SERious EPI

In this episode we talk to Dr. Timothy Lash of Emory University about Quantitative Bias Analysis (QBA). We talk about how QBA is any method that quantifies the impact of non-random error. We talk about direction magnitude and uncertainty. We differentiate from sensitivity analysis, and we talk about how to identify key sources of bias. We talk about bias models and bias parameters and how we draw inferences from bias analyses. We talk about validation data and where you can get it. We talk about why predictive values often aren’t as useful as classification values for bias analysis. We talk about how bias analysis can strengthen your results and that our intuition about the impact of biases is t always great. And we talk about how bias analysis can guide your future research. We differentiate between simple and probabilistic bias analysis. And we end with some examples of cases where bias analysis is really helpful.

In this episode Hailey and Matt talk about Matt’s technology troubles (including having his computer just decide not to let him log on) before we discuss regression discontinuity and difference in difference approaches as part of quasi experimental methods. We focus on what quasi experimental means and encompasses and its relation to natural experiments. We talk about who owns interrupted time series (epidemiologists, economists, other social scientists?). Matt again admits he can’t define exogeneity. We talk about how both designs exploit a threshold when there is a rapid change in the probability of being exposed and we think of those on either side of the discontinuity close to the threshold are exchangeable and we can estimate effects in that population under a set of assumptions. And we talk about how difference in difference takes this same approach but adds a control group. And we debate whether the last difference is singular or plural.

In this episode we talk to Dr. Usama Bilal of Drexel University about Regression Discontinuity Design (RDD) and Difference-in-Differences (DiD), two quasi experimental methods that fall under the instrumental variables framework which we discussed in previous episodes. We talk about what RDD is, the different types (fuzzy vs sharp) and what we are actually estimating (LATE vs CACE). We talk about the bias vs variance tradeoff in how far from the threshold we choose to draw inferences. We talk about the assumptions that are needed for these methods to give valid estimate of effects. Then we talk about DiD and how this is a form of RDD with a second group that does not experience the discontinuity as a control. And we talk about the additional assumptions needed for this approach (e.g. parallel trends).

In this episode, Hailey and Matt discuss whether IVs are rebellious or magical or the midlife crisis of methods. We talk about how they deal with confounding problems. We talk about how they are used to attempt to mimic randomization and the assumptions for IVs. We talk about why it’s so helpful to think about who gets the exposure and why for causal inference. We talk about how IVs fit in with the target trial framework and wham it might tell us about how to teach intro epi. We talk about what estimand IVs estimate. And we relitigate the soda vs pop discussion.

In this episode, we discuss instrumental variables with Dr. Rita Hamad of Harvard’s TH Chan School of Public Health. This episode is focused on the first part of Chapter 28 of Modern Epidemiology 4th edition on quasi experimental methods. We start with what quasi experimental designs are and why we would want to use them (and whether more epidemiologists are being exposed to them). We also talk about why these methods are more common in economics than in epi. We talk about how these methods try to take advantage of something that approximates randomization to estimate causal effects. We talk about what instrumental variables are and the conditions required to be met for a variable to be an instrument. We focus on the strengths and limitations of the methods and when they make the most sense to use them. We talk about what happens when you violate the assumptions of IV. We talk about weak and strong IVs and we talk about Mendelian randomization and its role in epi. And we ask the age-old question, how do you find the elusive instrumental variable?

In this episode we follow up on our conversation about mediation. We talk about what mediation is and when it is useful. We talk about the history of these methods. We debate what direct and indirect effects are. We describe natural and controlled effects. We discuss the importance of the number 666 in Matt’s life. We talk about exposure mediator interaction. Matt learns what kinesiology is. We discuss proportion mediated and proportion eliminated. And we talk about the confounding assumptions needed for mediation analysis.

