Randomized controlled trials (RCTs) are considered the gold standard approach for assessing the causal effect of an intervention on an outcome. However, they suffer from a lack of external validity when the population eligible for the RCT is significantly different from the target population of the intervention policy. On the other hand, observational data are often representative of the target population but suffer from confounding biais due to the lack of controlled experimental intervention. In this work, we combine the information gathered from experimental and observational data to generalize the results of an RCT to a target population.
In order to identify the average treatment effect, one requires an ignorability assumption that implies that we observe all variables that are treatment effects modifiers and are shifted between the two sets. Standard estimators then use either weighting (IPSW), outcome modeling (G-formula), or combine the two in doubly robust approaches (AIPSW). However such covariates are often not available in both sets.
Therefore, after completing existing proofs on the complete case consistency of those three estimators, we compute the expected bias
induced by a missing covariate, assuming a semi-parametric linear model for the outcome. This enables sensitivity analysis for each missing
covariate pattern (missing in the RCT, in the observational data or in both), giving the sign of the expected bias. We also show
that there is no gain in imputing a partially-unobserved covariate. We illustrate all these results on simulations, as well as on an example from critical care medicine.
This is a joint work with Benedicte Colnet, Erwan Scornet and Gael Varoquaux.

Séminaire en salle 109 (IMAG, bâtiment 9, salle de conférence).
Également retransmis sur zoom :
https://umontpellier-fr.zoom.us/j/94087408185