Estimation of the time-varying average causal effect: the challenge of non-positivity and the impact of model flexibility
When estimating the time-varying average causal effect, challenges to meeting the positivity condition required for making causal inference can easily arise. In this talk, we use as an example the estimation of the per-protocol effect for the Effects of Aspirin in Gestation and Reproduction trial, which quantifies the effect on incidence of pregnancy by 26 weeks had all women been assigned to aspirin and complied versus been assigned to placebo and complied. However, there were clear indications for a practical violation of the positivity assumption, which limited our ability to make causal interpretations. We further compare the results estimated using targeted minimum loss-based estimation and inverse probability weighting approaches and show how the level of flexibility in the estimator impacted the effects of that non-positivity. Finally, we discuss the solutions when faced with such non-positivity.
Jacqueline's research has largely sat at the intersection of causal inference and epidemiologic methods with substantive questions, with work spanning the areas of HIV/AIDS, environmental, and reproductive epidemiology. She is broadly interested in the application of advanced methods to overcome the challenges encountered in complex longitudinal data, such as competing events, time-varying data structures, and generalizability. For example, her post-doctoral work centered on different methodological complications surrounding per-protocol analyses in the Effects of Aspirin in Gestation and Reproduction trial. Currently, she is pursuing similar work within the context of HIV/AIDS observational cohort studies, focusing on the intersection of HIV and chronic comorbidities. She is also dedicated to the dissemination of novel, advanced methods within the epidemiologic community – making these methods more accessible to use while also making their theory, strengths, and weakness better understood. An incredibly important part of this is consistently rooting these methods within the broader context of the causal inference roadmap, both in her own research and when communicating these ideas to others.
This seminar will be held over Zoom and is being recorded.
To join, click here
Or, go to https://monash.zoom.us/join and enter:
Meeting ID: 837 2143 1619
Passcode: 784692