Directed Acyclic Graphs (DAGs) with Dr. Hailey Banack

Dr. McGavock: [00:00:00] Welcome to the Emerging Scholar Podcast Season 1. We hope you enjoy this series of great conversations with leading medical researchers that's designed for graduate students, medical trainees, residents, and fellows to help you on your journey towards your first scientific project. Welcome everyone to episode five of the Emerging Scholar podcast.

Today we are interviewing Dr. Hailey Banack, an associate professor at the Dalla Lana School of Public Health at the University of Toronto. Dr. Banack is an epidemiologist with interests in obesity, aging, epidemiological methods, causal inference, and life course epidemiology. Dr. Banack is also the co-host of the Serious Epi podcast series with Matt Fox.

Today we're gonna talk about directed acyclical graphs or DAGs and why they're important at the outset of your research project. This is something that, uh, we [00:01:00] introduced into our lab a couple of years ago, and Dr. Banack is, uh, someone who promotes this a great deal within her own, uh, podcast, and we think it's important for you as a, as an emerging scholar to learn more about DAGs, how you can use them, how you can, uh, apply them to the analyses that you'll be doing on your first, uh, research project.

Uh, looking forward to this discussion with Dr. Banack.

Okay. Welcome everyone. So today's episode, we're gonna talk about DAGs, directed acyclical graphs. And, uh, I'm very honored to have Dr. Hailey Banack here from the University of Toronto, Department of Epidemiology, School of Public Health. Welcome, Hailey.

Can you introduce yourself to us?

Dr. Banack: Hi. Yeah, thanks for having me on. So I am an assistant professor in epidemiology, and my specialty is, um, obesity and aging research mostly, but aside from that substantive area of interest, I love epi methods. I love talking about causal inference. A lot of my research h- is focused on [00:02:00] bias and how far our estimates are from the truth that we want to get to, and so, you know, topics like that.

Directed acyclic graphs are a tool that we use a lot in our research program to understand the relationship between variables and, and you really don't see those relationships until you lay them out in front of you, and so I, I really am a, a big fan of using DAGs in, in research projects.

Dr. McGavock: The reason I thought you would be one of the perfect people to talk about this is I listen avidly to your other podcast, which is called Serious Epi.

And, uh- Thanks ... you're often, uh, explaining how important it is to, to have a DAG and to start with a DAG and how to do a DAG. So for residents and young investigators who are listening right now, can you just tell us what is a DAG?

Dr. Banack: Sure. So a DAG, a, a directed acyclic graph, and it's a graphical representation of the relationships between the variables you're interested in.

We call it a, a causal mechanism. So if you have, uh, an exposure you're interested in an outcome, that would be the simplest kind of DAG that you could set up, and you'd have your exposure and an arrow leading to your [00:03:00] outcome. And so that would be sort of the, the simplest framework, and then you, you build on that.

Are there covariates or confounding variables that you're interested in adding in? Are there variables along the causal pathways, some might call it a mediator variable, in that mechanism? So there's lots of different ways to expand on that, but at its most basic format, it's a picture representing how you view the relationship between your variables.

Dr. McGavock: So for someone out there who's maybe read some manuscripts but actually never seen a DAG, one of the experiences I've had with trainees is they will come in and say, "I want to test this hypothesis between physical activity and type 2 diabetes risk." And everyone else has adjusted for age and sex and smoking status and SES.

And so they have this understanding that maybe we should adjust for these things. But you touched a little bit on the DAG isn't just the covariates, but it's the interrelationships between variables and the exposure, and also those variables and the outcome, and they [00:04:00] could be different. And so can you maybe walk us through how you start drawing DAGs when you're coming up with your hypothesis?

And maybe you could use one of your own experiences around sort of obesity and aging or something like that.

Dr. Banack: Yeah. So, um, I think actually your, the example you gave, um, about physical activity and diabetes is a nice clear example, so, so we can go with that one. So the first thing I do, uh, you need to grab an actual piece of paper.

Just trust me when I say you want to be using a pencil, uh, with an eraser. Don't use DAGitty or any of these online softwares. Don't do it in PowerPoint. Use an actual piece of paper with a pencil and an eraser. And really the first thing I always do is I write out my exposure, so physical activity, PA, or sometimes you might just use X to represent that.

