@EpiEllie's Threads
Time for a #tweetorial!
Today, I want to tell you about using simulation models for causal inference, why learning about mediation & mechanisms is hard, and what these two things have in common!
Buckle up, we’re going for a ride!
Today, I want to tell you about using simulation models for causal inference, why learning about mediation & mechanisms is hard, and what these two things have in common!
Buckle up, we’re going for a ride!
First up, who uses simulation models to inform causal decision-making? Lots of people! All kinds of scientists who study infectious disease, health economics, social epidemiology, computational sociology, political science, and more!
There are two main types of simulation models used for decision-making:
•group-level models: eg Markov cohorts, SIR models, decision trees, compartment models
•individual-level models: eg microsimulations, agent-based models, or just ‘individual-level simulation models’.
•group-level models: eg Markov cohorts, SIR models, decision trees, compartment models
•individual-level models: eg microsimulations, agent-based models, or just ‘individual-level simulation models’.
Who’s up for a #tweetorial about adherence, per-protocol effects, and randomized trials?
@_MiguelHernan and I have a new paper out in collaboration with the #CHARM trial team. Will this be the one that finally convinces you? 😄
OA Link: authors.elsevier.com/a/1aUFF_ZI2ghd…
@_MiguelHernan and I have a new paper out in collaboration with the #CHARM trial team. Will this be the one that finally convinces you? 😄
OA Link: authors.elsevier.com/a/1aUFF_ZI2ghd…
First up, what are per-protocol effects and how are they different from intention-to-treat effects?
The per-protocol effect is the difference in the outcome that we would see if everyone in our trial had followed treatment protocol A compared to if everyone had followed treatment protocol B (regardless of what protocol people did or didn’t actually follow!) 
This short paper is a really great example of the importance of being clear about your question & using that to drive your methods.
So, a #tweetorial on research that’s *not* causal, and why it’s OK (& maybe even really good!) that this study does NOT adjust for confounding!👇🏼
So, a #tweetorial on research that’s *not* causal, and why it’s OK (& maybe even really good!) that this study does NOT adjust for confounding!👇🏼
What is this paper all about? The authors describe in the introduction an interesting clinical observation: people who have cancer often develop shoulder problems, & certain types of cancer may be misdiagnosed as shoulder problems.
nature.com/articles/bjc20…
nature.com/articles/bjc20…
This leads them to ask: can shoulder problems be used as an early warning sign to indicate a group of people who might be at high risk of cancer?
Starting 2020 off right by getting the benefit of hindsight & reading some classic papers.
#epielliereads
#epielliereads
1st up, Take the Con out of Econometrics by Edward Leamer, 1983: all about pros & cons of randomized & observational studies plus answers to many of recurrent #econtwitter, #statstwitter, & #epitwitter arguments & even why #ML can’t do causal inference!
jstor.org/stable/1803924
jstor.org/stable/1803924
In Part I randomized experiments vs natural experiments, aka observational studies, Leamer provides a nice summary of some of the problems that can arise in RCTs including that randomization guarantees validity *on average* but that chance imbalance can make any given RCT biased
It’s been a long time since I did a new #cartoonepi #tweetorial, so let’s talk about how CAUSAL GRAPHS can help us think about MEDIATION!
Consider the following example:
•Drug X improves blood pressure.
•Exercise also improves blood pressure.
•People who take Drug X exercise more than those who don’t.
The causal graph (DAG) below shows these variables & their relationships.
•Drug X improves blood pressure.
•Exercise also improves blood pressure.
•People who take Drug X exercise more than those who don’t.
The causal graph (DAG) below shows these variables & their relationships.
Now suppose we want to know *how much* we can improve people’s blood pressure by having them take Drug X and how much of that improvement is not because people taking Drug X exercise more.
What can we do?
What can we do?
Good morning, #epitwitter!
I’m a co-author on @JessGeraldYoung’s new paper out now in Trials & I want to tell you all about it!
Do you use longitudinal data? Are your measurements often enough? Do you know what “often enough” is?
Time for a #tweetorial!
trialsjournal.biomedcentral.com/articles/10.11…
I’m a co-author on @JessGeraldYoung’s new paper out now in Trials & I want to tell you all about it!
Do you use longitudinal data? Are your measurements often enough? Do you know what “often enough” is?
