SlideShare une entreprise Scribd logo
1  sur  18
Télécharger pour lire hors ligne
. . . . . .
.
......
A Framework of Causal Inference in Meta-Analysis
Presentation for G8325 Causal Inference
Presentation for G8325 Causal Inference A Framework of Causal Inference in Meta-Analysis
. . . . . .
.. Meta-Analysis
Meta-analysis, which collect and combine ndings from several
different but related studies, is becoming increasingly popular in
elds such as evidence-based medicine.
Presentation for G8325 Causal Inference A Framework of Causal Inference in Meta-Analysis
. . . . . .
.. Meta-Analysis
Meta-analysis, which collect and combine ndings from several
different but related studies, is becoming increasingly popular in
elds such as evidence-based medicine.
However, several pitfalls are well-known in the practice of
meta-analysis, such as publication bias, uneven qualities and
heterogeneous results of of the included studies.
We will deal with the heterogeneity of the estimates of effect size
of the studies.
Presentation for G8325 Causal Inference A Framework of Causal Inference in Meta-Analysis
. . . . . .
.. Visualizing Heterogeneity of Meta-Analysis
Figure: e typical approach of visualizing a meta-analysis: forest plot. e
last two rows give Fixed Effects and Random Effects estimates.
Presentation for G8325 Causal Inference A Framework of Causal Inference in Meta-Analysis
. . . . . .
.. Dealing with Heterogeneity: Fixed Effects model
e most natural way of dealing with heterogeneity is to model it
as sampling errors, mathematically
ˆθi = θ + εi
in which θ is the true underlying effect of interest.
en a re-weighted method based on the standard errors of the
study-level estimates can be used to estimate θ
ˆθ =
∑
wi
ˆθi
∑
wi
in which
wi = 1/Var(ˆθ)
Unfortunately, this method is dubbed Fixed Effects model, which
might cause some confusions.
Presentation for G8325 Causal Inference A Framework of Causal Inference in Meta-Analysis
. . . . . .
.. Dealing with Heterogeneity: Random Effects model
e predominant way of dealing with heterogeneity of
meta-analysis is through Random Effect model, made popular by
DeSimonian and Laird (1986).
Let ˆθi denote the effect size estimate from the i−th study,
Random Effect model assumes that it is the sum of the true effect
size for the i−th study, θi, and sampling error εi
ˆθi = θi + εi
And the true effect size of each study comes from some
distribution with mean θ and sd σ. θ is considered as the ultimate
quantity of interest.
θi
iid
∼ G(θ, σ2
)
e estimate is given by
ˆθ =
∑
wi
ˆθi
∑
wi
in which wi = 1/[Var(ˆθ) + ˆσ2]
Presentation for G8325 Causal Inference A Framework of Causal Inference in Meta-Analysis
. . . . . .
.. A Potential Outcomes Framework for Meta-Analysis
To base the meta-analysis on a solid causal foundation, a
Potential Outcome framework can be constructed.
Assume that there are S trials, indexed by 1, 2, 3, . . . , s, and 2
treatments with Z = 0 as the control arm and Z = 1 as the
treatment arm, and the outcome of interest is Y. en set of the
potential outcomes for the individual i will be
⃗Yi =
(
Yi(Z = 0, S = 1), . . . , Yi(Z = 0, S = s)
Yi(Z = 1, S = 1), . . . , Yi(Z = 1, S = s)
)
Presentation for G8325 Causal Inference A Framework of Causal Inference in Meta-Analysis
. . . . . .
.. Identi cation Assumptions
A1.Extended SUTVA: For all Z and S, ⃗Yi(Z, S) ≡ ⃗Yi(Zi, Si), in
which Zi and Si are the treatment and trial assignments of
individual i.
A2. Ignorability within Trials
⃗Yi ⊥⊥ Zi, ∀i ∈ {Si = s}
So within each trial, it is a randomized experiments.
With these two assumptions, apparently we can identify
treatment effect inside any trial.
Presentation for G8325 Causal Inference A Framework of Causal Inference in Meta-Analysis
. . . . . .
.. Further Modeling of the Response
..
We can write the potential outcome as
Yi(z, s) = αi + βi(s) + τi(z) + γi(z, s) + εi(z, s)
en the unit level treatment effect in study s is
Yi(1, s)−Yi(0, s) = τi(1)−τi(0)+γi(1, s)−γi(0, s)+εi(1, s)−εi(0, s)
Aggregating to the study level, the treatment effect in study s is
θs
△
= Es[Yi(1, s)−Yi(0, s)] = Es[τi(1)−τi(0)]+Es[γi(1, s)−γi(0, s)]
Presentation for G8325 Causal Inference A Framework of Causal Inference in Meta-Analysis
. . . . . .
.. Study-by-Treatment Interaction and Selection into Studies
From the last formula, we can see two factors contribute to the
heterogeneity of study level estimated effect sizes. We can make
further Identi cation Assumptions about them.
Presentation for G8325 Causal Inference A Framework of Causal Inference in Meta-Analysis
. . . . . .
.. Study-by-Treatment Interaction and Selection into Studies
From the last formula, we can see two factors contribute to the
heterogeneity of study level estimated effect sizes. We can make
further Identi cation Assumptions about them.
A3. No Study-by-Treatment Interaction
γi(z, s) = 0
A4. No Selection into Studies
⃗Yi ⊥⊥ Si, ∀i
If A1-A4 all hold, then study level treatment effect θs should not
depend on study, which means the heterogeneity solely results
from sampling errors. us we should use Fixed Effect model.
Presentation for G8325 Causal Inference A Framework of Causal Inference in Meta-Analysis
. . . . . .
..
Interpretation of the “Treatment Effect” Estimate from
Random Effects model
If A3 or A4 doesn’t hold, then Fixed Effects model is not valid.
But how much sense does the Random Effects model estimate
make?
Some justi cation of Random Effects model is based on
exchangeability, Higgins et al. (2009), i.e., we cannot tell a priori
which θi is larger or smaller. us the existence of a distribution
G from which, conditioned on some parameters, θi’s can be seen
as iid sample.
But the pure theoretical nature of this argument also threatens its
practical relevance. e mean of distribution G might not has any
causal interpretation.
Presentation for G8325 Causal Inference A Framework of Causal Inference in Meta-Analysis
. . . . . .
.. A Selection into Studies Example
..
In a hypothetical example, where the entire population can be
divided into four sub-populations by age (young/elderly) and
gender (male/female). e response is continuous, and the
treatment effects by sub-population are given
Male Female
Young 1 2
Elderly 4 6
Table: Treatment Effects by Sub-populations.
We conduct a meta-analysis based on 4 studies. However, the rst
two studies focused on the treatment effects on elderly, and the
second two studies focused on the treatment effects on female.
Sub-population
Study YM YF EM EF Average Effect
1 0% 0% 40% 60% 5.2
2 0% 0% 60% 40% 4.8
3 0% 40% 0% 60% 4.4
3 0% 60% 0% 40% 3.6
Table: Compositions of sub-population and average effects of the 4 studies
Presentation for G8325 Causal Inference A Framework of Causal Inference in Meta-Analysis
. . . . . .
.. Example Cont’d
..
Further, we can assume that the research on elderly is better
funded, and thus has larger sample size and smaller standard
errors. And the effect sizes of all trials are accurately estimated.
Study Estimate Standard Error
1 5.2 0.21
2 4.8 0.18
3 4.4 0.42
4 3.6 0.40
Table: Estimated Effect Sizes and Standard Errors from each Studies
e pooled treatment effect estimate based on Random Effects
model is 4.60, based on Fixed Effects model is 4.79, and based on
unweighted average is 4.5.
But these estimates don’t correspond to a treatment effect for any
well-de ned population. In fact, it is even less informative than
the estimates from individual trials.
Presentation for G8325 Causal Inference A Framework of Causal Inference in Meta-Analysis
. . . . . .
.. Individual-Level Data Explaining the Heterogeneity away
Vioxx Trials Data: Vioxx is a drug that was withdrawn from
market because of severe adverse impact on cardiovascular
system. We conduct a Individual-Level meta-analysis with data
from 30 randomized trials to examine the impact of Vioxx on risk
of cardiovascular adverse event.
For ease of computation, we apply a Poisson regression model to
the survival data, which is equivalent to a Cox Proportional
Hazard model with constant baseline hazard function.
Presentation for G8325 Causal Inference A Framework of Causal Inference in Meta-Analysis
. . . . . .
.. Individual-Level Data Explaining the Heterogeneity away
Fitting a random effect model on individual-level data but
without individual-level demographic covariates gives a large
estimate of random effect for treatment.
Random effects:
Groups Name Variance Std.Dev. Corr
trial (Intercept) 0.402776 0.634646
treatment1 0.009353 0.096711 -1.000
Number of obs: 17254, groups: trial, 30
Fixed effects:
Estimate Std. Error z value Pr(>|z|)
(Intercept) -10.1515 0.2075 -48.92 < 2e-16 ***
treatment1 0.4622 0.1246 3.71 0.000209 ***
Presentation for G8325 Causal Inference A Framework of Causal Inference in Meta-Analysis
. . . . . .
.. Individual-Level Data Explaining the Heterogeneity away
However, once the demographic covariates are included, the
estimate of random effect becomes essentially zero.
Random effects:
Groups Name Variance Std.Dev. Corr
trial (Intercept) 0.0000e+00 0.0000e+00
treatment1 1.0398e-10 1.0197e-05 NaN
Number of obs: 17254, groups: trial, 30
Fixed effects:
Estimate Std. Error z value Pr(>|z|)
(Intercept) -15.343832 0.484044 -31.70 < 2e-16 ***
raceblack 0.259301 0.390801 0.66 0.507004
racewhite 0.244987 0.238631 1.03 0.304593
age 0.071338 0.005902 12.09 < 2e-16 ***
genderM 0.415846 0.124012 3.35 0.000799 ***
treatment1 0.431808 0.118046 3.66 0.000254 ***
Presentation for G8325 Causal Inference A Framework of Causal Inference in Meta-Analysis
. . . . . .
.. Summary
Standard practice of meta-analysis lacks a solid causal
foundation, although researchers are mostly asking causal
questions. Establishing a potential framework for meta-analysis
will help us better understand the questions and hopefully guide
us in nding better answers.
Predominant methods such as Random Effects model, made
popular by DeSimonian and Laird (1986), typically fail to provide
us an estimate that admits causal interpretation.
Treating meta-analysis naïvely as number crunching is a bad
practice that should be discouraged. Instead, we should try to dig
into the original studies for possible explanations of
heterogeneity. We demonstrate an example where including
individual demographic covariates explains heterogeneity away.
Presentation for G8325 Causal Inference A Framework of Causal Inference in Meta-Analysis