You draw that on the piece of paper, and then you draw a simple arrow leading towards your outcome, which we can call Y, uh, sometimes or you can just write diabetes or T2D or whatever you want to write it as, you know. You really just start as simple as possible let... putting out your variables, your exposure and [00:05:00] outcome, and an arrow connecting them.

One important thing, and the reason that DAGs are called acyclic, is that the arrows can only go in one direction. So we represent temporality between our exposure and our outcome, showing that arrow going in a particular direction. We don't include two-headed arrows. Sometimes you'll see those in- More casual conceptual models, um, if you think that there's, uh, uh, you know, a bidirectional relationship.

But in a DAG, we represent those by writing out our variables a second time. So if you have physical activity, and you think that leads to an increase or decrease risk of diabetes, and then you think someone's diabetes status influences whether or not they're physically active at a later time point, you would write physical activity again on your piece of paper after you've written out diabetes.

So it's, it's a little bit hard to talk through this stuff, um, you know, just using words, but, but that's really how I start, with your exposure and your outcome and a simple [00:06:00] arrow between them. And then when you want to start thinking about variables that are confounders, in epidemiology, we, you know, the, the most simple confounding structure we have basically looks like a triangle, where the left corner is your exposure, the right corner is your outcome, and the top of your triangle is the variable that you're worried or interested in exploring whether it's a confounder.

So let's say we're talking about age as a simple example. So on that piece of paper that I'm working on with my pencil, I add in the word age at the top of the triangle, um, that would connect my exposure and my outcome. And for a variable to be a confounder, it has to be associated with my exposure, so that means you would want to draw an arrow from that word age into the word physical activity or however you've written out your exposure.

The arrow has to go from age to physical activity. And you also need to put an arrow from age to your outcome, so from age to diabetes. The arrow has to go downwards [00:07:00] towards that. So if you're looking at this very simple DAG, you know, you can look at it and see a triangle where it goes your exposure, physical activity, your outcome, uh, diabetes, and the top of the triangle is that age covariant or whatever covary you're, you're interested in thinking about.

Dr. McGavock: Right. And so what's important there is, uh, there are no two-headed arrows.

Dr. Banack: Correct. Yep.

Dr. McGavock: And so you're starting to create what I would say is your worldview.

Dr. Banack: Absolutely. And that's I think a really important point about DAGs and, and why I think it's so important to, uh, use a pencil, and also to draw a DAG with someone else, someone else who knows the topic area and who will say to you, "Why is there an arrow there?"

Or, "Why didn't you put an arrow there?" In DAGs, um, if there's, uh, no arrow between variables, what you are saying is there's no relationship between these two variables. So the absence of an arrow is as important as the presence of an arrow. [00:08:00] And so being able to talk through, "This is my world's view. What do you think about it?"

With a mentor or a peer or a, you know, somebody who, who you're working with, an analyst, um, is very, very helpful to sort of understand where you need arrows and where you don't need arrows.

Dr. McGavock: Right. And which direction those arrows should go in. So- Yes ... one of the things I found the most helpful is trying to distinguish between a confounder and a mediator.

Yes. And if someone comes in and they've labeled something as a confounder, so this third variable has an arrow going to the exposure, also has an arrow going to the outcome, an analyst or another person in the group might say, "Does that really cause physical activity?"

Dr. Banack: Yeah.

Dr. McGavock: You're like, "Oh, actually, no. I think that's why physical activity is linked to diabetes."

And then you-

Dr. Banack: Yeah, this is something- ... change the direction of the arrow ... really drives me crazy and, and it happens a lot in the published literature. Um, my mentor in my PhD once said that when you start to look for this stuff, it kind of ruins research in some way because you begin [00:09:00] to see how many times people are adjusting for a variable.

So they, they're saying, "Oh, th- we believe this is a confounder." But then when you draw out the DAG, it becomes so clear that's not a confounder, that's a mediator, and you don't see that until you start to draw things out. An example that I see all the time that drives me just crazy in my obesity research is people will adjust for statins all the time, you know, especially with some kind of cardiovascular related outcome.

And you have to ask, does taking a statin make you obese, or does being obese make it more likely for you to be prescribed a statin? And surely in my worldview, I think it's the latter. I think it's having obesity makes it more likely for you to have some kind of issue with your cholesterol that requires a statin.

