Regency EF
A Comparison of Statistical Methods for Estimating Adjusted Hazard Ratio from the National Cancer Database (304089)
*Dongliang Wang, SUNY Upstate Medical University
Keywords: hazard ratio, Cox model, propensity score, NCDB
The National Cancer DataBase (NCDB) has been widely used to retrospectively evaluate different treatment regimens. However interpretation and generalization of the findings require careful justification given the bias and confounding arising from a variety of sources. For instance, Age plays an important role in treatment choice and also has been well known for its close association with survival outcomes. To evaluate the efficacy of neoadjuvent endocrine therapy (NET) as compared to neoadjuvent chemotherapy (NCT), using the NCDB data from 2004 to 2014, the hazard ratio of NET versus NCT is 1.290 if age >70 and 0.841 if age <=70, whereas the marginal HR is 1.492. Several commonly used methods for adjusting potential confounding effects in observational studies are considered, namely, stratified analysis, multivariate Cox regression, propensity score matching, and inverse treatment probability weighting. The performances of these methods are thoroughly compared and discussed.