Ruta Brazauskas1, Jennifer Le-Rademacher2. 1. Division of Biostatistics, Medical College of Wisconsin, 8701 Watertown Plank Road, Milwaukee, WI 53226, USA. Electronic address: ruta@mcw.edu. 2. Department of Health Sciences Research, Mayo Clinic, 200 First Street SW, Rochester, MN 55905, USA.
Abstract
BACKGROUND AND OBJECTIVES: Clustered competing risks data arise often in genetic studies, multicenter investigations, and matched-pairs studies. In the last two decades, major advances in competing risks theory had been made. Many new statistical methods need to be evaluated via simulation studies. Some mechanisms for simulating clustered competing risks data have been considered in the literature. However, most of them produce data where the strength of the dependence between individuals within a cluster is not clear. In this article, we aim to examine various techniques for generating bivariate competing risks data. METHODS: Theoretical framework for simulating dependent competing risks data using latent failure time approach, multistate models, and shared frailty models is described. The steps needed to implement each method are outlined. Properties of each technique are discussed and standard measures of association are provided in order to assess the degree of dependence in simulated paired competing risks data. RESULTS AND CONCLUSIONS: In addition to describing a variety of techniques to generate dependent competing risks data, the cross-hazard ratios from multiple scenarios for each method are computed. The cross-hazard ratios provide a means to compare the level of dependence of the generated data across methods. This acts as a guide for researchers to select an approach and the parameters needed to achieve the desired degree of dependence for their simulation studies.
BACKGROUND AND OBJECTIVES: Clustered competing risks data arise often in genetic studies, multicenter investigations, and matched-pairs studies. In the last two decades, major advances in competing risks theory had been made. Many new statistical methods need to be evaluated via simulation studies. Some mechanisms for simulating clustered competing risks data have been considered in the literature. However, most of them produce data where the strength of the dependence between individuals within a cluster is not clear. In this article, we aim to examine various techniques for generating bivariate competing risks data. METHODS: Theoretical framework for simulating dependent competing risks data using latent failure time approach, multistate models, and shared frailty models is described. The steps needed to implement each method are outlined. Properties of each technique are discussed and standard measures of association are provided in order to assess the degree of dependence in simulated paired competing risks data. RESULTS AND CONCLUSIONS: In addition to describing a variety of techniques to generate dependent competing risks data, the cross-hazard ratios from multiple scenarios for each method are computed. The cross-hazard ratios provide a means to compare the level of dependence of the generated data across methods. This acts as a guide for researchers to select an approach and the parameters needed to achieve the desired degree of dependence for their simulation studies.