And so adjusting for it is not correct in a standard regression model. And you don't see that until you draw it out in a DAG and you question whether the arrow should go from statin [00:10:00] to obesity or from obesity to statin. And, and these are the situations where DAGs are really helpful

Dr. McGavock: Exactly. And so in my experience, the reason why we got into DAGs, and I'd love to hear about your first introduction to DAGs, was speaking to an analyst.

So, um, one of the things we tell our trainees all the time is talk to the analyst first. Even when you're just thinking about that first hypothesis you wanna test, don't wait until all the data have been collected or you've got access to your data. Bring an analyst in right at the beginning. And by showing them the DAG, those analysts start asking those questions.

And when they see your worldview, it makes it much easier for the group to decide, you know, what our analytic strategy should be. And very often that's the question the analyst will Will bring in is I don't think that can actually cause your exposure. And then you say, "Oh, okay. Well, now we're gonna do a little deeper dive," and it makes more robust analysis.

It doesn't take any more time, more resources. It's just a better thought-out plan, and it starts with a DAG. So for me, the DAG is a way to translate a little bit-

Dr. Banack: Absolutely ...

Dr. McGavock: from the clinician [00:11:00] to the analyst. So do you want to talk a little more about that?

Dr. Banack: Yeah. So, you know, I think one of the strongest things you can do as a researcher is knowing where your expertise lies and knowing what areas you need to collaborate with others to complete a project.

And clinicians have such an amazing amount of expertise, but they're not trained as epidemiologists for the most part. They're not trained as statisticians or analysts. And so you really need to work with these other people, and part of being able to work with other people is translating your ideas. In the same way that as a clinician you have to translate ideas all the time for patients or patients' families.

You know, when you're explaining what's going on, you, you might not use the same medical terminology that you would use when talking to a colleague. And the same is true as, as talking to an analyst or a statistician that you're working with. You need to work on translating an idea, and a DAG is a tool to help you translate that idea.

You know? Even [00:12:00] just listening to me talk about DAGs, you know, you're thinking about writing on a page and arrows and, and this is really hard to translate using words. But translating it using a picture is much easier and is a much more helpful strategy to move the conversation forward. You know, when you're...

You want someone to do an analysis for you, it's a very specific task you're asking them to do, and you need to be as specific as possible in your direction to them so they can give you what you want, which is a result. And so DAGs are just such a useful tool in that regard.

Dr. McGavock: Right. So I've seen multiple types.

The project we were working on with the analyst was, uh, exposures early in life, outcomes in young adulthood. And as we started to do the DAG, it became 50, 60 boxes with 15,000 squiggly lines. I think that was one- Yeah. ... of your podcasts. Right? And so simple versus complex DAGs when you first go to sit down with your grad student or an analyst, what...

How do you do it?

Dr. Banack: Oh, simple for sure. As I [00:13:00] said, start with your exposure and outcome. Start with the variables that you are really sure are confounders or mediators in that relationship, especially if you're asking a question about mediation, uh, which I find is, is often of interest to clinicians. So I always start with simple, and you build layer by layer, level by level, adding in key variables that you're interested in, um, one at a time and seeing, okay, what does this one add, and where do the arrows from this one go?

And DAGs can get very, very complicated when you have 50 different variables and you have to think about the interrelationships between those. But- Those kind of questions I find Often you can keep it down to a simpler level to accomplish the same objective when you're trying to answer the research question, rather than just getting bogged down with there are 50 variables, how am I ever gonna do this?

Start with the five variable version and see where that conversation [00:14:00] takes you, and then you can build on that with time. But there's no need to start out with a, a super complex DAG. I think in a lot of the epidemiology literature, you know, even in, in very complex analyses, they will include a DAG, and they will say, "This is a simplified version of the causal mechanism that we are examining, you know, in this analysis."

Even if it's a complicated analysis, you can still boil it down to what is the actual exposure outcome question I'm interested in answering? You know, and here's a couple of key covariates that I've added, but it doesn't need to be this, you know, tangle of, of 50,000 arrows.

Dr. McGavock: Right. So I think about it like this.

I, I love that idea too, because I think we even published our, our worldview DAG of DOHaD and adult hypertension or adolescent hypertension. It's just almost impossible to look at, right? And understand. So I think about if I'm writing a grant or if a resident, you're writing a two-page proposal to a committee to say, "Are you okay with this?"

