Event Details


Conferenciante: Miguel Ángel Luque Fernández

Abstract: Modern Epidemiology has identified significant limitations of classical epidemiological methods, such as outcome regression analysis when estimating causal quantities for the average treatment effect (ATE) using observational data. A limitation of estimating the ATE with regression models is the assumption that the effect measure is constant across levels of confounders included in the model (i.e., that there is no effect modification). Another limitation of parametric modelling rests on the need for correct model specification to obtain unbiased estimates of the true ATE.

To overcome these limitations, Targeted Maximum Likelihood Estimation (TMLE) has been developed, which is a semi-parametric, double robust, efficient substitution estimator allowing for data-adaptive estimation while obtaining valid statistical inference based on the targeted minimum loss-based estimation. Moreover, TMLE allows inclusion of machine learning algorithms to minimise the risk of model misspecification, a problem that persists for competing estimators.

eltmle is the only Stata program implementing TMLE for the ATE for a binary or continuous outcome and binary treatment. eltmle includes the use of an R-based super-learner called from the SuperLearner package v.2.0-2.1 (Polley E., et al. 2011) to calculate predictions of the treatment and outcome models. We are developing the program to be Stata native using Lasso and also calling the Super Learner from Python.

Evidence shows that TMLE typically provides the least unbiased estimates of the ATE compared with other double robust estimators. Nonetheless, recent developments support the use of cross-fit double-robust estimators for data adaptive estimation and we are planning to update eltmle with these functionalities.

The following links provides access to a TMLE tutorial: https://migariane.github.io/TMLE.nb.html and the GitHub repository for eltmle Stata package: https://github.com/migariane/meltmle

Organizadores del ciclo de conferencias: IMAG y el Departamento de Estadística e Investigación Operativa de la Universidad de Granada