Exciting Slides from Presentation
Target trial emulation
Causal inference from observational data
Joy Shi, PhD
Instructor of Epidemiology
CAUSAL and Department of Epidemiology
Harvard T.H. Chan School of Public Health
ISPOR
May 9, 2023
Disclosures
2
This research was supported by the U.S. Department of Veterans
Affairs (VA) Office of Research and Development (ORD)
Cooperative Studies Program (CSP) Epidemiology Center at the
VA Boston Healthcare System through CSP #2032, by resources
and the use of facilities at the VA Boston Healthcare System and
VA Informatics and Computing Infrastructure (VINCI) (VA HSR
RES 13-457).
The VA CAUSAL Methods Core is a collaboration between the
Massachusetts Veterans Epidemiology, Research, and
Information Center (MAVERIC) Division of Population Health
and Data Sciences and the CAUSALab at the Harvard T.H. Chan
School of Public Health.
For clinical practice and public health,
decisions are necessary
E.g., treatment decisions:
•
Treat with A or B?
•
Treat now or later?
•
When to stop treatment?
E.g., implementing a screening
program:
•
Will this screening program work?
•
How often do we screen?
•
Who do we screen?
3
For clinical practice and public health,
decisions are necessary
E.g., treatment decisions:
•
Treat with A or B?
•
Treat now or later?
•
When to stop treatment?
E.g., implementing a screening
program:
•
Will this screening program work?
•
How often do we screen?
•
Who do we screen?
To make better decisions, we need to know what works.
This is inherently a causal question about comparative
effectiveness and safety.
4
Too expensive
(Who will fund the trial?)
Ideally, we would conduct a randomized trial to
inform our decisions…
But randomized trials aren’t always available.
5
Infeasible
(How will we recruit
enough participants to be
adequately powered?)
Not timely enough
(We need to make a decision
now!)
Unethical
(Exposure to harm must be
limited)
Alternative: analyze observational data
•
Imagine a hypothetical
(i.e., target) trial that we
would prefer to conduct
•
A causal analysis of
observational data is an
attempt to emulate that
target trial
6
Alternative: analyze observational data
•
Imagine a hypothetical
(i.e., target) trial that we
would prefer to conduct
•
A causal analysis of
observational data is an
attempt to emulate that
target trial
7
Simple algorithm for causal
inference:
1.
Ask the causal question (i.e.,
specify the protocol of the
target trial)
2.
Answer the causal question
(i.e., conduct the trial or
emulate the trial using
observational data)
Why bother?
Deviations from the target trial can lead to bias
Components of a target trial protocol
Eligibility criteria
Treatment strategies
Assignment
Follow-up
Outcomes
Causal contrasts
Data analysis
8
Let’s consider an example:
Prostate cancer screening with prostate-
specific antigen (PSA) test and prostate cancer
mortality
Target trial summary:
PSA screening and prostate cancer mortality
Eligibility criteria
Treatment strategies
Assignment
Follow-up
Outcomes
Causal contrasts
Data analysis
9
Participants should
be eligible to receive
all treatment
strategies of interest
Target trial summary:
PSA screening and prostate cancer mortality
Eligibility criteria
Treatment strategies
Assignment
Follow-up
Outcomes
Causal contrasts
Data analysis
10
Target trial summary:
PSA screening and prostate cancer mortality
Eligibility criteria
Treatment strategies
Assignment
Follow-up
Outcomes
Causal contrasts
Data analysis
11
Target trial summary:
PSA screening and prostate cancer mortality
Eligibility criteria
Treatment strategies
Assignment
Follow-up
Outcomes
Causal contrasts
Data analysis
12
Point or sustained strategy? Static or dynamic?
Subject matter knowledge needed to identify
strategies that are relevant
Target trial summary:
PSA screening and prostate cancer mortality
Eligibility criteria
Treatment strategies
Assignment
Follow-up
Outcomes
Causal contrasts
Data analysis
13
Target trial summary:
PSA screening and prostate cancer mortality
Eligibility criteria
Treatment strategies
Assignment
Follow-up
Outcomes
Causal contrasts
Data analysis
14
To adjust for confounders is to emulate random
assignment
Alternative: instrumental variable methods
Target trial summary:
PSA screening and prostate cancer mortality
Eligibility criteria
Treatment strategies
Assignment
Follow-up
Outcomes
Causal contrasts
Data analysis
15
Target trial summary:
PSA screening and prostate cancer mortality
Eligibility criteria
Treatment strategies
Assignment
Follow-up
Outcomes
Causal contrasts
Data analysis
16
Need alignment at
time zero
:
•
Eligibility assessment
•
Treatment assignment
•
Start of outcome ascertainment
Otherwise,
immortal time bias
or
selection bias
Target trial summary:
PSA screening and prostate cancer mortality
Eligibility criteria
Treatment strategies
Assignment
Follow-up
Outcomes
Causal contrasts
Data analysis
Target trial summary:
PSA screening and prostate cancer mortality
Eligibility criteria
Treatment strategies
Assignment
Follow-up
Outcomes
Causal contrasts
Data analysis
Target trial summary:
PSA screening and prostate cancer mortality
Eligibility criteria
Treatment strategies
Assignment
Follow-up
Outcomes
Causal contrasts
Data analysis
Observational analyses fail when they deviate from the
target trial protocol
•
Lack of randomized assignment? Maybe…
•
Often, time zero is misspecified
The target trial approach allows you to articulate the tradeoffs
that you are willing to accept in your observational analysis.
20
Observational analyses fail when they deviate from the
target trial protocol
•
Lack of randomized assignment? Maybe…
•
Often, time zero is misspecified
The target trial approach allows you to articulate the tradeoffs
that you are willing to accept in your observational analysis.
We analyze observational data because we have to, not
necessarily because we want to.
21
In general, we can only emulate pragmatic trials
No placebo control (at most, compare against “usual
care”)
No blinding
Treatment strategies must exist in the real world
No tight monitoring or enforcement of adherence
22
What does successful specification and emulation of a
target trial require?
Subject matter knowledge
•
What treatment strategies are relevant?
•
What confounders are we concerned about?
High quality data on treatment, outcome and confounders
Knowledgeable users of the data
•
How are these data collected?
•
How to extract relevant information from complex
databases?
Appropriate causal inference analytics
23
Acknowledgements
24
•
Miguel Hernán
•
Barbra Dickerman
•
Sonja Swanson
•
CAUSALab Group
•
VA-CAUSAL Methods Core
(CSP #2032)
Questions
Email:
joyshi@hsph.harvard.edu
@joy_shi1
25
Slide Note
A collection of visually appealing slides showcasing various images. Each slide features a different image with a description. These high-quality images can be a great addition to any presentation or project.
Download Presentation
Please find below an Image/Link to download the presentation.
The content on the website is provided AS IS for your information and personal use only. It may not be sold, licensed, or shared on other websites without obtaining consent from the author.If you encounter any issues during the download, it is possible that the publisher has removed the file from their server.
You are allowed to download the files provided on this website for personal or commercial use, subject to the condition that they are used lawfully. All files are the property of their respective owners.
The content on the website is provided AS IS for your information and personal use only. It may not be sold, licensed, or shared on other websites without obtaining consent from the author.
E N D
