aspirin
(benefit > harm)
(benefit > harm)
Shared decision
(benefit ≈ harm)
(benefit ≈ harm)
Avoid aspirin
(harm > benefit)
(harm > benefit)
Why did published meta-analyses miss this?
聚合的盲目性
Published trials reported average benefits.
他们无法表明高风险患者获得
while low-risk patients lost.
Only by examining each patient's baseline risk,
each patient's outcomes,
could the interaction be revealed.
他们无法表明高风险患者获得
while low-risk patients lost.
Only by examining each patient's baseline risk,
each patient's outcomes,
could the interaction be revealed.
🔀
初级预防困境
一名 52 岁男子询问阿司匹林是否可以预防心脏病发作。他既往没有心血管疾病，胆固醇处于临界值，血压得到控制，不吸烟。他的 10 年 CVD 风险为 8%。
他在网上读到“阿司匹林可以预防心脏病”并希望得到您的建议。
他在网上读到“阿司匹林可以预防心脏病”并希望得到您的建议。
您告诉他什么？
A
"Yes, take aspirin daily - it's proven to prevent heart attacks"
B
"At your risk level, aspirin's bleeding risk roughly equals its heart benefit - I'd recommend against it"
C
"Take baby aspirin - the lower dose is safer"
“拯救濒临灭绝的药物
可能会伤害安全。
在开处方前了解患者的风险。
这是个人数据告诉我们的。“
可能会伤害安全。
在开处方前了解患者的风险。
这是个人数据告诉我们的。“
95,000 patients. One truth: risk determines benefit.
你没看到他们争论了几十年吗：
“老人应该和年轻人一样对待吗？”
Some said: "Lower is always better."
还有人说：“老人很脆弱，要小心。”
谁是对的？
“老人应该和年轻人一样对待吗？”
Some said: "Lower is always better."
还有人说：“老人很脆弱，要小心。”
谁是对的？
J 曲线争议
CARDIOLOGY CONFERENCES, 1980s-2000s
多年来，医生们争论：
激进阵营： "Every mmHg of BP reduction saves lives.
Treat everyone to target 120/80."
保守阵营： "In the elderly, low BP causes falls, strokes, death.
There's a J-curve—too low is dangerous."
Individual trials were too small to settle it. 直到BPLTTC 收集了个人数据。
激进阵营： "Every mmHg of BP reduction saves lives.
Treat everyone to target 120/80."
保守阵营： "In the elderly, low BP causes falls, strokes, death.
There's a J-curve—too low is dangerous."
Individual trials were too small to settle it. 直到BPLTTC 收集了个人数据。
The Blood Pressure Lowering Treatment Trialists
48
Trials Combined
344,716
Individual Patients
15+
Years of Follow-up
THE QUESTION
Does the benefit of BP lowering vary by age?
Is there a J-curve at very low pressures?
Do the very old (>80 years) still benefit?
Is there a J-curve at very low pressures?
Do the very old (>80 years) still benefit?
BPLTTC. Lancet 2021;397:1625-36
最后的答案
Benefit at Every Age
从 55 岁到 55 岁，比例风险降低是一致的85+
WHAT IPD SHOWED
Every 5 mmHg BP reduction:
Age 55-64: 10% lower major CV events
Age 65-74: 10% lower major CV events
Age 75-84: 10% lower major CV events
Age ≥85: Still 10% lower
pinteraction = 0.85。没有年龄修改的证据。
Age 55-64: 10% lower major CV events
Age 65-74: 10% lower major CV events
Age 75-84: 10% lower major CV events
Age ≥85: Still 10% lower
pinteraction = 0.85。没有年龄修改的证据。
J 曲线怎么样？
THE RESOLUTION
BPLTTC 通过检查患者来测试 J 曲线
who achieved very low blood pressures.
Result: No J-curve in randomized comparisons.
观察数据中的表观 J 曲线 reverse causation— sick patients have low BP because they're sick, not sick because their BP is low.
只有随机对照试验中的 IPD 才能解决这个问题。
who achieved very low blood pressures.
Result: No J-curve in randomized comparisons.
观察数据中的表观 J 曲线 reverse causation— sick patients have low BP because they're sick, not sick because their BP is low.
只有随机对照试验中的 IPD 才能解决这个问题。
Treating Hypertension by Age
Should This Elderly Patient Get BP Treatment?
Patient Age?
↓
55-74
75-84
85+
↓
Treat if BP elevated
Proportional benefit maintained across all ages
Proportional benefit maintained across all ages
THE LESSON
Age alone is not a reason to withhold treatment.
Consider frailty, life expectancy, patient preference—but not age.
Consider frailty, life expectancy, patient preference—but not age.