Contenu connexe

Tendances

Repurposing Classification & Regression Trees for Causal Research with High-D...
Repurposing Classification & Regression Trees for Causal Research with High-D...Repurposing Classification & Regression Trees for Causal Research with High-D...
Repurposing Classification & Regression Trees for Causal Research with High-D...Galit Shmueli
 
To Explain, To Predict, or To Describe?
To Explain, To Predict, or To Describe?To Explain, To Predict, or To Describe?
To Explain, To Predict, or To Describe?Galit Shmueli
 
Chapter 4(2) Hypothesisi Testing
Chapter 4(2) Hypothesisi TestingChapter 4(2) Hypothesisi Testing
Chapter 4(2) Hypothesisi TestingSumit Prajapati
 
Hypothesis Testing: Proportions (Compare 1:Standard)
Hypothesis Testing: Proportions (Compare 1:Standard)Hypothesis Testing: Proportions (Compare 1:Standard)
Hypothesis Testing: Proportions (Compare 1:Standard)Matt Hansen
 
Hypothesis Testing: Central Tendency – Non-Normal (Compare 1:Standard)
Hypothesis Testing: Central Tendency – Non-Normal (Compare 1:Standard)Hypothesis Testing: Central Tendency – Non-Normal (Compare 1:Standard)
Hypothesis Testing: Central Tendency – Non-Normal (Compare 1:Standard)Matt Hansen
 
Hypothesis Testing: Spread (Compare 1:Standard)
Hypothesis Testing: Spread (Compare 1:Standard)Hypothesis Testing: Spread (Compare 1:Standard)
Hypothesis Testing: Spread (Compare 1:Standard)Matt Hansen
 