In this episode, Matt and Hailey talk with Dr. Kara Rudolph and Dr. Ivan Diaz about mediation analysis. We talk through what it is, what it means and when we want to do it. We talk about mechanism of causation and how mediation can help. We cover things like natural direct and indirect effects and controlled direct effects (and why there isn’t a controlled indirect effect – a thing that stumped Matt for some time). And we discuss the different assumptions need to draw valid inferences in a mediation analysis, like all the many no confounding assumptions and the cross world assumption. And we talk about what Matt refers to as mediated moderation (interaction in the effect on the outcome between the exposure and mediator).

In this episode, Hailey and Matt continue on their discussion on study efficiency and realize that we think about efficiency in very different ways. We talk about the difference between statistical efficiency and cost efficiency and we each make our case for one of them being the driving force in how we design and analyze studies. It may be the biggest disagreement we’ve had yet (though maybe that was interaction).We talk about matching and its impact of efficiency and also why we do matching. And we try to understand when matching is useful. Studies mentioned in the podcast: Rothman KJ, Poole C. A strengthening programme for weak associations. Int J Epidemiol. 1988 Dec;17(4):955-9. doi: 10.1093/ije/17.4.955. PMID: 3225112.

In this episode we are joined by Professor Robert Platt of McGill University to talk about study efficiency and the ways we can think about this in terms of study design. We talk about hierarchies of evidence and its relationship to things like target validity. We get into why we think case control studies are so often misunderstood, particularly with respect to missing that they should be nested within a cohort. We talked about the varying definitions of efficient (variance, efficiency of confounding control, cost efficient, etc.) and how they relate to different study designs, and we disagreed about which definition is the most useful. And we talk about sampling and how it affects study efficiency and also what question we are asking. The paper that Rob reads over and over is: Kurth T, Walker AM, Glynn RJ, Chan KA, Gaziano JM, Berger K, Robins JM. Results of multivariable logistic regression, propensity matching, propensity adjustment, and propensity-based weighting under conditions of nonuniform effect. Am J Epidemiol. 2006;163:262-70. We also referenced: Westreich D, Edwards JK, Lesko CR, Cole SR, Stuart EA. Target Validity and the Hierarchy of Study Designs. Am J Epidemiol. 2019;188:438-443. Kramer MS, Guo T, Platt RW, Shapiro S, Collet JP, Chalmers B, Hodnett E, Sevkovskaya Z, Dzikovich I, Vanilovich I; PROBIT Study Group. Breastfeeding and infant growth: biology or bias? Pediatrics. 2002;110(2 Pt 1):343-7.

We kick off season 4 by reminiscing about the origins of the podcast and preview what’s upcoming for season 4 where we will continue on our last season of reviewing Modern Epidemiology 4th edition. We touch on a few of the topics we are most excited about for the coming season and we preview some small formatting changes. But then we put each other through the fun questions that we ask our guests so you all can get to know us better (spoiler: Matt has no idea what the word non-fiction means). We are excited for our upcoming guests this season and the fun conversations we have in store.

It’s hard to believe this is the final episode of season 3! In this season finale episode, we continue our discussion of topics related to Chapter 26 in Modern Epidemiology (4th Edition) with Dr. Eric Tchetgen Tchetgen. In this conversation we ask Dr. Tchetgen Tchetgen to help us better understand several issues related to interaction, including why it’s so important to study interaction. He provides a helpful framework for thinking about interaction: start simple and then move on to more complex questions. As part of this framework, he emphasizes the distinction between total effects and main effects, how confounding plays into conversations about interaction, and the role of scale dependence when interpretating interaction.

