Skip to main content

Study Design and Analysis for Causal Inference With Time-Varying Exposures

East Baltimore
4th Term
Academic Year
2023 - 2024
Instruction Method
Class Time(s)
M, W, 10:30 - 11:50am
Auditors Allowed
Yes, with instructor consent
Available to Undergraduate
Grading Restriction
Letter Grade or Pass/Fail
Course Instructor(s)
Contact Name
Frequency Schedule
Every Year
Next Offered
2024 - 2025

Prior enrollment in 140.664 or concurrent enrollment in 140.665.
AND Prior enrollment in 140.621-623 with concurrent enrollment in 140.624 OR Prior enrollment in 140.651-653 with concurrent enrollment in 140.654. Prior enrollment in 340.774.

Many exposures in epidemiology change over time. For example, how does disease risk vary with cumulative exposure to psychosocial stress? When should clinical guidelines recommend switching to a different therapeutic approach? Research questions like these face several challenges. Study designs must be sculpted from data without inducing immortal person-time. Traditional regression analysis can fail under feedback between exposures and confounders. This course will present students with the concepts, study designs, and analytic methods to ask and answer precise and policy-relevant scientific questions about time-varying exposures, applicable to trials, cohorts, and administrative records.
Presents a holistic framework for studying causal effects of time-varying exposures. Builds on 140.664 and 340.774 and explores how to articulate causal questions and clarifies assumptions needed to identify the effects of time-varying exposures. Distinguishes total effects of exposures at a given point in time from those that involve cumulative doses or adherence to dynamic treatment rules. Outlines design parameters such as eligibility, start of follow-up, and artificial censoring with data from cohorts or administrative healthcare records. Reviews the motivation, intuition, and application of advanced methods such as time-dependent propensity scores, marginal structural models, and the parametric g-formula to overcome time-varying confounding and selection-bias. Emphasizes practical application and robustness checks, guideposts for choosing among study designs and analytic methods, and comparative strengths for studies with an etiologic vs. translational focus.
Learning Objectives
Upon successfully completing this course, students will be able to:
  1. Formulate well-defined causal questions in terms of total, direct, sustained, and cumulative effects of time-varying exposures
  2. Encode scientific knowledge into a causal model and identify what data are needed to identify causal effects of time-varying exposures
  3. Craft appropriate study designs from complex longitudinal data that overcome immortal person-time, selection-bias, and feedback between time-varying exposures and confounders
  4. Compare the practical and inferential strengths and weaknesses of advanced methods for studying time-varying exposures and identify when simpler methods suffice
  5. Apply study designs and statistical models to study the effects of cumulative exposure and adherence to complex treatment rules using cohort and administrative healthcare data
Methods of Assessment
This course is evaluated as follows:
  • 80% Group Work
  • 10% Participation
  • 10% In-class Exercises