Time for a #tweetorial!
trialsjournal.biomedcentral.com/articles/10.11…
If exposure happens once, then we only need to worry about confounders once too.
But if exposure can happen over time, so can confounding! & if exposure happens every day, so can confounding😰
But we usually don’t have data everyday. Is that bad? Classic epi answer: it depends!
But if exposure can happen over time, so can confounding! & if exposure happens every day, so can confounding😰
But we usually don’t have data everyday. Is that bad? Classic epi answer: it depends!
How often do we need to measure our exposure and confounders to be able to adequately adjust for confounding?
We did a simulation study to find out!
We did a simulation study to find out!
I know I’m bad with change but I don’t I’m being unreasonable when I say I hate the new version of the Twitter list page on mobile. It’s ugly & also what’s the point of the pins if it won’t let me pin anything. Plus, why can’t I choose my own header pic?!
Okay well today it’s letting me pin things, but yesterday that part wasn’t working at all. I still hate it tho 😠
Oh snap, pinning the lists makes them easy to access by swiping & I don’t even have to look at the ugly new list page when I do that. I take it back, I love the new lists 😆😆😆
Speaking of things that are frightening, I just listened to the whole recording of me interviewing Dr Krieger to make sure the audio quality was okay. Why is listening to your own voice so weird?! Something about skull echos?
Anyway, I’ll upload the interview to YouTube later tonight, so stay tuned!
Good morning #EpiPeoplesHealth #epibookclub!
We’re reading chapter 7 this week, which means we’re almost done the book. Can you believe it?!
Chapter 7 is all about Dr Krieger’s theory about population health: the ecosocial model.
We’re reading chapter 7 this week, which means we’re almost done the book. Can you believe it?!
Chapter 7 is all about Dr Krieger’s theory about population health: the ecosocial model.
I think this is my favorite chapter so far, because it finally starts to answer the question of how to incorporate a broad view of social, societal, environmental, and biological determinants of health into our theory.
My biology undergrad focused pretty heavily on ecology, so this theory really resonated with me & I realized that it makes explicit a lot of the assumptions & background I’d been working from all along.
An interesting question that arises from that attractiveness & endometriosis study, is to what extent should researchers be considering the history of other published articles when choosing a journal vs assessing the audience that it reaches?
Are there any journals that have really never ever published a bad take study? Does boycotting journals penalize early career researchers, for example by imposing a requirement to spend extra time reviewing historic articles before choosing where to submit?
Could an argument be made that journals that have good circulation & publish bad takes might even be *more* important places to publish because their audiences may have more of a need to be educated on changing norms in the field?
Whatcha doing twitter? It’s 2:30am & I cant sleep...
Quick 15 min poll: can’t sleep, what should I do?
Welp, looks like you’re in agreement that I should go read, so I’m putting down twitter
Hey, #epitwitter allow me to introduce the #epigif! A whole new level of #epicartoon to inform and delight 😆
If you’d prefer a non-gif version, the inaugural #epigif is based on this Medium post of mine👇🏼 link.medium.com/Bic3m9sw3Y
So, what do you think? More #epigifs? Just stick with #epicartoons? Other ideas?
A #tweetorial about defining your protocol for per-protocol effects, using the LIVE trial of personalized mechanical ventilation for acute respiratory distress syndrome (ARDS).
h/t @otavio_ranzani @f_g_zampieri
Original article in @LancetRespirMed👇🏼
thelancet.com/journals/lanre…
h/t @otavio_ranzani @f_g_zampieri
Original article in @LancetRespirMed👇🏼
thelancet.com/journals/lanre…
First, what is the research question?
The Lung Imaging for Ventilator Setting (LIVE) trial was designed to test if choosing ventilator settings based on lung features (focal vs non-focal) could help improve ARDS survival, relative to a one-size-fits-all ventilation approach.
The Lung Imaging for Ventilator Setting (LIVE) trial was designed to test if choosing ventilator settings based on lung features (focal vs non-focal) could help improve ARDS survival, relative to a one-size-fits-all ventilation approach.
Of 1080 potential participants w/ ARDS assessed for eligibility, 420 were eventually randomized to either the control arm (one-size-fits-all settings) or personalized arm (settings based on focal vs non-focal).