Matt and Hailey take a deep dive into Chapter 26 in Modern Epidemiology, 4th Edition, Analysis of Interaction. This episode needs a content warning- it is among the most advanced and conceptually complex topics we have ever covered on SERious Epi. Interaction occurs when the effect of one exposure on outcome depends in some way on the presence or absence of another exposure. Seems like a simple enough concept, right? However, as you’ll see in this episode, there are many different layers of complexity to consider related to terminology, scale, and interpretation of interaction analyses. A note from Matt and Hailey: since this material is very complex, we reached out to Dr. Jay Kaufman for his perspective on the episode before releasing it. He had some very helpful thoughts, and we would like to share them with you (paraphrasing with his permission): Part of what is confusing about this topic is the terminology differences, with Hailey using terminology (“interaction”) that lines up with that used by VanderWeele, ME4, and the Hernán and Robins textbook chapter and Matt using terminology (“interdependence”) from other articles in the literature, such as Greenland and Poole (1988). When there are joint effects that are exactly multiplicative, or supermultiplicative, you know it’s a causal interaction (i.e., synergistic or biologic interaction) because multiplicativity is necessarily super-additive as long as both exposures meet consistency, exchangeability, and positivity assumptions. However, knowing that joint effects are submultiplicative is not informative about additive interaction or synergism. It is also not possible to make a conclusion about additive interaction when a results section tells you only that in a logistic or Cox regression analysis there is “no significant interaction effect (p<0.05)” as that just tells you an effect is not exactly multiplicative. Multiplicativity has some causal implications because it is super additive as long as the causal assumptions listed above are plausibly satisfied. There are several proposed causal mechanisms that would generate multiplicative joint effects especially from the cancer epidemiology literature (e.g., Koopman 1990). In general, considering interaction on the additive scale is more useful for assessing public health relevance (e.g. Panagiotou and Wacholder 2014). Some of these concepts are difficult to convey in podcast format, so we’re including some helpful resources for anyone interested in learning more about this topic. Thanks again to Dr. Kaufman for helping us put this list together: Greenland S, Poole C. Invariants and noninvariants in the concept of interdependent effects. Scand J Work Environ Health. 1988 Apr;14(2):125-9. doi: 10.5271/sjweh.1945. PMID: 3387960. VanderWeele TJ. On the distinction between interaction and effect modification. Epidemiology. 2009 Nov;20(6):863-71. doi: 10.1097/EDE.0b013e3181ba333c. VanderWeele TJ. The Interaction Continuum. Epidemiology. 2019 Sep;30(5):648-658. doi: 10.1097/EDE.0000000000001054. PMID: 31205287; PMCID: PMC6677614. Greenland S, Poole C. Invariants and noninvariants in the concept of interdependent effects. Scand J Work Environ Health. 1988 Apr;14(2):125-9. doi: 10.5271/sjweh.1945. PMID: 3387960. Koopman JS, Weed DL. Epigenesis theory: a mathematical model relating causal concepts of pathogenesis in individuals to disease patterns in populations. Am J Epidemiol. 1990 Aug;132(2):366-90. doi: 10.1093/oxfordjournals.aje.a115666. PMID: 2372013. Panagiotou OA, Wacholder S. Invited commentary: How big is that interaction (in my community)--and in which direction? Am J Epidemiol. 2014 Dec 15;180(12):1150-8. PMID: 25395027.

In this episode, we are joined by Dr. Sonia Hernandez Diaz for a discussion on Chapter 25 in Modern Epidemiology, 4th edition. This chapter is focused on methods for causal inference in longitudinal settings, with a particular focus on time varying exposures. Dr. Hernandez-Diaz helps to explain some of the conceptual and methodological challenges related to time-varying exposures, including the advanced analytic strategies required and the careful conceptual considerations about defining the exposure of interest and causal questions. Papers referenced in this episode: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3731075/ https://academic.oup.com/aje/article/183/8/758/1739860

This episode is focused on Chapter 25 of Modern Epidemiology 4th edition, Causal Inference with Time Varying Exposures. In this episode, Matt and Hailey talk about how we should think about exposures that change over time. We discuss the concept of feedback loops- scenarios where the exposure affects outcome which affects a later time point of exposure and then that exposure affects a later outcome. We think about whether biologic (mechanistic) conceptualizations of feedback loop the same as the epidemiologic notion presented in the chapter. We then follow the chapter to continue our discussion about how time varying exposures change our frameworks for thinking about causal inference and analytic strategies (e.g., marginal structural models, g-formula, and structural mean models). A historical note about Andrew James Rhodes, whose picture is hanging up in the conference room that Hailey was recording from: https://discoverarchives.library.utoronto.ca/index.php/rhodes-andrew-james

Recording from across the globe, in Melbourne, Australia, Dr. Margarita Moreno-Betancur joins us for an episode on Chapter 22 in Modern Epidemiology (4th edition) on Time-to-Event Analyses. This is a chapter focused on the methods we use when the timing of the occurrence of the event is of central importance. Dr. Moreno-Betancur answers all our questions about these types of analyses, including: the importance of the time scale, defining the origin (time zero), censoring vs. truncation. We also ask Dr. Moreno-Betancur to weigh-in on a hot take about whether the Cox Proportional Hazard model is overused in the health sciences literature.