Joint causal inference on observational and experimental data - NIPS 2016 "Wh...
Joint causal inference on observational and experimental data - NIPS 2016 "Wh...Joint causal inference on observational and experimental data - NIPS 2016 "Wh...
Joint causal inference on observational and experimental data - NIPS 2016 "Wh...Sara Magliacane
 
Factor analysis using spss 2005
Factor analysis using spss 2005Factor analysis using spss 2005
Factor analysis using spss 2005jamescupello
 
Hypothesis Testing: Central Tendency – Normal (Compare 2+ Factors)
Hypothesis Testing: Central Tendency – Normal (Compare 2+ Factors)Hypothesis Testing: Central Tendency – Normal (Compare 2+ Factors)
Hypothesis Testing: Central Tendency – Normal (Compare 2+ Factors)Matt Hansen
 
Hypothesis Tests in R Programming
Hypothesis Tests in R ProgrammingHypothesis Tests in R Programming
Hypothesis Tests in R ProgrammingAtacan Garip
 
Hypothesis Testing: Proportions (Compare 1:1)
Hypothesis Testing: Proportions (Compare 1:1)Hypothesis Testing: Proportions (Compare 1:1)
Hypothesis Testing: Proportions (Compare 1:1)Matt Hansen
 
Hypothesis Testing: Spread (Compare 1:1)
Hypothesis Testing: Spread (Compare 1:1)Hypothesis Testing: Spread (Compare 1:1)
Hypothesis Testing: Spread (Compare 1:1)Matt Hansen
 
What's Significant? Hypothesis Testing, Effect Size, Confidence Intervals, & ...
What's Significant? Hypothesis Testing, Effect Size, Confidence Intervals, & ...What's Significant? Hypothesis Testing, Effect Size, Confidence Intervals, & ...
What's Significant? Hypothesis Testing, Effect Size, Confidence Intervals, & ...Pat Barlow
 

Tendances (20)

Repurposing Classification & Regression Trees for Causal Research with High-D...
Repurposing Classification & Regression Trees for Causal Research with High-D...Repurposing Classification & Regression Trees for Causal Research with High-D...
Repurposing Classification & Regression Trees for Causal Research with High-D...
 
To Explain, To Predict, or To Describe?
To Explain, To Predict, or To Describe?To Explain, To Predict, or To Describe?
To Explain, To Predict, or To Describe?
 
Chapter 4(2) Hypothesisi Testing
Chapter 4(2) Hypothesisi TestingChapter 4(2) Hypothesisi Testing
Chapter 4(2) Hypothesisi Testing
 
Two Means Independent Samples
Two Means Independent Samples  Two Means Independent Samples
Two Means Independent Samples
 
Hypothesis Testing: Proportions (Compare 1:Standard)
Hypothesis Testing: Proportions (Compare 1:Standard)Hypothesis Testing: Proportions (Compare 1:Standard)
Hypothesis Testing: Proportions (Compare 1:Standard)
 
Les5e ppt 08
Les5e ppt 08Les5e ppt 08
Les5e ppt 08
 
Hypothesis Testing: Central Tendency – Non-Normal (Compare 1:Standard)
Hypothesis Testing: Central Tendency – Non-Normal (Compare 1:Standard)Hypothesis Testing: Central Tendency – Non-Normal (Compare 1:Standard)
Hypothesis Testing: Central Tendency – Non-Normal (Compare 1:Standard)
 
Spss session 1 and 2
Spss session 1 and 2Spss session 1 and 2
Spss session 1 and 2
 
Hypothesis Testing: Spread (Compare 1:Standard)
Hypothesis Testing: Spread (Compare 1:Standard)Hypothesis Testing: Spread (Compare 1:Standard)
Hypothesis Testing: Spread (Compare 1:Standard)
 
Joint causal inference on observational and experimental data - NIPS 2016 "Wh...
Joint causal inference on observational and experimental data - NIPS 2016 "Wh...Joint causal inference on observational and experimental data - NIPS 2016 "Wh...
Joint causal inference on observational and experimental data - NIPS 2016 "Wh...
 
Factor analysis using spss 2005
Factor analysis using spss 2005Factor analysis using spss 2005
Factor analysis using spss 2005
 
Hypothesis Testing: Central Tendency – Normal (Compare 2+ Factors)
Hypothesis Testing: Central Tendency – Normal (Compare 2+ Factors)Hypothesis Testing: Central Tendency – Normal (Compare 2+ Factors)
Hypothesis Testing: Central Tendency – Normal (Compare 2+ Factors)
 
Hypothesis Tests in R Programming
Hypothesis Tests in R ProgrammingHypothesis Tests in R Programming
Hypothesis Tests in R Programming
 
Les5e ppt 09
Les5e ppt 09Les5e ppt 09
Les5e ppt 09
 
Hypothesis Testing: Proportions (Compare 1:1)
Hypothesis Testing: Proportions (Compare 1:1)Hypothesis Testing: Proportions (Compare 1:1)
Hypothesis Testing: Proportions (Compare 1:1)
 
Les5e ppt 10
Les5e ppt 10Les5e ppt 10
Les5e ppt 10
 
Hypothesis Testing: Spread (Compare 1:1)
Hypothesis Testing: Spread (Compare 1:1)Hypothesis Testing: Spread (Compare 1:1)
Hypothesis Testing: Spread (Compare 1:1)
 
Testing of hypothesis
Testing of hypothesisTesting of hypothesis
Testing of hypothesis
 
Les5e ppt 11
Les5e ppt 11Les5e ppt 11
Les5e ppt 11
 
What's Significant? Hypothesis Testing, Effect Size, Confidence Intervals, & ...
What's Significant? Hypothesis Testing, Effect Size, Confidence Intervals, & ...What's Significant? Hypothesis Testing, Effect Size, Confidence Intervals, & ...
What's Significant? Hypothesis Testing, Effect Size, Confidence Intervals, & ...
 

