Abstract:
|
Competing risks studies can make use of the Fine-Gray(FG) competing risk model or the cause-specific Cox model. Because the FG model cannot account for survey design factors, we propose a solution utilizing a function of survey design weight and FG weight. Method: We analyzed a subset from NHIS and NCHS public-use mortality data. Administrative censoring allows the competing risks model to simplify to a Cox model which allows survey design(M1). To compare to this model, we used a product of the survey design weight and FG weight from the censoring distribution in the partial likelihood function of the subdistribution hazard (model M2). Estimates from M2 were compared to those of M1. Result: Compared to those with a BMI of 18.5-25(normal), the log hazard ratio (and SE) of the estimates were 0.25(0.04), 0.42(0.06), and 0.63(0.08) for those who had higher respective BMIs (25-30, 30-35, and >35) for model M1. The estimates for our alternative approach (M2) were 0.25(0.05), 0.41(0.06), and 0.62(0.08) respectively. Accounting for sampling weights had small effect on the results. Conclusion: The censoring mechanism and survey design weight appear to have small effects on the estimation.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.