Or three, we do [00:15:00] things called works in progress. So now you're gonna- Oh, yeah ... present your work to your other residents and the department, and they don't wanna see 50 variables and 50 lines. They just wanna see five, right? And so you need to just boil that down to something that anyone could digest in about a 30-second viewing of it.

Do you include these in your grants or in some of your, like, early proposals- Oh, yeah ... or do your trainees do that?

Dr. Banack: Absolutely. So I, I almost always include it in a grant. I find it a very helpful tool, again, to translate your idea to a reviewer who's scoring your grant. They can get a quick glance at what you're, you're doing and say, "Okay, this is the r- the exposure outcome they're interested in.

And, you know, this is, this is very helpful for me to understand the big picture of what they're trying to sell me in their grant." So I, I find that very helpful. In presentations, once you've done it on paper and pencil and you kind of have, like, a idea of what the relationships looks like, when I'm presenting on it, I almost always use animation in PowerPoint to describe my DAG because I like to, to walk [00:16:00] them through, this is my thought process.

Here is my exposure. I animate it. I add an arrow with animation. Here's my outcome. Here are the covariates. I'm gonna add them one at a time using animation so those people can understand my thought process and see in the end this is what the whole DAG looks like. But I've built it up in my presentation to help them understand, um, you know, what exactly I'm, I'm trying to accomplish.

And so I almost always use animation if, if that's one tip for presentations.

Dr. McGavock: I agree that when you have an audience in front of you, that's exactly the way to do it. In a grant, it's a little more challenging. Yes. But one of the advantages could also be if you can draw it in a way that if your reviewers are also clinicians.

Mm-hmm. You're grounding your theory in a worldview that they know. So you're saying, "You know physical activity, and you know that people who are active also have these other variables." And I think as a reviewer, you start to feel comfortable, "Oh, yeah, I understand where this person's going with it."

Dr. Banack: And I think that as more people start to use them, [00:17:00] it will become, you know, reviewers will become more comfortable seeing them, and I hope at some point expect them.

You know? I as a reviewer would feel more comfortable that they've really thought out the relationships that they're interested in by including a DAG. Even if I might not agree with everything in the DAG, at least they've kind of laid out, these are the hypotheses I'm interested in, these are what ... This is what I think is going on, and I'd like to do this research to test these hypotheses.

You know? It's another way of, of representing your worldview, your hypothesis on the topic in a, a graphical approach.

Dr. McGavock: And theory over time, if you stay on the same topic, you can actually ... Your DAGs can evolve over time, you know- Oh, yeah ... based on your own work, and you can start putting maybe references in there, which is kind of my next question.

There are some debates about how to do a DAG properly, and should we be advocating for evidence-based DAGs? And so in some of my grants and papers, we're trying to put references to support the lines that we're drawing. Is that something you do or something you [00:18:00] advocate for?

Dr. Banack: So yes, DAGs should be evidence-based because you are saying, "I believe there's a relationship between physical activity and diabetes."

And although, you know, for you and I this is sort of s- old hat, like of course physical activity is so ... But a reviewer might not know that. And so, uh, in the text when I'm describing the DAG, I include a citation that demonstrates that, yes, physical activity is actually associated with diabetes. So I do include references when I'm describing a DAG because I think it's important for reviewers to understand what is known on the topic and what I am trying to test in this research that I'm doing.

One thing I am not a fan of, and maybe it's because I'm a, a DAG purist, but sometimes people try to use, uh, dashed lines, let's say, to show this is what I'm hypothesizing or they try to, uh, put a question mark over the arrow. Yeah. You know, I, I'm not a fan [00:19:00] of accoutrements in my DAG. Um, I just like variables and arrows.

And we do not put a box around a variable unless it's indicating that you are conditioning, so controlling let's say, for that variable. And so it really should just be a word connected to another word with an arrow. Very simple. You know, no colors for things you're unsure about, et cetera. It should be a very simple representation.

We don't need to junk it up with, you know, shading and all this kind of stuff.

Dr. McGavock: So you wouldn't do the DAGitty approach, uh, mediators, moderators, and confounders all have different colors?

Dr. Banack: I don't. I don't use DAGitty. You know, I, I find- Right ... that the time I've spent doing things on DAGitty, uh, just do it in PowerPoint with text boxes and arrows and, you know, you'll be just as, as well served.