Similaire à A Causal Framework for Meta-Analysis, drafty

Statistice Chapter 02[1]
Statistice  Chapter 02[1]Statistice  Chapter 02[1]
Statistice Chapter 02[1]plisasm
 
Histograms and Descriptive Statistics Scoring GuideCRITERIANON.docx
Histograms and Descriptive Statistics Scoring GuideCRITERIANON.docxHistograms and Descriptive Statistics Scoring GuideCRITERIANON.docx
Histograms and Descriptive Statistics Scoring GuideCRITERIANON.docxpooleavelina
 
Regression shrinkage: better answers to causal questions
Regression shrinkage: better answers to causal questionsRegression shrinkage: better answers to causal questions
Regression shrinkage: better answers to causal questionsMaarten van Smeden
 
Probability distribution Function & Decision Trees in machine learning
Probability distribution Function  & Decision Trees in machine learningProbability distribution Function  & Decision Trees in machine learning
Probability distribution Function & Decision Trees in machine learningSadia Zafar
 
Inference via Bayesian Synthetic Likelihoods for a Mixed-Effects SDE Model of...
Inference via Bayesian Synthetic Likelihoods for a Mixed-Effects SDE Model of...Inference via Bayesian Synthetic Likelihoods for a Mixed-Effects SDE Model of...
Inference via Bayesian Synthetic Likelihoods for a Mixed-Effects SDE Model of...Umberto Picchini
 
Novelties in social science statistics
Novelties in social science statisticsNovelties in social science statistics
Novelties in social science statisticsJiri Haviger
 
Model validation strategies ftc 2018
Model validation strategies ftc 2018Model validation strategies ftc 2018
Model validation strategies ftc 2018Philip Ramsey
 
Comparison between two statistical tests of significance
Comparison between two statistical tests of significanceComparison between two statistical tests of significance
Comparison between two statistical tests of significanceinventionjournals
 
Lecture 7 guidelines_and_assignment
Lecture 7 guidelines_and_assignmentLecture 7 guidelines_and_assignment
Lecture 7 guidelines_and_assignmentDaria Bogdanova
 
Advanced business mathematics and statistics for entrepreneurs
Advanced business mathematics and statistics for entrepreneursAdvanced business mathematics and statistics for entrepreneurs
Advanced business mathematics and statistics for entrepreneursDr. Trilok Kumar Jain
 
Sample Size Determination.23.11.2021.pdf
Sample Size Determination.23.11.2021.pdfSample Size Determination.23.11.2021.pdf
Sample Size Determination.23.11.2021.pdfstatsanjal
 
BioSHaRE: Analysis of mixed effects models using federated data analysis appr...
BioSHaRE: Analysis of mixed effects models using federated data analysis appr...BioSHaRE: Analysis of mixed effects models using federated data analysis appr...
BioSHaRE: Analysis of mixed effects models using federated data analysis appr...Lisette Giepmans
 
To Explain Or To Predict?
To Explain Or To Predict?To Explain Or To Predict?
To Explain Or To Predict?Galit Shmueli
 

Similaire à A Causal Framework for Meta-Analysis, drafty (20)

Statistice Chapter 02[1]
Statistice  Chapter 02[1]Statistice  Chapter 02[1]
Statistice Chapter 02[1]
 
Histograms and Descriptive Statistics Scoring GuideCRITERIANON.docx
Histograms and Descriptive Statistics Scoring GuideCRITERIANON.docxHistograms and Descriptive Statistics Scoring GuideCRITERIANON.docx
Histograms and Descriptive Statistics Scoring GuideCRITERIANON.docx
 
Causal Inference Opening Workshop - Latent Variable Models, Causal Inference,...
Causal Inference Opening Workshop - Latent Variable Models, Causal Inference,...Causal Inference Opening Workshop - Latent Variable Models, Causal Inference,...
Causal Inference Opening Workshop - Latent Variable Models, Causal Inference,...
 
Regression shrinkage: better answers to causal questions
Regression shrinkage: better answers to causal questionsRegression shrinkage: better answers to causal questions
Regression shrinkage: better answers to causal questions
 
Probability distribution Function & Decision Trees in machine learning
Probability distribution Function  & Decision Trees in machine learningProbability distribution Function  & Decision Trees in machine learning
Probability distribution Function & Decision Trees in machine learning
 
isi
isiisi
isi
 
Inference via Bayesian Synthetic Likelihoods for a Mixed-Effects SDE Model of...
Inference via Bayesian Synthetic Likelihoods for a Mixed-Effects SDE Model of...Inference via Bayesian Synthetic Likelihoods for a Mixed-Effects SDE Model of...
Inference via Bayesian Synthetic Likelihoods for a Mixed-Effects SDE Model of...
 
Novelties in social science statistics
Novelties in social science statisticsNovelties in social science statistics
Novelties in social science statistics
 
Prueba z en ingles
Prueba z en ingles Prueba z en ingles
Prueba z en ingles
 
Model validation strategies ftc 2018
Model validation strategies ftc 2018Model validation strategies ftc 2018
Model validation strategies ftc 2018
 
Hypothesis testing
Hypothesis testingHypothesis testing
Hypothesis testing
 
Comparison between two statistical tests of significance
Comparison between two statistical tests of significanceComparison between two statistical tests of significance
Comparison between two statistical tests of significance
 
Causal Inference Opening Workshop - Bayesian Nonparametric Models for Treatme...
Causal Inference Opening Workshop - Bayesian Nonparametric Models for Treatme...Causal Inference Opening Workshop - Bayesian Nonparametric Models for Treatme...
Causal Inference Opening Workshop - Bayesian Nonparametric Models for Treatme...
 
Lecture 7 guidelines_and_assignment
Lecture 7 guidelines_and_assignmentLecture 7 guidelines_and_assignment
Lecture 7 guidelines_and_assignment
 
Advanced business mathematics and statistics for entrepreneurs
Advanced business mathematics and statistics for entrepreneursAdvanced business mathematics and statistics for entrepreneurs
Advanced business mathematics and statistics for entrepreneurs
 
Sample Size Determination.23.11.2021.pdf
Sample Size Determination.23.11.2021.pdfSample Size Determination.23.11.2021.pdf
Sample Size Determination.23.11.2021.pdf
 
Math 300 MM Project
Math 300 MM ProjectMath 300 MM Project
Math 300 MM Project
 
BioSHaRE: Analysis of mixed effects models using federated data analysis appr...
BioSHaRE: Analysis of mixed effects models using federated data analysis appr...BioSHaRE: Analysis of mixed effects models using federated data analysis appr...
BioSHaRE: Analysis of mixed effects models using federated data analysis appr...
 
ecir2019tutorial
ecir2019tutorialecir2019tutorial
ecir2019tutorial
 
To Explain Or To Predict?
To Explain Or To Predict?To Explain Or To Predict?
To Explain Or To Predict?
 