🔀
82 岁老人的血压
一位 82 岁女性的血压为 158/88。她很独立，认知能力完好，并且没有跌倒史。一位同事说：“她太老了，不适合积极的血压治疗。J 曲线，你知道。”
你记得 BPLTTC IPD 荟萃分析。
你记得 BPLTTC IPD 荟萃分析。
What do you do?
A
同意你同事的观点 - 接受老年人更高的血压目标
B
引用 BPLTTC 证据表明老年人仍能受益，并发起治疗
C
Wait for a cardiovascular event before treating
“他们说老人太脆弱了，无法接受药物。
但个人数据显示并非如此：
At every age, the benefit endures.
Do not let age alone deny protection."
但个人数据显示并非如此：
At every age, the benefit endures.
Do not let age alone deny protection."
有您没有目睹与时间的赛跑
when a clot blocks the brain?
Every minute, two million neurons die.
But when does the window close?
When is it too late to intervene?
when a clot blocks the brain?
Every minute, two million neurons die.
But when does the window close?
When is it too late to intervene?
Time Is Brain
1.9M
Neurons Lost Per Minute
14B
Synapses Lost Per Minute
THE QUESTION
溶栓可以溶解血栓并恢复血流。
But given too late, it causes bleeding into dying brain tissue.
What is the time window? 3 hours? 4.5 hours? 6 hours?
Individual trials disagreed. Guidelines were uncertain.
But given too late, it causes bleeding into dying brain tissue.
What is the time window? 3 hours? 4.5 hours? 6 hours?
Individual trials disagreed. Guidelines were uncertain.
中风溶栓试验者
STT COLLABORATIVE GROUP, 2014
STT 从 6,756 patients
across 9 major alteplase trials.
They knew exactly when each patient's stroke began. They knew exactly when thrombolysis was given. They knew exactly who lived, who died, who recovered.
They could map benefit against time, minute by minute.
They knew exactly when each patient's stroke began. They knew exactly when thrombolysis was given. They knew exactly who lived, who died, who recovered.
They could map benefit against time, minute by minute.
STT Collaborative Group. Lancet 2014;384:1929-35
衰落窗口
0-3h
OR 1.75 获得良好结果
3-4.5h
OR 1.26 获得良好结果良好的结果
4.5-6h
OR 1.15 (not significant)
效益的衰减
Every 15 minutes of delay reduced the benefit.
Treated at 90 min: 1 in 4 achieve excellent outcome
Treated at 180 min: 1 in 7 achieve excellent outcome
Treated at 270 min: 1 in 14 achieve excellent outcome
Treated at 90 min: 1 in 4 achieve excellent outcome
Treated at 180 min: 1 in 7 achieve excellent outcome
Treated at 270 min: 1 in 14 achieve excellent outcome
溶栓决定
Acute Ischemic Stroke: Give Thrombolysis?
Time Since Symptom Onset?
↓
0-3h
Strong benefit
3-4.5h
Moderate benefit
4.5-6h
Uncertain
>6h
Harm likely
↓
TREAT
URGENTLY
URGENTLY
TREAT
if eligible
if eligible
Consider
imaging
imaging
Generally
avoid
avoid
Why did we need individual data?
的局限性平均值
Trial A: "Patients treated within 3 hours" (average: 2.1 hours)
Trial B: "Patients treated within 6 hours" (average: 4.2 hours)
These overlapping, inconsistent windows couldn't be compared.
Only by knowing each patient's exact time
could the continuous decay of benefit be mapped.
Trial B: "Patients treated within 6 hours" (average: 4.2 hours)
These overlapping, inconsistent windows couldn't be compared.
Only by knowing each patient's exact time
could the continuous decay of benefit be mapped.
🔀
4 小时中风
一名 68 岁的患者在 4 小时前出现症状。CT 显示没有出血，NIHSS 评分为 12（中等）。家人问：“使用溶栓药物是否太晚了？”
Door-to-needle time will add 30 minutes, making total time ~4.5 hours.
Door-to-needle time will add 30 minutes, making total time ~4.5 hours.
What do you do?
A
"It's too late - the 3-hour window has passed"
B
“我们仍在 4.5 小时窗口内 - 紧急进行溶栓治疗”
C
"Wait for MRI to see if there's salvageable tissue"
"Every minute the clot remains,
two million neurons perish.
IPD 向我们展示了逐渐消失的窗口。
Act quickly, or the window closes forever."
two million neurons perish.
IPD 向我们展示了逐渐消失的窗口。
Act quickly, or the window closes forever."
您已收集了各个数据。
您打开充满希望的文件。
然后您会看到它： empty cells.
Age: 67. Sex: Male. Smoking status: missing.
Outcome at 1 year: missing.
