Competing risks analysis of multiple cancers with application to vitamin D in PLCO
Principal Investigator
Name
Yei Eun Shin
Degrees
Ph.D.
Institution
Biostatistics Branch, Division of Cancer Epidemiology and Genetics, National Cancer Institute
Position Title
Principal Investigator
Email
About this CDAS Project
Study
PLCO
(Learn more about this study)
Project ID
PLCO-904
Initial CDAS Request Approval
Feb 2, 2022
Title
Competing risks analysis of multiple cancers with application to vitamin D in PLCO
Summary
Nested case-control design (also known as risk set sampling or incidence density sampling) is one of the most popular epidemiological subsampling designs to obtain expensive exposures for a subset of a cohort. In nested case-control designs, a fixed number of controls are sampled for each case arising in a cohort from among the set of individuals still at risk at the time of case diagnosis (Langholz and Thomas, 1990). All covariate information is complete only for the sample that consists of all cases and sampled controls. Conditional logistic regression is a traditional estimation method for estimating relative hazards in nested case-control data (Breslow et al., 1978). Of note, it can be shown with a simple exercise that the resulting estimates are equivalent to the estimates from a Cox regression model with a partial data structure.
In survival analyses, it is common to consider one type of cause and treat other causes as independent censoring. However, there needs to consider the possibility that other causes can occur apart from censoring, where those are called competing risks. To this end, it is straightforward to specify a Cox regression model for each cause (also known as cause-specific hazard models). On the other hand, Lunn and McNeil (1995) proposed to apply a Cox regression model to competing risks by augmenting data such that one observation is duplicated into two or more observations with different types of causes and fitting a traditional Cox model with an additional variable indicating the type of events. Although this approach requires assumptions that the ratio of baseline hazard functions of different causes is constant, it enables easier computation for competing risk estimation with existing software and simpler interpretation on parameter estimates.
The aim of our study is to investigate Lunn and McNeil’s competing risk modeling for multiple nested case-control designs where the case-control sampling is performed for each cause. Simulation studies will be performed to justify bias and efficiency of estimation, comparing to a traditional way of fitting cause-specific hazard models, and to explore sensitivity to model assumptions, considering various scenarios.
Langholz B & Thomas DC. Nested case-control and case-cohort methods of sampling from a cohort: a critical comparison. American Journal of Epidemiology. 1990; 131(1): 169–176.
Breslow NE, Day NE, Halvorsen KT, Prentice RL, & Sabai C. (1978). Estimation of multiple relative risk functions in matched case-control studies. American Journal of Epidemiology. 1978; 108(4), 299-307.
Lunn M & McNeil D. Applying Cox regression to competing risks. Biometrics. 1995; 524-532.
In survival analyses, it is common to consider one type of cause and treat other causes as independent censoring. However, there needs to consider the possibility that other causes can occur apart from censoring, where those are called competing risks. To this end, it is straightforward to specify a Cox regression model for each cause (also known as cause-specific hazard models). On the other hand, Lunn and McNeil (1995) proposed to apply a Cox regression model to competing risks by augmenting data such that one observation is duplicated into two or more observations with different types of causes and fitting a traditional Cox model with an additional variable indicating the type of events. Although this approach requires assumptions that the ratio of baseline hazard functions of different causes is constant, it enables easier computation for competing risk estimation with existing software and simpler interpretation on parameter estimates.
The aim of our study is to investigate Lunn and McNeil’s competing risk modeling for multiple nested case-control designs where the case-control sampling is performed for each cause. Simulation studies will be performed to justify bias and efficiency of estimation, comparing to a traditional way of fitting cause-specific hazard models, and to explore sensitivity to model assumptions, considering various scenarios.
Langholz B & Thomas DC. Nested case-control and case-cohort methods of sampling from a cohort: a critical comparison. American Journal of Epidemiology. 1990; 131(1): 169–176.
Breslow NE, Day NE, Halvorsen KT, Prentice RL, & Sabai C. (1978). Estimation of multiple relative risk functions in matched case-control studies. American Journal of Epidemiology. 1978; 108(4), 299-307.
Lunn M & McNeil D. Applying Cox regression to competing risks. Biometrics. 1995; 524-532.
Aims
1. To illustrate the proposed method with application to vitamin D in PLCO.
2. To study competing risks of multiple cancer sites associated with vitamin D.
Collaborators
Stephanie Weinstein, NCI/DCEG/MEB