I know DAGitty also puts boxes and circles and, and things like that, and that's not the formal DAG approach. If you're just, you know, having a conversation, let's say, and you want to represent the [00:20:00] relationships of interest, w- okay, maybe use DAGitty and throw it in there. But as a DAG purist, um, I am not a huge fan of those things.

Dr. McGavock: So then how do you take your reviewer or your audience from your DAG to your hypothesis? What ... Do you draw that out? Do you just write that out? What does that look like for you?

Dr. Banack: Uh, so typically I explain it in words in the text. So this is the relationship that I'm interested in. DAG figure one shows the causal mechanism that I'm hypothesizing, and I'm interested in understanding how much physical activity or how much sedentary behavior contributes to diabetes risk or, you know, something like that.

I'm, I'm still posing my, my research question and my hypothesis in words supported with the figure over here. It's, it's very similar to the way that when you write a results section of a paper or you're reading someone else's results section, they have their results described in words in the text, and then they have a, a scatter plot [00:21:00] or a bar graph or whatever, you know, figure they're using to support the words that they are representing.

And so I, I think the same is true of DAGs in a introduction section or a beginning of a, a grant.

Dr. McGavock: In there, one of the concepts that we were talking about in another podcast is about data matching. And so one of the ways that I've tried to bring DAGs into a conversation is to say, "Draw out some of the major confounders here."

Right? So you're looking at early life exposure to diabetes and heart disease when you're in your 30s, and do you think that shared genetics plays a big role, and do you think shared environment plays a big role, right? And so you would draw that in a DAG. And then we would take a look at our dataset and say, "Do we have shared genetics?"

And then the analyst- Yes ... might say, "We can't test this hypothesis-

Dr. Banack: Yes ...

Dr. McGavock: based on your worldview- And that's- ... we shouldn't do it with

Dr. Banack: this" ... and that's the brilliance. And that's the brilliance of where bringing in an analyst. You've translated your idea, and you're basically saying, "Is this possible?" You know, as it...

to come back to the example I was just using about sedentary [00:22:00] behavior, you have to have a measurement of physical activity or sedentary behavior in order to test a hypothesis grounded in that exposure. And so if an analyst says, A, "We don't have it," you're kind of in trouble. Uh- Yeah ... or B, you know, "Oh, we have it five years ago," or, "We have it on self-report," or best case, we have accelerometry, you know, d- it, it opens the door to a conversation that you might not have had at that point in your research plan, or you might not have had at all, and, you know, left it up to the analyst, and that's not the best approach either.

And so it's, again, this tool for facilitating, uh, conversations or knowledge translation, um, that, that is really, really helpful. You know, and, and so not just deciding What are the confounders? But do we have measurements of all our, our confounders and the variables of interest? And also, I find them particularly useful when you're dealing with issues related to temporality.

So, you know, understanding [00:23:00] that my exposure precedes my outcome temporally, um, I have a measurement of my exposure temporally prior to my outcome. These are really essential for, uh, establishing causal effects. You have to have that temporal sequence, uh, whether it's time varying. So if you have your exposure measured at multiple time points, do you have your outcome measured at multiple time points?

This is where it's phenomenally useful to draw things out in a DAG because you'll begin to see the temporal ordering between variables.

Dr. McGavock: Does it bother you when someone puts a DAG together and then they use cross-sectional data to, uh, to test that DAG?

Dr. Banack: No, not, no, because I think a DAG could be set up to describe a cross-sectional relationship.

It, it can be set up to describe if I went out today and did a big survey of people in November about physical activity and diabetes. I could set up a DAG representing that survey [00:24:00] done today where I would still draw an arrow from the exposure to the outcome. It's just you're limited in the inferences you can draw and the generalizability of that data, um, you know, based on how it's been collected.

So not, uh, it doesn't bother me if that, uh, whether it's cross-sectional or not, d- just depending on the data you have and how you're setting it up for your question.

Dr. McGavock: Do you think we should be peer reviewing the DAGs, or should we be mandating DAGs so that when you respond to the journal and say, "This person drew a DAG, but they left out this major confounder," right?

In my mind, it's a lot easier when you see the DAG in front of you and you see, oh my God, they left out shared genetics, versus you see, uh, survival curves that have just a list of variables that were adjusted for underneath, or they just said, "Oh, we propensity score matched." And you're like, "Oh, I'm sure you did, but you missed this massive thing," right?