Dernier

How to Get Started in Social Media for Art League City
How to Get Started in Social Media for Art League CityHow to Get Started in Social Media for Art League City
How to Get Started in Social Media for Art League CityEric T. Tung
 
Call Girls Service In Old Town Dubai ((0551707352)) Old Town Dubai Call Girl ...
Call Girls Service In Old Town Dubai ((0551707352)) Old Town Dubai Call Girl ...Call Girls Service In Old Town Dubai ((0551707352)) Old Town Dubai Call Girl ...
Call Girls Service In Old Town Dubai ((0551707352)) Old Town Dubai Call Girl ...allensay1
 
Call Girls Ludhiana Just Call 98765-12871 Top Class Call Girl Service Available
Call Girls Ludhiana Just Call 98765-12871 Top Class Call Girl Service AvailableCall Girls Ludhiana Just Call 98765-12871 Top Class Call Girl Service Available
Call Girls Ludhiana Just Call 98765-12871 Top Class Call Girl Service AvailableSeo
 
Call Girls Zirakpur👧 Book Now📱7837612180 📞👉Call Girl Service In Zirakpur No A...
Call Girls Zirakpur👧 Book Now📱7837612180 📞👉Call Girl Service In Zirakpur No A...Call Girls Zirakpur👧 Book Now📱7837612180 📞👉Call Girl Service In Zirakpur No A...
Call Girls Zirakpur👧 Book Now📱7837612180 📞👉Call Girl Service In Zirakpur No A...Sheetaleventcompany
 
Eluru Call Girls Service ☎ ️93326-06886 ❤️‍🔥 Enjoy 24/7 Escort Service
Eluru Call Girls Service ☎ ️93326-06886 ❤️‍🔥 Enjoy 24/7 Escort ServiceEluru Call Girls Service ☎ ️93326-06886 ❤️‍🔥 Enjoy 24/7 Escort Service
Eluru Call Girls Service ☎ ️93326-06886 ❤️‍🔥 Enjoy 24/7 Escort ServiceDamini Dixit
 
Chandigarh Escorts Service 📞8868886958📞 Just📲 Call Nihal Chandigarh Call Girl...
Chandigarh Escorts Service 📞8868886958📞 Just📲 Call Nihal Chandigarh Call Girl...Chandigarh Escorts Service 📞8868886958📞 Just📲 Call Nihal Chandigarh Call Girl...
Chandigarh Escorts Service 📞8868886958📞 Just📲 Call Nihal Chandigarh Call Girl...Sheetaleventcompany
 
Call Girls Electronic City Just Call 👗 7737669865 👗 Top Class Call Girl Servi...
Call Girls Electronic City Just Call 👗 7737669865 👗 Top Class Call Girl Servi...Call Girls Electronic City Just Call 👗 7737669865 👗 Top Class Call Girl Servi...
Call Girls Electronic City Just Call 👗 7737669865 👗 Top Class Call Girl Servi...amitlee9823
 
Falcon Invoice Discounting: The best investment platform in india for investors
Falcon Invoice Discounting: The best investment platform in india for investorsFalcon Invoice Discounting: The best investment platform in india for investors
Falcon Invoice Discounting: The best investment platform in india for investorsFalcon Invoice Discounting
 
Malegaon Call Girls Service ☎ ️82500–77686 ☎️ Enjoy 24/7 Escort Service
Malegaon Call Girls Service ☎ ️82500–77686 ☎️ Enjoy 24/7 Escort ServiceMalegaon Call Girls Service ☎ ️82500–77686 ☎️ Enjoy 24/7 Escort Service
Malegaon Call Girls Service ☎ ️82500–77686 ☎️ Enjoy 24/7 Escort ServiceDamini Dixit
 
Uneak White's Personal Brand Exploration Presentation
Uneak White's Personal Brand Exploration PresentationUneak White's Personal Brand Exploration Presentation
Uneak White's Personal Brand Exploration Presentationuneakwhite
 
Marel Q1 2024 Investor Presentation from May 8, 2024
Marel Q1 2024 Investor Presentation from May 8, 2024Marel Q1 2024 Investor Presentation from May 8, 2024
Marel Q1 2024 Investor Presentation from May 8, 2024Marel
 
The Path to Product Excellence: Avoiding Common Pitfalls and Enhancing Commun...
The Path to Product Excellence: Avoiding Common Pitfalls and Enhancing Commun...The Path to Product Excellence: Avoiding Common Pitfalls and Enhancing Commun...
The Path to Product Excellence: Avoiding Common Pitfalls and Enhancing Commun...Aggregage
 
Falcon's Invoice Discounting: Your Path to Prosperity
Falcon's Invoice Discounting: Your Path to ProsperityFalcon's Invoice Discounting: Your Path to Prosperity
Falcon's Invoice Discounting: Your Path to Prosperityhemanthkumar470700
 
BAGALUR CALL GIRL IN 98274*61493 ❤CALL GIRLS IN ESCORT SERVICE❤CALL GIRL
BAGALUR CALL GIRL IN 98274*61493 ❤CALL GIRLS IN ESCORT SERVICE❤CALL GIRLBAGALUR CALL GIRL IN 98274*61493 ❤CALL GIRLS IN ESCORT SERVICE❤CALL GIRL
BAGALUR CALL GIRL IN 98274*61493 ❤CALL GIRLS IN ESCORT SERVICE❤CALL GIRLkapoorjyoti4444
 
Call Girls in Delhi, Escort Service Available 24x7 in Delhi 959961-/-3876
Call Girls in Delhi, Escort Service Available 24x7 in Delhi 959961-/-3876Call Girls in Delhi, Escort Service Available 24x7 in Delhi 959961-/-3876
Call Girls in Delhi, Escort Service Available 24x7 in Delhi 959961-/-3876dlhescort
 
Call Girls Hebbal Just Call 👗 7737669865 👗 Top Class Call Girl Service Bangalore
Call Girls Hebbal Just Call 👗 7737669865 👗 Top Class Call Girl Service BangaloreCall Girls Hebbal Just Call 👗 7737669865 👗 Top Class Call Girl Service Bangalore
Call Girls Hebbal Just Call 👗 7737669865 👗 Top Class Call Girl Service Bangaloreamitlee9823
 
