Title: A nonparametric survival estimation method for dependent competing risks: An application in relative survival analysis
Authors: Reuben Adatorwovor - University of Kentucky (United States) [presenting]
Abstract: Quantifying disease-specific survival in patients with competing risk is generally done by disease-specific survival analysis when the cause of the event is known. Latent variable model is another method formulated for the unobserved event time. This approach may be the better one for population-based cancer survival studies because disease-specific survival estimates are invalid for the unreliable, misclassified or missing cause of death information. The cause of death due to disease competes with other causes of death, which creates a dependence between the event times. To relax the independence assumption, we formulate the dependence between the time to disease-specific death and the time to other causes of competing mortality using copula. A nonparametric copula-based methodology is used to fit the distributions of disease-specific death and other cause mortality using a function of the Kaplan-Meier estimator. Since the dependence structure for disease-related and other-cause mortality is unknown, we treat the copula as known with a sensitivity analysis conducted across a range of assumed dependence structures. We demonstrate the practical utility of our method through simulation studies with an application to French breast cancer data where we estimated the net and crude survival probabilities which are used for determining prognosis and treatment regimen for disease-specific survival.