Online Program

Return to main conference page

Wednesday, January 10
Wed, Jan 10, 5:30 PM - 7:00 PM
Crystal Ballroom CD & Prefunction
Welcome Reception & Poster Session I

Comparison of three methods to control selection bias: an example from a smoking cessation study (304212)

Kathleen Cartmell, Medical University of South Carolina 
Michael Cummings, Medical University of South Carolina 
Clara Dismuke, Veterans Health Adminstration 
*Mary Dooley, Medical University of South Carolina 
Martina Mueller, Medical University of South Carolina 
Georges Nahhas, Medical University of South Carolina 
Annie Simpson, Medical University of South Carolina 
Graham Warren, Medical University of South Carolina 

Keywords: propensity score, inverse probablility treatment weight, selection bias

The goal of this project is to understand potential selection bias in a non-randomized smoking cessation intervention by evaluating three methods. METHODS: A hospital-based tobacco dependence treatment service (TDTS) was implemented to provide bedside and phone follow-up consultations. Thirty-day readmission was compared using: 1) regression with covariate adjustment, 2) inverse probability of treatment weighted (IPTW) propensity score regression with no adjustment and 3) IPTS score regression with covariate adjustment. RESULTS: 1640 received TDTS and 1441 smokers did not. When using method 1, TDTS was significant with a 25% reduction in odds of readmission (p=.015) compared to no intervention. After method 2 results remained significant with a 21% lower odds (p= .043). However, after method 3 the statistical significance approached the a-level with no effect size change (21%, p= .048). CONCLUSION: The use of IPTW indicated that there appeared to be selection bias between those that received TDTS service and those that did not. Even when controlling for the probability of receiving service, without also controlling for potential confounders, treatment effect can be inflated.