FULL ENJOY Call Girls In Mahipalpur Delhi Contact Us 8377877756
FULL ENJOY Call Girls In Mahipalpur Delhi Contact Us 8377877756FULL ENJOY Call Girls In Mahipalpur Delhi Contact Us 8377877756
FULL ENJOY Call Girls In Mahipalpur Delhi Contact Us 8377877756dollysharma2066
 
Call Girls Jp Nagar Just Call 👗 7737669865 👗 Top Class Call Girl Service Bang...
Call Girls Jp Nagar Just Call 👗 7737669865 👗 Top Class Call Girl Service Bang...Call Girls Jp Nagar Just Call 👗 7737669865 👗 Top Class Call Girl Service Bang...
Call Girls Jp Nagar Just Call 👗 7737669865 👗 Top Class Call Girl Service Bang...amitlee9823
 

Dernier (20)

How to Get Started in Social Media for Art League City
How to Get Started in Social Media for Art League CityHow to Get Started in Social Media for Art League City
How to Get Started in Social Media for Art League City
 
Call Girls Service In Old Town Dubai ((0551707352)) Old Town Dubai Call Girl ...
Call Girls Service In Old Town Dubai ((0551707352)) Old Town Dubai Call Girl ...Call Girls Service In Old Town Dubai ((0551707352)) Old Town Dubai Call Girl ...
Call Girls Service In Old Town Dubai ((0551707352)) Old Town Dubai Call Girl ...
 
Call Girls Ludhiana Just Call 98765-12871 Top Class Call Girl Service Available
Call Girls Ludhiana Just Call 98765-12871 Top Class Call Girl Service AvailableCall Girls Ludhiana Just Call 98765-12871 Top Class Call Girl Service Available
Call Girls Ludhiana Just Call 98765-12871 Top Class Call Girl Service Available
 
Call Girls Zirakpur👧 Book Now📱7837612180 📞👉Call Girl Service In Zirakpur No A...
Call Girls Zirakpur👧 Book Now📱7837612180 📞👉Call Girl Service In Zirakpur No A...Call Girls Zirakpur👧 Book Now📱7837612180 📞👉Call Girl Service In Zirakpur No A...
Call Girls Zirakpur👧 Book Now📱7837612180 📞👉Call Girl Service In Zirakpur No A...
 
Eluru Call Girls Service ☎ ️93326-06886 ❤️‍🔥 Enjoy 24/7 Escort Service
Eluru Call Girls Service ☎ ️93326-06886 ❤️‍🔥 Enjoy 24/7 Escort ServiceEluru Call Girls Service ☎ ️93326-06886 ❤️‍🔥 Enjoy 24/7 Escort Service
Eluru Call Girls Service ☎ ️93326-06886 ❤️‍🔥 Enjoy 24/7 Escort Service
 
Chandigarh Escorts Service 📞8868886958📞 Just📲 Call Nihal Chandigarh Call Girl...
Chandigarh Escorts Service 📞8868886958📞 Just📲 Call Nihal Chandigarh Call Girl...Chandigarh Escorts Service 📞8868886958📞 Just📲 Call Nihal Chandigarh Call Girl...
Chandigarh Escorts Service 📞8868886958📞 Just📲 Call Nihal Chandigarh Call Girl...
 
Call Girls Electronic City Just Call 👗 7737669865 👗 Top Class Call Girl Servi...
Call Girls Electronic City Just Call 👗 7737669865 👗 Top Class Call Girl Servi...Call Girls Electronic City Just Call 👗 7737669865 👗 Top Class Call Girl Servi...
Call Girls Electronic City Just Call 👗 7737669865 👗 Top Class Call Girl Servi...
 
Falcon Invoice Discounting: The best investment platform in india for investors
Falcon Invoice Discounting: The best investment platform in india for investorsFalcon Invoice Discounting: The best investment platform in india for investors
Falcon Invoice Discounting: The best investment platform in india for investors
 
Malegaon Call Girls Service ☎ ️82500–77686 ☎️ Enjoy 24/7 Escort Service
Malegaon Call Girls Service ☎ ️82500–77686 ☎️ Enjoy 24/7 Escort ServiceMalegaon Call Girls Service ☎ ️82500–77686 ☎️ Enjoy 24/7 Escort Service
Malegaon Call Girls Service ☎ ️82500–77686 ☎️ Enjoy 24/7 Escort Service
 
Uneak White's Personal Brand Exploration Presentation
Uneak White's Personal Brand Exploration PresentationUneak White's Personal Brand Exploration Presentation
Uneak White's Personal Brand Exploration Presentation
 
Marel Q1 2024 Investor Presentation from May 8, 2024
Marel Q1 2024 Investor Presentation from May 8, 2024Marel Q1 2024 Investor Presentation from May 8, 2024
Marel Q1 2024 Investor Presentation from May 8, 2024
 
unwanted pregnancy Kit [+918133066128] Abortion Pills IN Dubai UAE Abudhabi
unwanted pregnancy Kit [+918133066128] Abortion Pills IN Dubai UAE Abudhabiunwanted pregnancy Kit [+918133066128] Abortion Pills IN Dubai UAE Abudhabi
unwanted pregnancy Kit [+918133066128] Abortion Pills IN Dubai UAE Abudhabi
 
The Path to Product Excellence: Avoiding Common Pitfalls and Enhancing Commun...
The Path to Product Excellence: Avoiding Common Pitfalls and Enhancing Commun...The Path to Product Excellence: Avoiding Common Pitfalls and Enhancing Commun...
The Path to Product Excellence: Avoiding Common Pitfalls and Enhancing Commun...
 
