I would like to thank Dr. Caniglia and Dr. Schisterman for the opportunity to talk in this session. Today I will discuss the implementation of the ttf in the study of social determinants in dementia research. I should note that through out my talk I will make more questions than give answers, I am confident that this is the perfect place to pose them and take this talk as a conversation starter.
The national institute of aging has outlined this framework including different exposures or risk factors that drive disparities in dementia and other aging-related outcomes. The final goal, and here I quote: to understand the etiologic pathways and ultimately identify effective targets for intervention... which means, asking causal questions.
However there has been a long resistance to connect causal methods to structural factors is because the best study design is a randomized trial.
While we are mostly focusing on RCTs to answer questions for biological mechanisms, we still need the best of these methods to study structural factors.
L. Paloma Rojas-Saunero MD, PhD
Postdoctoral scholar
Mayeda Research Group, Department of Epidemiology
I would like to thank Dr. Caniglia and Dr. Schisterman for the opportunity to talk in this session. Today I will discuss the implementation of the ttf in the study of social determinants in dementia research. I should note that through out my talk I will make more questions than give answers, I am confident that this is the perfect place to pose them and take this talk as a conversation starter.
Hill et al. Ethnicity and disease. 2015
The national institute of aging has outlined this framework including different exposures or risk factors that drive disparities in dementia and other aging-related outcomes. The final goal, and here I quote: to understand the etiologic pathways and ultimately identify effective targets for intervention... which means, asking causal questions.
However there has been a long resistance to connect causal methods to structural factors is because the best study design is a randomized trial.
While we are mostly focusing on RCTs to answer questions for biological mechanisms, we still need the best of these methods to study structural factors.
No loss to follow-up
Full adherence through out the study duration
No loss to follow-up
Full adherence through out the study duration
Double blind assignment
No loss to follow-up
Full adherence through out the study duration
Double blind assignment
Time points aligned by design:
Before we go into the nuance of social exposures, I would like to highlight some of the features of an ideal randomized trial.
With this in mind, we can consider the ttf as a method to analyze observational data, that motivates researchers to conceptualize the hypothetical randomized trial that we would like to conduct that answers our question
That way we can start by defining a clear estimand, and outline the protocol elements of that target trial
And then attempt to emulate the target trial with observational data as close as possible. Finally under certain assumptions and with adequate statistical methods, we can estimate a causal effect.
This workflow shows that an essential part of this framework relies on the study design. Sometimes we heavily focus on this last part, but the essence of the ttf is to be more transparent about the study design decisins we face when we use observational data that has not been collected specifically to answer one specific research question.
With that prior, I would like to discuss an ongoing project, lead by my amazing collaborator TM.
What is the effect of early-life racial residencial segregation in 1940, measured by the dissimilarity index, on later-life memory decline in US population, represented by the Health and Retirement Study (HRS)?
Work led by Taylor Mobley (UCLA, Mayeda's research lab)
THe HRS is a national representative data with oversampling of black, latino population. Recruitment started in the 90's and collects all kind of measurements, including memory scores every 2 years.
So now I am going to walk you through the outline of the target trial and how we would emulate it... so hold on to your seats because we are going to be creative
Section | Target trial protocol | Emulation using observational data |
---|---|---|
Eligibility criteria | Population who self-identify as white or Black, born in the US by 1940, who resided in a county with Black residents | Same + participated in HRS, with linked data for 1940 census, living in counties with > 1 enumeration district and with memory assessment measured by 1998 |
First, lets imagine we can take a time machine that transport us to 1940.
Section | Target trial protocol | Emulation using observational data |
---|---|---|
Eligibility criteria | Population who self-identify as white or Black, born in the US by 1940, who resided in a county with Black residents | Same + participated in HRS, with linked data for 1940 census, living in counties with > 1 enumeration district and with memory assessment measured by 1998 |
Policy-level interventions | 0. Do nothing (Status quo) 1. Reduce racial residencial segregation at a county level by certain amount |
Same |
Section | Target trial protocol | Emulation using observational data |
---|---|---|
Eligibility criteria | Population who self-identify as white or Black, born in the US by 1940, who resided in a county with Black residents | Same + participated in HRS, with linked data for 1940 census, living in counties with > 1 enumeration district and with memory assessment measured by 1998 |
Policy-level interventions | 0. Do nothing (Status quo) 1. Reduce racial residencial segregation at a county level by certain amount |
Same |
Randomized assignment | Random assignment to either strategy in 1940 | Random assignment in 1940, within levels of county level covariates |
Section | Target trial protocol | Emulation using observational data |
---|---|---|
Eligibility criteria | Population who self-identify as white or Black, born in the US by 1940, who resided in a county with Black residents | Same + participated in HRS, with linked data for 1940 census, living in counties with > 1 enumeration district and with memory assessment measured by 1998 |
Policy-level interventions | 0. Do nothing (Status quo) 1. Reduce racial residencial segregation at a county level by certain amount |
Same |
Randomized assignment | Random assignment to either strategy in 1940 | Random assignment in 1940, within levels of county level covariates |
Start/End of follow-up | 1998 (1st memory assessment) until 2018, including every wave of data in between | Same |
Section | Target trial protocol | Emulation using observational data |
---|---|---|
Eligibility criteria | Population who self-identify as white or Black, born in the US by 1940, who resided in a county with Black residents | Same + participated in HRS, with linked data for 1940 census, living in counties with > 1 enumeration district and with memory assessment measured by 1998 |
Policy-level interventions | 0. Do nothing (Status quo) 1. Reduce racial residencial segregation at a county level by certain amount |
Same |
Randomized assignment | Random assignment to either strategy in 1940 | Random assignment in 1940, within levels of county level covariates |
Start/End of follow-up | 1998 (1st memory assessment) until 2018, including every wave of data in between | Same |
Outcome | Composite memory score | Same |
Section | Target trial protocol | Emulation using observational data |
---|---|---|
Eligibility criteria | Population who self-identify as white or Black, born in the US by 1940, who resided in a county with Black residents | Same + participated in HRS, with linked data for 1940 census, living in counties with > 1 enumeration district and with memory assessment measured by 1998 |
Policy-level interventions | 0. Do nothing (Status quo) 1. Reduce racial residencial segregation at a county level by certain amount |
Same |
Randomized assignment | Random assignment to either strategy in 1940 | Random assignment in 1940, within levels of county level covariates |
Start/End of follow-up | 1998 (1st memory assessment) until 2018, including every wave of data in between | Same |
Outcome | Composite memory score | Same |
Causal contrast | Intention-to-treat | Same |
E[Ya=11998]−E[Ya=01998]
E[Ya=11998]−E[Ya=01998]
E[Ya=1,c=02000]−[Ya=0,c=02000]
....
