Josy Xi Zhou
The University of North Carolina, USA
Title: Select Potential Covariates to Propensity Score Model for Assessing Cancer Treatment Effect
Biography
Biography: Josy Xi Zhou
Abstract
Background:
Previous studies have shown that including irrelevant covariates in Propensity Score (PS) methods reduces the effect of treatment. For instance, the simulation study illustrates “including variables that are associated with the treatment but not the outcome will decrease the precision of the estimated treatment effect without decreasing bias.” Therefore, these many individual comorbidities have power to change the cancer treatment effect.
Method:
This study conducts an essential algorithm “Combination of Association and Bias (CAB)” in standardized difference (SD), odds ratio (OR), and relative risk (RR), and effect modification (Bias) compared adjustment to un-adjustment. It rests on the exploitation of information of correlation that is usually untapped and includes a variable selection component to national Medicare patients≥65 years with first primary diagnosis 2007-2011. CAB examines 328 covariates including baseline parameters, patients’ characteristics, empirical factors, and other comorbidities associated with radiation-endocrine treatment (RET) within 6 months of diagnosis and the primary outcome recurrence after 1 year of diagnosis.
Results:
Of 9096 patients have 51% received RET. Selected 135 pre-observational covariates modify the effect of RET and recurrence including 38 correlated to RET, 17 related with outcome. RET (RR: 0.5, 95% CI 0.42-0.58) has significant different effect on recurrence by adjusting the covariates related treatment and outcome. If adjusting non-selections (irrelative) covariates in propensity model, the estimation is significantly increased standard deviation and variance compared to selected covariates.
Conclusion:
With large datasets containing many variables, especially cancer studies, the PS theory indicates the most indispensable covariates are those highly correlated with both the selection process of treatment and outcomes. CAB promotes the precision of estimation of treatment effect and limit selection bias.
Keywords: Covariates selection, Propensity Score, Association, Breast Cancer, Treatment Effect