The treatment matrix is sorta complex, so I’m pasting it here👇🏼
The treatment matrix is sorta complex, so I’m pasting it here👇🏼
I have a matching question, #epitwitter:
In cohort studies, matching is supposed to control for confounding so you don’t need to adjust for the matching factors, but in case-control studies, matching just improves efficiency & you still need to adjust for matching factors... 1/3
In cohort studies, matching is supposed to control for confounding so you don’t need to adjust for the matching factors, but in case-control studies, matching just improves efficiency & you still need to adjust for matching factors... 1/3
In a DAG, the case-control situation seems clear to me👇🏼... selection is a consequence of the outcome and the match factor and is therefore a collider.
So, makes sense that we need to adjust for the match factor in a case-control study to block that path & the confounding. 2/3
So, makes sense that we need to adjust for the match factor in a case-control study to block that path & the confounding. 2/3
But, when I draw a similar DAG for a cohort study, isn’t selection a consequence of *exposure* & the match factor & therefore *still* a collider?
Why don’t we need to adjust for the match factor as a confounder / open collider path in a cohort study? What am I missing? 3/3.
Why don’t we need to adjust for the match factor as a confounder / open collider path in a cohort study? What am I missing? 3/3.
Yesterday, @BUSPHEpi #epibookclub was joined by Dr Nancy Krieger
It was so great to be able to discuss with her in person! I definitely understand so much more now!
So, here’s a thread of our convo👇🏼
#EpiPeoplesHealth
It was so great to be able to discuss with her in person! I definitely understand so much more now!
So, here’s a thread of our convo👇🏼
#EpiPeoplesHealth
The 1st question was posed by @ProfMattFox: what was Dr Krieger’s goal in writing this book?
Long story short: “this book is what I wish I had been taught as a student”
She encouraged students to explore their interests even if they don’t fit traditional boundaries of the field
Long story short: “this book is what I wish I had been taught as a student”
She encouraged students to explore their interests even if they don’t fit traditional boundaries of the field
Dr Krieger said something which really resonated with me & is a big part of why I #epitwitter
“The role of educators is to transmit the lessons you’ve learned to others”
If we dont share what we’ve learned, be it theory or methods or experiences, we’re not fulfilling our roles.
“The role of educators is to transmit the lessons you’ve learned to others”
If we dont share what we’ve learned, be it theory or methods or experiences, we’re not fulfilling our roles.
I’m literally melting, #epitwitter! So, let’s have some #epifoodwar fun!
What’s your favorite cool-down treat?
What’s your favorite cool-down treat?
If you had to pick only one of these 4 ice cream flavors to eat for the rest of forever, which would it be?
This is a #epifoodwar, so you gotta pick one! No 5th options allowed! 😆
This is a #epifoodwar, so you gotta pick one! No 5th options allowed! 😆
At #ser2018, I saw an iced custard stand which sounded weird but now I regret not trying. Where can I get some near Boston?
What other variants on ice-based dessert treats have I never heard of? Share your local fav!!
#epifoodwar
What other variants on ice-based dessert treats have I never heard of? Share your local fav!!
#epifoodwar
It’s week 2 of the #EpiPeoplesHealth #epibookclub, so that means it’s time to talk about chapter 2!
This week is about early theories of the patterns and occurrence of disease.
I’m gonna do a short recap & then @nabuelezam and I will have some discussion questions all week!
This week is about early theories of the patterns and occurrence of disease.
I’m gonna do a short recap & then @nabuelezam and I will have some discussion questions all week!
In Ch 2, Dr Krieger introduces us to 4 early theories of disease.
The first is the humoral theory of disease causation of the ancient Greeks.
Under this theory, health requires balance of the 4 humors, which are in turn affected by the environment & the season.
The first is the humoral theory of disease causation of the ancient Greeks.
Under this theory, health requires balance of the 4 humors, which are in turn affected by the environment & the season.
Every person has their own natural state within the system, and based on whether you are more yellow bile or more pleghm or less black bile, there are different expectations for how your health will change throughout the year & following life events... a humoral disease horoscope
I’ve been getting a lot of new followers lately, so how about an #epitwitter #ScholarSunday so newcomers can get connected to our lovely community!
Here’s a thread of some of my suggestions for who to follow & why! Add your own in the comments👇🏼
Here’s a thread of some of my suggestions for who to follow & why! Add your own in the comments👇🏼
Looking for some academic career advice? Have questions about writing grants? Want tips on making delicious pizza or pie? Head on over to @BillMiller_Epi!
