BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Orthogonal prediction of counterfactual outcomes - Stijn Vansteela
 ndt (Ghent University)
DTSTART:20231013T130000Z
DTEND:20231013T140000Z
UID:TALK206005@talks.cam.ac.uk
CONTACT:Qingyuan Zhao
DESCRIPTION:Many patients in critical care are at high risk of acute kidne
 y injury. Renal replacement therapy can be life-saving\, but also puts pat
 ients at risk\, besides imposing a high financial and logistical burden to
  critical care units. Motivated by this\, we studied the question of when 
 to start renal replacement therapy\, in a close collaboration with intensi
 ve care clinicians and nephrologists. In this talk\, I will reflect on our
  experience with regard to learning an optimal treatment policy based on t
 he Ghent University Intensive Care database. This has motivated methodolog
 ical work on variable importance\, which I will mention briefly\, and on c
 ausal prediction\, which I will discuss in detail. Orthogonal meta-learner
 s\, such as DR-learner\, R-learner and IF-learner\, are increasingly used 
 to predict treatment effects (conditional on patient characteristics). The
 se improve convergence rates relative to naive meta-learners through de-bi
 asing procedures that involve applying standard learners to specifically t
 ransformed outcome data. However\, these transformations lead them to disr
 egard the possibly constrained outcome space\, which can be particularly p
 roblematic for dichotomous outcomes: these typically get transformed to va
 lues that are no longer constrained to the unit interval\, making it diffi
 cult for standard learners to guarantee predictions within the unit interv
 al. To address this\, I will show how to construct Neyman-orthogonal meta-
 learners for the prediction of counterfactual outcomes which respect the o
 utcome space. As such\, the obtained i-learner or imputation-learner is mo
 re generally expected to outperform existing learners\, even when the outc
 ome is unconstrained\, as we confirm empirically in simulation studies and
  illustrate in an analysis of critical care data. 
LOCATION:MR12\, Centre for Mathematical Sciences
END:VEVENT
END:VCALENDAR