Lectures and seminars Sfo Epi Seminar in Biostatistics: Dr. Karla Diaz-Ordaz

Zoom Link:https://ki-se.zoom.us/j/69615386873 

*Webinar format: unless you are on the panel you will be muted and with no video. Participants can raise their hands during the Q&A to be recognized to ask questions.

Abstract

Machine learning (ML) methods have received a lot of attention in recent years especially in settings with a large number of variables. However, causal effect estimation often involves counterfactuals, and prediction tools from the ML literature cannot be readily used for causal inference. In the last decade, major innovations have taken place incorporating supervised ML tools into estimators for causal parameters such as the average treatment effect (ATE). This holds the promise of attenuating model misspecification issues, and enhancing researchers' knowledge with variable selection.  

In this talk, I will review some of these developments incorporating machine learning in the estimation of the ATE of a binary treatment, under the “no unobserved confounding” and positivity assumptions.   In particular,  I will illustrate machine learning for estimation of ATE and will explain why such data-adaptive methods should be avoided in outcome regression or propensity score models. We will then see that doubly robust estimators are valid even when using data adaptive models (such as targeted maximum likelihood estimation with super-learner). 

Throughout, I will use as an illustrative example the evaluation of cancer immunotherapy in Non-Small-Cell Lung Cancer patients, using electronic health records and tumour genomic data from a large USA cohort.

If you have any questions, please contact Erin Gabriel at Erin.Gabriel@ki.se