(Anamika) VIP Call Girls Napur Call Now 8617697112 Napur Escorts 24x7
(Anamika) VIP Call Girls Napur Call Now 8617697112 Napur Escorts 24x7(Anamika) VIP Call Girls Napur Call Now 8617697112 Napur Escorts 24x7
(Anamika) VIP Call Girls Napur Call Now 8617697112 Napur Escorts 24x7
 
Falcon's Invoice Discounting: Your Path to Prosperity
Falcon's Invoice Discounting: Your Path to ProsperityFalcon's Invoice Discounting: Your Path to Prosperity
Falcon's Invoice Discounting: Your Path to Prosperity
 
BAGALUR CALL GIRL IN 98274*61493 ❤CALL GIRLS IN ESCORT SERVICE❤CALL GIRL
BAGALUR CALL GIRL IN 98274*61493 ❤CALL GIRLS IN ESCORT SERVICE❤CALL GIRLBAGALUR CALL GIRL IN 98274*61493 ❤CALL GIRLS IN ESCORT SERVICE❤CALL GIRL
BAGALUR CALL GIRL IN 98274*61493 ❤CALL GIRLS IN ESCORT SERVICE❤CALL GIRL
 
Call Girls in Delhi, Escort Service Available 24x7 in Delhi 959961-/-3876
Call Girls in Delhi, Escort Service Available 24x7 in Delhi 959961-/-3876Call Girls in Delhi, Escort Service Available 24x7 in Delhi 959961-/-3876
Call Girls in Delhi, Escort Service Available 24x7 in Delhi 959961-/-3876
 
Call Girls Hebbal Just Call 👗 7737669865 👗 Top Class Call Girl Service Bangalore
Call Girls Hebbal Just Call 👗 7737669865 👗 Top Class Call Girl Service BangaloreCall Girls Hebbal Just Call 👗 7737669865 👗 Top Class Call Girl Service Bangalore
Call Girls Hebbal Just Call 👗 7737669865 👗 Top Class Call Girl Service Bangalore
 
FULL ENJOY Call Girls In Mahipalpur Delhi Contact Us 8377877756
FULL ENJOY Call Girls In Mahipalpur Delhi Contact Us 8377877756FULL ENJOY Call Girls In Mahipalpur Delhi Contact Us 8377877756
FULL ENJOY Call Girls In Mahipalpur Delhi Contact Us 8377877756
 
Call Girls Jp Nagar Just Call 👗 7737669865 👗 Top Class Call Girl Service Bang...
Call Girls Jp Nagar Just Call 👗 7737669865 👗 Top Class Call Girl Service Bang...Call Girls Jp Nagar Just Call 👗 7737669865 👗 Top Class Call Girl Service Bang...
Call Girls Jp Nagar Just Call 👗 7737669865 👗 Top Class Call Girl Service Bang...
 