Yeah. I find it's a lot easier for the [00:25:00] reader to say, to do- identify some of those things.

Dr. Banack: Absolutely. And, you know, I think DAGs are a really helpful tool in peer review at the initial design stage. Peer review in the sense of like, "Hey, colleague next door, can you take a look at this? What do you think?" Or, "Hey, analyst, I'd like to work with you on this and, and let's talk about it."

So those are all components of the peer review. But sure, if I'm doing a peer review of an-- a paper, I almost always comment on the DAG, uh, whether I think there's something missing or whether it's helpful or, you know, how can they, uh, modify it to make it more clear or, you know, whatever is going on. But I do think DAGs are, are particularly helpful in the peer review process.

And when it gets to final publication, as you're mentioning, you know, it's really hard to understand the relationship between variables that's not your core research area, and a list of, of covariates is pretty unhelpful for doing that.

Dr. McGavock: And as I've reviewed papers recently, I, I ask for a DAG in an appendix, or I ask for a DAG- [00:26:00] Yeah in a paper. Um, or if I see a grant and they don't have it, I say, "This grant would be a lot stronger if you started with a DAG so that we understood where it was going."

Dr. Banack: Yeah, and to connect to, uh, an earlier question you asked about how simple or complex, uh, DAGs should be. A paper I wrote a few years ago, we included a pretty simple DAG in the paper, but then one of the reviewers said, "I'd like you to include a more complex DAG showing the true relationships between all of these variables in a supplemental appendix."

And sure, it looks like a, a tangle of arrows and, and it's kind of hard to, to parse apart, but it's a good accompaniment for the simplified version that is in the, the published text. Mm-hmm. Um, so that's another good feature of the, the era that we're living in, which is the era of unlimited online appendices, um, where you can basically put anything, um, that's relevant to the paper and, and would be helpful for a reader to understand.

Dr. McGavock: I think for me, one of the take-home messages I've always adopted and, and applied [00:27:00] is that a picture is worth a thousand words. And- For sure ... the more complex this gets. And the last paper I read of yours actually was around collider bias- Mm-hmm ... around pediatric obesity and diabetes risk. And I, I've always had a hard time trying to teach collider bias or try to bring that into the group.

The PhD student in my lab is really good at it. But the graph you put up in that paper, I love the intro too. You said, "We're tired of seeing this obesity paradox that can be- Yeah ... completely explained by collider bias." If I'm right on that, it is a collider bias because of- Yeah ... the way they selected the population, and you put it in a DAG, and I was like, "Oh yeah, that makes a lot of sense to me."

Dr. Banack: Yeah. In particular, I find it's helpful for, you know, issues related to collider bias or selection bias. You know, how you've selected your study sample can have an important impact on the validity of your findings and the generalizability of your findings, and so DAGs are, are particularly useful in that regard as well.

Dr. McGavock: They are, but they're not the be-all and end-all. So the other areas- No ... measurement bias, residual confounding, all those things we can't really pick up with the DAG. You still have to [00:28:00] be vigilant as a reader, but also- Oh, yeah ... vigilant as a designer when you're thinking about your study.

Dr. Banack: Yeah. I mean, DAGs are not going to single-handedly save research.

They're not a, a, a fix for every problem that we have in research. They're just a helpful tool for our toolbox, especially as you are designing studies to understand what's really going on here. Sure, y- you know, if you have a great DAG, well grounded in research, you've workshopped it with some colleagues, and you really think this is what's going on, if you have bad measurements of your variables, you are still in a lot of trouble, and you are probably not going to get a valid answer to the question you're interested in.

You know, the same is true of confounding variables. If you don't measure the covariates, the potential confounders, you can't adjust for them. Even if you've drawn it out in your DAG, you can't adjust for them. So certainly there's many, many, many other problems that can affect your research study that don't get fixed by using a DAG.

Dr. McGavock: And so for residents, uh, trainees who are [00:29:00] trying to get into this and they want to read more, and we're gonna do some, you know, links to different papers, where would you suggest we send these trainees? Which first papers would you recommend that we read?

Dr. Banack: Um, so there is a great paper, um, by Steve Cole, and I call it the egg salad sandwich paper.