What now?
您打开充满希望的文件。
然后您会看到它： empty cells.
Age: 67. Sex: Male. Smoking status: missing.
Outcome at 1 year: missing.
What now?
三种缺失
1
Missing Completely at Random (MCAR)
Lab machine broke randomly. No relation to patient characteristics. Safe to ignore (but wasteful).
2
Missing at Random (MAR)
Older patients more likely to miss follow-up. Missingness related to observed variables. Imputation can help.
3
非随机缺失(MNAR)
结果不佳的患者退出。与 missing value itself. 危险相关的失踪。需要敏感性分析。
处理缺失数据
如何处理缺失值？
How much is missing?
↓
<5%
Minor
5-20%
Moderate
>20%
Substantial
↓
Complete case
may suffice
may suffice
Multiple
imputation
imputation
Impute +
sensitivity
sensitivity
多重插补：标准
THE PROCESS
1. Impute → 2. Analyze → 3. Pool
Create M imputed datasets, analyze each, combine using Rubin's rules
WHY MULTIPLE?
单一插补假装您知道缺失值。
Multiple imputation (M=20-50 datasets) reflects uncertainty 关于缺失值可能是什么。
这保留了有效的标准误差和 p 值。
Multiple imputation (M=20-50 datasets) reflects uncertainty 关于缺失值可能是什么。
这保留了有效的标准误差和 p 值。
IPD 特定缺失数据挑战
1
Systematically Missing Variables
Trial A measured biomarker X. Trial B didn't. Can't impute what was never collected.
2
Multilevel Structure
Patients nested within trials. Imputation model must account for clustering.
3
Different Follow-up Durations
试验 A 持续了 2 年。试验B为期5年。生存分析需要小心。
When a Variable Is Missing Entirely
Trial Didn't Collect Your Key Covariate
Is it an effect modifier or confounder?
↓
Effect Modifier
↓
Restrict interaction
analysis to trials
使用变量
analysis to trials
使用变量
Confounder
↓
Sensitivity analysis
varying assumptions
about unmeasured
varying assumptions
about unmeasured
🔀
缺失的生物标志物
您的 IPD-MA 测试新的癌症药物在生物标志物阳性患者中是否效果更好。您在 8 次试验中获得了 IPD。但 3 项试验（40% 的患者）根本没有测量生物标志物。
这 3 项试验往往时间较长且规模较小。
这 3 项试验往往时间较长且规模较小。
如何分析相互作用？
A
Impute the biomarker status based on other patient characteristics
B
仅分析 5 项试验中生物标志物数据的相互作用；敏感性分析，包括所有试验的总体效果
C
从荟萃分析中完全排除 3 项试验
“空单元格并不是什么都没有。
It is a question: 为什么这是未知的？
在填写之前回答该问题差距——
因为缺席的原因决定了解决方案。”
It is a question: 为什么这是未知的？
在填写之前回答该问题差距——
因为缺席的原因决定了解决方案。”
你没见过研究人员
谁吗？从 willing trialists
收集数据并宣布胜利？
But the unwilling held secrets.
这些秘密改变了一切。
谁吗？从 willing trialists
收集数据并宣布胜利？
But the unwilling held secrets.
这些秘密改变了一切。
达菲的故事
A TRUE STORY
多年来，政府储备了用于治疗大流行流感的达菲，花费了数十亿美元。
Cochrane reviewers requested trial data to verify efficacy. Roche refused, citing confidentiality.
五年来，英国医学杂志 (BMJ) 一直致力于提高透明度。当完整的临床研究报告最终于 2014 年发布时，情况发生了变化：达菲将症状持续时间缩短了不到一天，并且不能预防并发症。
在这种药物上花费了数十亿美元，但其完整证据却被锁住了。
达菲传奇改变了人们的期望——如今，临床试验透明度正在成为常态，而不是例外。
Cochrane reviewers requested trial data to verify efficacy. Roche refused, citing confidentiality.