A Causal Framework for Meta-Analysis, drafty

  • 1. . . . . . . . ...... A Framework of Causal Inference in Meta-Analysis Presentation for G8325 Causal Inference Presentation for G8325 Causal Inference A Framework of Causal Inference in Meta-Analysis
  • 2. . . . . . . .. Meta-Analysis Meta-analysis, which collect and combine ndings from several different but related studies, is becoming increasingly popular in elds such as evidence-based medicine. Presentation for G8325 Causal Inference A Framework of Causal Inference in Meta-Analysis
  • 3. . . . . . . .. Meta-Analysis Meta-analysis, which collect and combine ndings from several different but related studies, is becoming increasingly popular in elds such as evidence-based medicine. However, several pitfalls are well-known in the practice of meta-analysis, such as publication bias, uneven qualities and heterogeneous results of of the included studies. We will deal with the heterogeneity of the estimates of effect size of the studies. Presentation for G8325 Causal Inference A Framework of Causal Inference in Meta-Analysis
  • 4. . . . . . . .. Visualizing Heterogeneity of Meta-Analysis Figure: e typical approach of visualizing a meta-analysis: forest plot. e last two rows give Fixed Effects and Random Effects estimates. Presentation for G8325 Causal Inference A Framework of Causal Inference in Meta-Analysis
  • 5. . . . . . . .. Dealing with Heterogeneity: Fixed Effects model e most natural way of dealing with heterogeneity is to model it as sampling errors, mathematically ˆθi = θ + εi in which θ is the true underlying effect of interest. en a re-weighted method based on the standard errors of the study-level estimates can be used to estimate θ ˆθ = ∑ wi ˆθi ∑ wi in which wi = 1/Var(ˆθ) Unfortunately, this method is dubbed Fixed Effects model, which might cause some confusions. Presentation for G8325 Causal Inference A Framework of Causal Inference in Meta-Analysis
  • 6. . . . . . . .. Dealing with Heterogeneity: Random Effects model e predominant way of dealing with heterogeneity of meta-analysis is through Random Effect model, made popular by DeSimonian and Laird (1986). Let ˆθi denote the effect size estimate from the i−th study, Random Effect model assumes that it is the sum of the true effect size for the i−th study, θi, and sampling error εi ˆθi = θi + εi And the true effect size of each study comes from some distribution with mean θ and sd σ. θ is considered as the ultimate quantity of interest. θi iid ∼ G(θ, σ2 ) e estimate is given by ˆθ = ∑ wi ˆθi ∑ wi in which wi = 1/[Var(ˆθ) + ˆσ2] Presentation for G8325 Causal Inference A Framework of Causal Inference in Meta-Analysis
  • 7. . . . . . . .. A Potential Outcomes Framework for Meta-Analysis To base the meta-analysis on a solid causal foundation, a Potential Outcome framework can be constructed. Assume that there are S trials, indexed by 1, 2, 3, . . . , s, and 2 treatments with Z = 0 as the control arm and Z = 1 as the treatment arm, and the outcome of interest is Y. en set of the potential outcomes for the individual i will be ⃗Yi = ( Yi(Z = 0, S = 1), . . . , Yi(Z = 0, S = s) Yi(Z = 1, S = 1), . . . , Yi(Z = 1, S = s) ) Presentation for G8325 Causal Inference A Framework of Causal Inference in Meta-Analysis
  • 8. . . . . . . .. Identi cation Assumptions A1.Extended SUTVA: For all Z and S, ⃗Yi(Z, S) ≡ ⃗Yi(Zi, Si), in which Zi and Si are the treatment and trial assignments of individual i. A2. Ignorability within Trials ⃗Yi ⊥⊥ Zi, ∀i ∈ {Si = s} So within each trial, it is a randomized experiments. With these two assumptions, apparently we can identify treatment effect inside any trial. Presentation for G8325 Causal Inference A Framework of Causal Inference in Meta-Analysis
  • 9. . . . . . . .. Further Modeling of the Response .. We can write the potential outcome as Yi(z, s) = αi + βi(s) + τi(z) + γi(z, s) + εi(z, s) en the unit level treatment effect in study s is Yi(1, s)−Yi(0, s) = τi(1)−τi(0)+γi(1, s)−γi(0, s)+εi(1, s)−εi(0, s) Aggregating to the study level, the treatment effect in study s is θs △ = Es[Yi(1, s)−Yi(0, s)] = Es[τi(1)−τi(0)]+Es[γi(1, s)−γi(0, s)] Presentation for G8325 Causal Inference A Framework of Causal Inference in Meta-Analysis
  • 10. . . . . . . .. Study-by-Treatment Interaction and Selection into Studies From the last formula, we can see two factors contribute to the heterogeneity of study level estimated effect sizes. We can make further Identi cation Assumptions about them. Presentation for G8325 Causal Inference A Framework of Causal Inference in Meta-Analysis
  • 11. . . . . . . .. Study-by-Treatment Interaction and Selection into Studies From the last formula, we can see two factors contribute to the heterogeneity of study level estimated effect sizes. We can make further Identi cation Assumptions about them. A3. No Study-by-Treatment Interaction γi(z, s) = 0 A4. No Selection into Studies ⃗Yi ⊥⊥ Si, ∀i If A1-A4 all hold, then study level treatment effect θs should not depend on study, which means the heterogeneity solely results from sampling errors. us we should use Fixed Effect model. Presentation for G8325 Causal Inference A Framework of Causal Inference in Meta-Analysis
  • 12. . . . . . . .. Interpretation of the “Treatment Effect” Estimate from Random Effects model If A3 or A4 doesn’t hold, then Fixed Effects model is not valid. But how much sense does the Random Effects model estimate make? Some justi cation of Random Effects model is based on exchangeability, Higgins et al. (2009), i.e., we cannot tell a priori which θi is larger or smaller. us the existence of a distribution G from which, conditioned on some parameters, θi’s can be seen as iid sample. But the pure theoretical nature of this argument also threatens its practical relevance. e mean of distribution G might not has any causal interpretation. Presentation for G8325 Causal Inference A Framework of Causal Inference in Meta-Analysis
  • 13. . . . . . . .. A Selection into Studies Example .. In a hypothetical example, where the entire population can be divided into four sub-populations by age (young/elderly) and gender (male/female). e response is continuous, and the treatment effects by sub-population are given Male Female Young 1 2 Elderly 4 6 Table: Treatment Effects by Sub-populations. We conduct a meta-analysis based on 4 studies. However, the rst two studies focused on the treatment effects on elderly, and the second two studies focused on the treatment effects on female. Sub-population Study YM YF EM EF Average Effect 1 0% 0% 40% 60% 5.2 2 0% 0% 60% 40% 4.8 3 0% 40% 0% 60% 4.4 3 0% 60% 0% 40% 3.6 Table: Compositions of sub-population and average effects of the 4 studies Presentation for G8325 Causal Inference A Framework of Causal Inference in Meta-Analysis
  • 14. . . . . . . .. Example Cont’d .. Further, we can assume that the research on elderly is better funded, and thus has larger sample size and smaller standard errors. And the effect sizes of all trials are accurately estimated. Study Estimate Standard Error 1 5.2 0.21 2 4.8 0.18 3 4.4 0.42 4 3.6 0.40 Table: Estimated Effect Sizes and Standard Errors from each Studies e pooled treatment effect estimate based on Random Effects model is 4.60, based on Fixed Effects model is 4.79, and based on unweighted average is 4.5. But these estimates don’t correspond to a treatment effect for any well-de ned population. In fact, it is even less informative than the estimates from individual trials. Presentation for G8325 Causal Inference A Framework of Causal Inference in Meta-Analysis
  • 15. . . . . . . .. Individual-Level Data Explaining the Heterogeneity away Vioxx Trials Data: Vioxx is a drug that was withdrawn from market because of severe adverse impact on cardiovascular system. We conduct a Individual-Level meta-analysis with data from 30 randomized trials to examine the impact of Vioxx on risk of cardiovascular adverse event. For ease of computation, we apply a Poisson regression model to the survival data, which is equivalent to a Cox Proportional Hazard model with constant baseline hazard function. Presentation for G8325 Causal Inference A Framework of Causal Inference in Meta-Analysis
  • 16. . . . . . . .. Individual-Level Data Explaining the Heterogeneity away Fitting a random effect model on individual-level data but without individual-level demographic covariates gives a large estimate of random effect for treatment. Random effects: Groups Name Variance Std.Dev. Corr trial (Intercept) 0.402776 0.634646 treatment1 0.009353 0.096711 -1.000 Number of obs: 17254, groups: trial, 30 Fixed effects: Estimate Std. Error z value Pr(>|z|) (Intercept) -10.1515 0.2075 -48.92 < 2e-16 *** treatment1 0.4622 0.1246 3.71 0.000209 *** Presentation for G8325 Causal Inference A Framework of Causal Inference in Meta-Analysis
  • 17. . . . . . . .. Individual-Level Data Explaining the Heterogeneity away However, once the demographic covariates are included, the estimate of random effect becomes essentially zero. Random effects: Groups Name Variance Std.Dev. Corr trial (Intercept) 0.0000e+00 0.0000e+00 treatment1 1.0398e-10 1.0197e-05 NaN Number of obs: 17254, groups: trial, 30 Fixed effects: Estimate Std. Error z value Pr(>|z|) (Intercept) -15.343832 0.484044 -31.70 < 2e-16 *** raceblack 0.259301 0.390801 0.66 0.507004 racewhite 0.244987 0.238631 1.03 0.304593 age 0.071338 0.005902 12.09 < 2e-16 *** genderM 0.415846 0.124012 3.35 0.000799 *** treatment1 0.431808 0.118046 3.66 0.000254 *** Presentation for G8325 Causal Inference A Framework of Causal Inference in Meta-Analysis
  • 18. . . . . . . .. Summary Standard practice of meta-analysis lacks a solid causal foundation, although researchers are mostly asking causal questions. Establishing a potential framework for meta-analysis will help us better understand the questions and hopefully guide us in nding better answers. Predominant methods such as Random Effects model, made popular by DeSimonian and Laird (1986), typically fail to provide us an estimate that admits causal interpretation. Treating meta-analysis naïvely as number crunching is a bad practice that should be discouraged. Instead, we should try to dig into the original studies for possible explanations of heterogeneity. We demonstrate an example where including individual demographic covariates explains heterogeneity away. Presentation for G8325 Causal Inference A Framework of Causal Inference in Meta-Analysis