And so i- it's published in the International Journal of Epidemiology, and it's a great paper describing how DAGs can help you understand why bias, selection bias in particular, is arising. And so I, I... that would be one of my go-to papers. A more advanced resource would be, uh, Modern Epidemiology, the textbook.

That's kind of the epidemiology bible, um, let's say. Uh, so I think that that would be sort of the, the more advanced version of anyone who wants to get more experience with DAGs.

Dr. McGavock: Excellent. And so- But

Dr. Banack: there's tons of papers out there now.

Dr. McGavock: Yeah. And we'll link 'em. There's a nice one, uh, that's sort of an intro to DAGs for pediatrics, and they, they use similar, you know, s- very simple epi- epi questions that, that I think a, a resident could relate to.

Any other things that we haven't [00:30:00] talked about DAGs or about, uh, introduction to causal inference you might wanna share with residents before we close out?

Dr. Banack: No, I think we've covered it all. I mean, I think the genius of DAGs is helping you think through things before you dive in headfirst. And oftentimes, if you don't think through these things beforehand, it's much harder to fix at a later stage.

And so this is sort of take the time to sit and draw it out now as you're getting started so it saves you time later on. And so I think that that's sort of where all researchers should start, is by drawing out a DAG.

Dr. McGavock: Yeah, in the same way that someone who's designing a home draws it out and then shows it to the- Absolutely potential buyer first before they start hammering that first nail.

Dr. Banack: Yeah. No, and you know, um, an architect draws out plans to discuss with the, the home buyer, and so, you know, that's kind of the same underlying idea. Knowledge translation is a helpful tool for that.

Dr. McGavock: Awesome. Uh, [00:31:00] can't thank you enough. I think this is a great start for residents who are trying to get into this, and, uh, probably the best way you can start to plan out your project.

And so super grateful that you took time to speak with us today. I can't recommend highly enough, uh, the Serious Epi podcast. Uh, you and Matt Fox- Oh, thank you ... do an amazing job. You're great interviewers. Your back and forth is really nice, so I would also recommend that. Any other recommendations, any other tools or, or resources you go to that are great for epi that, uh, we might want to put it as a tagline in here?

Dr. Banack: Oh, I don't think so. Um, let me... I can, I can get back to you on some other resources like that.

Dr. McGavock: All right, sounds good. Okay. Thanks. Thanks again, Hailey, and, uh, we're really grateful. Uh, take

Dr. Banack: care. No problem. Okay. Bye bye.

Dr. McGavock: Bye. That was our conversation with Dr. Hailey Banack about DAGs, an extremely useful tool to add to your research toolbox.

Be sure to join us for episode six when Dr. Norm Rosenblum will give us some insight on how to find a great mentor and how to foster a great mentoring relationship. Thanks for listening. [00:32:00] The Emerging Scholar Podcast was recorded in the UMFM studios located at the University of Manitoba in Winnipeg, Manitoba, Canada.

This is Treaty 1 territory, the original homeland of the Anishinaabe, the Ininew, and on the national homeland of the Red River Métis. Manitoba is also homeland to the Anishininew, the Dakota Oyate, and the Dene peoples. We acknowledge that our water is sourced from Shoal Lake 40 First Nation. As a settler scientist working in health research, I acknowledge TRC Call to Action number 18, that health inequities between Indigenous and non-Indigenous people are the direct result of colonization.

Winnipeg is the center for reconciliation in Canada, and where we are speaking from today is at the crossroads of Canada's history and Canada's future. We are three kilometers away from one of the most sacred meeting grounds for Indigenous people on the plains and within Turtle Island. We are also several kilometers away from one of the last residential schools in the province.

We are working [00:33:00] collectively on reconcili-action with youth, elders, and community leaders to heal from the trauma of colonization, past and present, honor the original treaties, and create a Canada that addresses and eliminates the inequities and injustices of the past and present to create a future for Indigenous children to live and thrive.

On behalf of the team at the Emerging Scholar Podcast, we want to thank Osani Balkaran for the great production for our opening and closing themes. We also want to thank UMFM 101.5, especially co-producer Jared McKetiak for technical assistance and production on this project. We also want to thank all of our guests for the discussion and the insight that they've provided.

Once again, thank you for listening. Be sure to subscribe to the Emerging Scholar Podcast on Apple Podcasts, Spotify, and umfm.com.