五年来，英国医学杂志 (BMJ) 一直致力于提高透明度。当完整的临床研究报告最终于 2014 年发布时，情况发生了变化：达菲将症状持续时间缩短了不到一天，并且不能预防并发症。
在这种药物上花费了数十亿美元，但其完整证据却被锁住了。
达菲传奇改变了人们的期望——如今，临床试验透明度正在成为常态，而不是例外。
When IPD Is Selectively Available
经验发现
研究比较了共享 IPD 的试验与不共享 IPD 的试验。请勿：
Industry-sponsored trials: Less likely to share
负面结果的试验： Less likely to share
Older trials: 数据经常丢失
If these trials systematically differ in effect size,
您的 IPD-MA 存在偏差。
Industry-sponsored trials: Less likely to share
负面结果的试验： Less likely to share
Older trials: 数据经常丢失
If these trials systematically differ in effect size,
您的 IPD-MA 存在偏差。
Ahmed I, et al. BMJ 2012;344:d7762
Always Compare: IPD vs. Non-IPD Trials
Assessing Availability Bias
Compare AD effects: trials with IPD vs. without
↓
Similar Effects
↓
Low concern
for bias
for bias
Different Effects
↓
High concern
敏感性分析需要
敏感性分析需要
可用性偏差检查表
✓
Report IPD retrieval rate
“我们从 12/15 次试验中获得了 IPD (80%)”
✓
Compare IPD vs. non-IPD trial characteristics
样本大小、资金、发表日期、汇总效果大小数据
✓
Sensitivity analysis including non-IPD trials
结合非共享试验的IPD + AD的两阶段分析
✓
讨论非共享的原因
Data lost? Refused? Never requested? Each has different implications.
🔀
不情愿的试验者
您的手术干预的IPD-MA从12个合格试验中的8个获得数据（67%）。 4 项非共享试验均由行业资助，与 8 项 IPD 试验（汇总 OR 0.72）相比，其出版物中显示的效益较小（汇总 OR 0.95）。
您的 IPD 分析显示 OR 0.70。
您的 IPD 分析显示 OR 0.70。
您如何报告此情况？
A
Report IPD result (OR 0.70) as the main finding - IPD is gold standard
B
报告 IPD结果带有关于可用性偏差和敏感性分析的显着警告，包括来自非共享试验的 AD
C
Wait until you get 100% IPD before publishing
两阶段混合解决方案
COMBINING IPD + AD
Stage 1: Analyze IPD trials → get θIPD
Stage 2: Meta-analyze θIPD + θAD 来自非 IPD 试验
Uses all available evidence, even when IPD incomplete
THE PRINCIPLE
IPD 不是全有或全无。
在您拥有的地方使用 IPD（用于交互测试）。
用 AD 进行补充以进行总体效果估计。
关于来源的透明度。
在您拥有的地方使用 IPD（用于交互测试）。
用 AD 进行补充以进行总体效果估计。
关于来源的透明度。
“意愿可能并不代表意愿
敞开的门可能会隐藏真相。
Always ask: Who refused to share?
And what might they be hiding?"
敞开的门可能会隐藏真相。
Always ask: Who refused to share?
And what might they be hiding?"
早期乳腺癌试验小组发现个体试验可以发现什么不是吗？
Tamoxifen works better than placebo
Tamoxifen only works in estrogen receptor positive tumors
Tamoxifen has significant side effects
Tamoxifen works better in younger women
What is the "ecological fallacy" in meta-analysis?
Assuming environmental factors don't affect treatment
Assuming between-study relationships apply to within-study (individual) relationships
来自不同生态区域的汇集研究
Using outdated trials in modern meta-analyses
什么时候一阶段 IPD 分析优于两阶段？
When trials have sparse events or rare outcomes
When you have many large trials
When you want to create forest plots
当结果常见且随访时间较短时
在分析多个 IPD 之前应该做什么试验？
立即汇集所有数据并运行主要分析
Exclude trials with missing variables
从 IPD 重现每个试验已发布的结果以验证数据完整性
Convert all continuous variables to categorical
References
本课程引用的主要来源
- Riley RD, et al. 个人参与者数据元分析：医疗保健研究手册。 Wiley, 2021.
- Stewart LA 等al. PRISMA-IPD：对个体参与者数据进行系统评价和荟萃分析的首选报告项目。 JAMA 2015;313:1657-65.
- Early Breast Cancer Trialists' Collaborative Group. Tamoxifen for early breast cancer. Cochrane Database Syst Rev 2001.
- Cholesterol Treatment Trialists' Collaboration. Efficacy and safety of LDL-lowering therapy. Lancet 2010;376:1670-81.
- Roberts D, et al. Antenatal corticosteroids for accelerating fetal lung maturation. Cochrane Database Syst Rev 2017.
- IMPACT Study Group. Predicting outcome after traumatic brain injury. PLoS Med 2008;5:e165.
- Debray TPA, et al. Get real in individual participant data meta-analysis. Int J Epidemiol 2015;44:1287-97.
- Burke DL, et al. Meta-analysis using individual participant data. Stat Med 2017;36:320-38.
✔
Course Complete
“总的来说，个体消失了。
But you have learned to find them.
You have learned to ask: Who benefits? Who is harmed?
现在开始——不要让任何患者消失。”
But you have learned to find them.
You have learned to ask: Who benefits? Who is harmed?
现在开始——不要让任何患者消失。”
隐藏的病人 — 现在你看到他们了。