Want support for how to #liveyourbestlife, learn to say ‘no’, or throw an amazing party? You’ll be wanting to follow @lisabodnar!
Related to the discussion in this thread👇🏼, here’s a short thread on something I else I’ve been mulling over lately re: Hill’s criteria
#epibookclub #EpiPeoplesHealth #epitwitter
#epibookclub #EpiPeoplesHealth #epitwitter
Is there any way in which they be considered a theory of epidemiologic causation?
If not, & they’re instead just tool for thoughtful evaluation of evidence, can they be applied to non-quantitative evidence or non-causal questions?
Heard about the #EpiPeoplesHealth #epibookclub and want to know how it works? This is the thread for you!
1. We’re reading Nancy Krueger’s Epidemiology and the People’s Health.
2. We started this week, and are reading one chapter per week all summer.
1/n
1. We’re reading Nancy Krueger’s Epidemiology and the People’s Health.
2. We started this week, and are reading one chapter per week all summer.
1/n
3. @nabuelezam and I are co-moderating. This means we’ll post our thoughts, discussion questions, reflections each week.
4. To participate, comment on those threads or post ur own.
5. Make sure to include the tags: #epibookclub and #EpiPeoplesHealth so we can find ur tweets!
2/n
4. To participate, comment on those threads or post ur own.
5. Make sure to include the tags: #epibookclub and #EpiPeoplesHealth so we can find ur tweets!
2/n
6. Follow me & @nabuelezam to make sure you don’t miss our threads!
7. Don’t forget to check out the tags to see & discuss other’s posts
8. In the twitter web version, you can save the hashtags as searches for easy access anywhere, anytime.
9. Last but not least, have fun!
3/n
7. Don’t forget to check out the tags to see & discuss other’s posts
8. In the twitter web version, you can save the hashtags as searches for easy access anywhere, anytime.
9. Last but not least, have fun!
3/n
What is @HarvardChanSPH doing about this, @HarvardChanDean?
And while we’re at it, what has @HarvardChanSPH done for Flint, Michigan?
#thisispublichealth #FlintWaterCrisis
And while we’re at it, what has @HarvardChanSPH done for Flint, Michigan?
#thisispublichealth #FlintWaterCrisis
And then maybe head over to @LittleMissFlint’s page & help her with her latest fundraising goal
#FlintWaterCrisis
#FlintWaterCrisis
At #ser2019, @zepidpy gave a very nice talk about how to avoid bias from left truncation.
But the whole talk I thought he meant left censoring. Turns out censoring & truncation are two completely different ideas! Who knew?!
In case you also didn’t, here’s a #tweetorial!
But the whole talk I thought he meant left censoring. Turns out censoring & truncation are two completely different ideas! Who knew?!
In case you also didn’t, here’s a #tweetorial!
First, let’s make sure we’re all on the same page.
The setting: longitudinal data
The problem: not all data are observed
The setting: longitudinal data
The problem: not all data are observed
So, what is censoring?
Censoring is when you don’t know exactly when (or if) the outcome variable happens for someone.
Censoring is when you don’t know exactly when (or if) the outcome variable happens for someone.
At the #SER2019 plenary, @sandrogalea & @_MiguelHernan are *not* debating about #socialepi and #causalinference because, despite what everyone thinks, they don’t actually disagree!
So, instead, they’re here to set the record straight.
So, instead, they’re here to set the record straight.
If we’re talking about #socialepi, we need a definition. @sandrogalea starts broad: “social epi is the health effects of forces ‘above the skin’” aka everything outside your body
And similarly, #causalinference is about drawing conclusions about causal effects.
My next session on methods for complex data is happening now in Nicollet A!
I’ll be talking about pragmatic trials & per-protocol effects.
I’ll be talking about pragmatic trials & per-protocol effects.
But first, @JessiePBuckley for Alex Keil on a new g-computation approach for causal effects of exposure mixtures.
One benefit over existing approaches: unlike weighted quartile sums regression, this method doesn’t assume all components work in the same direction.
One benefit over existing approaches: unlike weighted quartile sums regression, this method doesn’t assume all components work in the same direction.
There’s a preprint, so I’ll for sure be checking that out! The link is: arxiv.org/abs/1902.04200
