Events

DSECT/DSEN Monthly Seminar Series - Mar 23, 2017

“Comparing approaches for Confounding Adjustment in Secondary Database Analyses:
High-Dimensional Propensity Score versus machine learning algorithms”

Presented by

Dr. Ehsan Karim, PhD
Scientist and Biostatistician, Centre for Health Evaluation and Outcome Sciences (CHÉOS), Saint Paul's Hospital, Vancouver

Thursday, March 23rd at 3-4pm EST

Online via GoToWebinar

RSVP: https://attendee.gotowebinar.com/register/7362182906301773827 

 

Download Event Flyer (PDF) 

Abstract:

The uses of retrospective healthcare claims datasets are frequently criticised for the lack of complete information on potential confounders. Utilising patient's health status related information from claims datasets as proxies of unobserved confounders, the high-dimensional propensity score (hd-PS) algorithm enables us to reduce bias. Using plasmode framework that mimicked a previously published cohort of post-myocardial infarction statin use, we compare the performance of the hd-PS algorithm with a few popular machine learning approaches for confounder selection in high-dimensional covariate spaces: random forest, LASSO and elastic net (compared with respect to bias and MSE of the estimated treatment effect).

This study was funded by CNODES

Learning Objectives:

  1. What is hd-PS and when is it usefulness?
  2. Can we improve the performance of hd-PS by utilising machine learning methods?.

Resources: