A0988
Title: Competing risks analysis using mixture cure cause-specific hazard models with partly interval censoring
Authors: Serigne Lo - University of Sydney / Melanoma Institute Australia (Australia) [presenting]
Joseph Descallar - School of Mathematical and Physical Sciences - Macquarie University - Sydney NSW 2109 (Australia)
Houying Zhu - Macquarie University (Australia)
Jun Ma - Macquarie University (Australia)
Abstract: In medical studies, competing risks survival data are commonly estimated using partial likelihood methods, with the assumption that any right-censored patient will experience one of the competing events beyond the conclusion of the study period. In some cases, however, a patient may be considered "cured" from the risks of interest, meaning that none of the risks will occur, resulting in a cured fraction. Furthermore, if disease progression is the event of interest, the exact event times are unknown and often subject to interval censoring. When dealing with such data, employing the standard analysis approach based on cause-specific hazards models while ignoring the cured fraction could lead to biased parameter estimates. A novel approach is introduced that accommodates interval censoring and a cured fraction within cause-specific Cox models when analyzing competing risks data. More specifically, a new maximum penalized likelihood approach is proposed to simultaneously estimate logistic regression parameters for the cured fraction, cause-specific Cox models regression coefficients, and their non-parametric baseline hazards. Asymptotic properties are developed, and simulation studies show reduced bias and improved coverage probability compared with the partial likelihood approach with a mid-point imputation.