In this episode Matt and Hailey discuss Chapter 22 of the 4th edition of Modern Epidemiology. This is a chapter focused on time to event analyses including core concepts related to time scales, censoring, and understanding rates. We discuss the issues and challenges related to time to event analyses and analytic approaches in this setting including Kaplan Meier, Cox Proportional Hazards, and other types of fancy models that are frequently taught in advanced epi courses (e.g., Weibull, Accelerated Failure Time) but infrequently used in the real-world. The chapter ends with a brief discussion of competing risks. It’s clear that Matt and Hailey need to brush up on concepts related to competing risks and semi-competing risks, and fortunately next month we’ll have an expert join us to answer all of our questions!

In this episode we discuss Chapter 18 in the Modern Epidemiology (4th Ed) textbook focused on stratification and standardization with Dr. Rich MacLehose. We invited the illustrious Dr. MacLehose to be the guest for this chapter because it is one of the most important in the book, linking the theoretical concepts discussed in the early chapters with the advanced analytic techniques discussed in subsequent chapters. In this episode we cover topics such as standardization, stratification, pooling, the use and interpretation of relative and absolute effect estimates, and p-values to evaluate effect heterogeneity.

This is an episode focused on ME4 Chapter 18 (Stratification and Standardization). This is a pretty formula-heavy chapter and I’m sure all of our listeners are tuning in to hear Matt’s voice read them to you: “The sum of M1i times T0i….”. So sorry to disappoint, but instead, we focused this issue on big picture conceptual issues discussed in the chapter. Matt and Hailey talk about the importance of stratification, compare pooling and standardization, discuss Mantel Haenszel and maximum likelihood estimation, and then finish the episode talking about homogeneity and heterogeneity.

In this episode we feature a super expert on all things related to selection bias, Dr. Chanelle Howe. There are a lot of confusing issues related to selection bias: how it’s defined, how it relates to collider stratification bias, whether it’s a threat to internal or external validity (or both!). Chanelle helps us understand many of the nuances related to selection bias and provides helpful resources for readers interested in learning more about the topic. Is a lack of exchangeability related to confounding bias or selection? How can DAGs help us decipher the difference between confounding bias and selection? Can you have selection bias in a prospective cohort study? Join us to find out the answers to all of these questions and much more! Resources: Hernán MA. Invited Commentary: Selection Bias Without Colliders. Am J Epidemiol. 2017 Jun 1;185(11):1048-1050. doi: 10.1093/aje/kwx077. PMID: 28535177; PMCID: PMC6664806. Lu H, Cole SR, Howe CJ, Westreich D. Toward a Clearer Definition of Selection Bias When Estimating Causal Effects. Epidemiology. 2022 Sep 1;33(5):699-706. doi: 10.1097/EDE.0000000000001516. Epub 2022 Jun 6. PMID: 35700187; PMCID: PMC9378569. Howe CJ, Cole SR, Chmiel JS, Muñoz A. Limitation of inverse probability-of-censoring weights in estimating survival in the presence of strong selection bias. Am J Epidemiol. 2011 Mar 1;173(5):569-77. doi: 10.1093/aje/kwq385. Epub 2011 Feb 2. PMID: 21289029; PMCID: PMC3105434.

In this episode, Matt and Hailey discuss all things selection bias. This chapter on selection bias and generalizability is the shortest of the bias chapters in the Modern Epidemiology textbook. Does that mean it’s the simplest? Listen to this episode and decide for yourself!