Speaker: Erin Gabriel (https://eegabriel.github.io/) from University of Copenhagen
Abstract: Recently, it has become common for applied works to weight commonly used outcome regression modeling methods, such as the multivariable Cox model or GLMs, and propensity score weights, with the intention of forming a doubly robust estimator that is unbiased in large samples when either the outcome model or the propensity score model is correctly specified for confounding. Unfortunately, this combination does not, in general, produce a doubly robust estimator, even after regression standardization, when there is truly a causal effect. I will demonstrate via simulation this lack of robustness for the semiparametric Cox model, the Weibull proportional hazards model, and a simple proportional hazards flexible parametric model, with both the latter models fit via maximum likelihood in addition to the non-canonical link GLMs. Interestingly, an IPTW canonical link GLM is doubly robust after standardization or application of the G-formula, and is, in fact, asymptotically equivalent to AIPW estimators, which are optimally efficient if both the propensity and outcome models are correctly specified. I will offer alternative survival modeling methods and suggestions for using doubly robust GLMs.
Seminar series .ics calendar:
https://indico.math.cnrs.fr/category/711/events.ics
Program of the Biostatistics seminars:
https://indico.math.cnrs.fr/e/biostat/
Subscribe to the seminar mailing list:
https://diff.u-bordeaux.fr/sympa/subscribe/seminaire.biostat.bph
Former e-seminars on our YouTube channel (mostly in French): https://www.youtube.com/channel/UCURp-hEQL7k23UzGfqgEurA/videos
Biostatistics seminar series from the Department of Public Health from the University of Bordeaux and the Bordeaux Population Health UMR 1219 research center
Boris Hejblum