So now we can outline our estimand or paratemer of interest. Since we our outcome of interest is memory decline, we discretized our estimand, by each time point. So first, we would be interested in evaluating the mean memory score in 1998 had everyone had followed intervention 1) compared to the mean score at baseline had everyone had followed intervention 0)
Next we would look at the mean difference on the memory score, under the two treatment strategies, and on and on we would do this until the last wave of data, and we could plot these point estimates over time, either as by treatment arm or the difference between them.
E[Ya=11998]−E[Ya=01998]
E[Ya=1,c=02000]−[Ya=0,c=02000]
....
E[Ya=1,c=02018]−E[Ya=0,c=02018]
So now we can outline our estimand or paratemer of interest. Since we our outcome of interest is memory decline, we discretized our estimand, by each time point. So first, we would be interested in evaluating the mean memory score in 1998 had everyone had followed intervention 1) compared to the mean score at baseline had everyone had followed intervention 0)
Next we would look at the mean difference on the memory score, under the two treatment strategies, and on and on we would do this until the last wave of data, and we could plot these point estimates over time, either as by treatment arm or the difference between them.
Consistency
Exchangeability
Positivity
No interference
Refers to having a well defined intervention.
Refers to having a well defined intervention.
What does it mean to intervene on residential segregation in 1940?
Is the dissimilarity index a good a proxy of all the racist policies that resulted in residential segregation?
Refers to having a well defined intervention.
What does it mean to intervene on residential segregation in 1940?
Is the dissimilarity index a good a proxy of all the racist policies that resulted in residential segregation?
What is a relevant intervention on the dissimilarity index, reducing under certain threshold or by certain value?
The target trial framework can help us expand the narrative and conversation around these complex exposures.
At best, this hypothetical intervention would be reducing the impact of all the racist policies that influenced residential segregation, determined by the frequency of these in the studied population.
The target trial framework can help us expand the narrative and conversation around these complex exposures.
At best, this hypothetical intervention would be reducing the impact of all the racist policies that influenced residential segregation, determined by the frequency of these in the studied population.
It requires deep interdisciplinary dialogue, and only with subject matter expertise we can inform how well we feel about satisfying this assumption.
Considering what variables to adjust for is more challenging.
Since there are many factors that can influence residential segregation, whatever we adjust for is not accounted into how we would intervene.
Considering what variables to adjust for is more challenging.
Since there are many factors that can influence residential segregation, whatever we adjust for is not accounted into how we would intervene.
There is a moral and ethical value in terms of what variables to adjust. "Non-allowable" sources are considered unfair and contribute to disparities (Jackson J. Epidemiology. 2021).
The target trial framework helps us in being more transparent about our questions, assumptions and interpretations.
We can use the target trial framework even if the intervention of interest is not measured (or exists!).
The target trial framework helps us in being more transparent about our questions, assumptions and interpretations.
We can use the target trial framework even if the intervention of interest is not measured (or exists!).
Having a better design can already prevent some sources of bias, and help us detect other sources that could be prevented (or quantified) through the analytic strategies.
The target trial framework helps us in being more transparent about our questions, assumptions and interpretations.
We can use the target trial framework even if the intervention of interest is not measured (or exists!).
Having a better design can already prevent some sources of bias, and help us detect other sources that could be prevented (or quantified) through the analytic strategies.
It is a dynamic process, since it requires a deep understanding of the data sources and a constant check that the causal contrasts and subsequent results are informative.
Buck C. Popper's Philosophy for Epidemiologists, IJE. 1975
Mayeda Research Group
Study Contributors:
NIA: R01AG074359
Hill et al. Ethnicity and disease. 2015
The national institute of aging has outlined this framework including different exposures or risk factors that drive disparities in dementia and other aging-related outcomes. The final goal, and here I quote: to understand the etiologic pathways and ultimately identify effective targets for intervention... which means, asking causal questions.
However there has been a long resistance to connect causal methods to structural factors is because the best study design is a randomized trial.
While we are mostly focusing on RCTs to answer questions for biological mechanisms, we still need the best of these methods to study structural factors.
Keyboard shortcuts
↑, ←, Pg Up, k | Go to previous slide |
↓, →, Pg Dn, Space, j | Go to next slide |
Home | Go to first slide |
End | Go to last slide |
Number + Return | Go to specific slide |
b / m / f | Toggle blackout / mirrored / fullscreen mode |
c | Clone slideshow |
p | Toggle presenter mode |
t | Restart the presentation timer |
?, h | Toggle this help |
Esc | Back to slideshow |