Abstract
Competing risks refer to the situation where there are multiple causes of failure and the occurrence of one type of event prohibits the occurrence of the other types of event or alters the chance to observe them. In large cohort studies with long-term follow-up, there are often competing risks. When the failure events are rare, or the information on certain risk factors is difficult or costly to measure for the full cohort, a case-cohort study design can be a desirable approach. In this paper, we consider a semiparametric proportional subdistribution hazards model in the presence of competing risks in case-cohort studies. The subdistribution hazards function, unlike the cause-specific hazards function, gives the advantage of outlining the marginal probability of a particular type of event. We propose estimating equations based on inverse probability weighting techniques for the estimation of the model parameters. In the estimation methods, we considered a weighted availability indicator to properly account for the case-cohort sampling scheme. We also proposed a Breslow-type estimator for the cumulative baseline subdistribution hazard function. The resulting estimators are shown, using empirical processes and martingale properties, to be consistent and asymptotically normally distributed. The performance of the proposed methods in finite samples are examined through simulation studies by considering different levels of censoring and event of interest percentages. The simulation results from the different scenarios suggest that the parameter estimates are reasonably close to the true values of the respective parameters in the model. Finally, the proposed estimation methods are applied to a case-cohort sample from the Sister Study, in which we illustrated the proposed methods by studying the association between selected CpGs and invasive breast cancer in the presence of ductal carcinoma in situ as competing risk.
Highlights
Large epidemiologic cohort studies that require the followup of thousands of subjects for a prolonged period of time can generally be expensive as data collection from participating subjects is resource-demanding
Because blood DNA methylations are expensive to measure, it was only available for a case-cohort sample which included: (1) 335 non-Hispanic white women who were diagnosed with incident breast cancer, i.e., either invasive breast cancer or ductal carcinoma in situ (DCIS), during the time interval between their blood draw during baseline data collection and May 2008, and (2) a random sample of 620 non-Hispanic white women drawn from the 29,026 participants enrolled in the study by June 2007
We implemented the proposed methodology in this dataset to investigate the association between three CpGs, which were identified in a previous study based on the Sister Study [27], and invasive breast cancer (IBC) risk in the presence of DCIS as a competing risk
Summary
Large epidemiologic cohort studies that require the followup of thousands of subjects for a prolonged period of time can generally be expensive as data collection from participating subjects is resource-demanding. A standard approach for competing risks data involves modeling the cause-specific hazard functions of the different competing events under the proportional hazards assumption [4]. Sørensen & Andersen [17] considered proportional cause-specific hazards model, where they generalized the pseudolikelihood approach proposed by Prentice [1] and Self & Prentice [10] for a single event to competing risks setting. We consider a proportional subdistribution hazards model in the presence of competing risks in casecohort studies and examine a weighted estimating equation approach for parameter estimation.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
