| To analyze bivariate time-to-event data from matched or naturally paired study designs, researchers frequently rely on a random effect called the frailty. This effect, which represents a risk factor that is shared between response measurements in a pair, induces a within-pair dependence.;We identify the statistical properties and features of the model, including a close connection with familiar bivariate copula families. Via theoretical and computational work, we explore aspects of the accuracy and precision of parameter estimation, as well as the consequences, for estimation, of model misspecification.;To address the problem of assessing model fit, we define a, residual for each response measurement pair via the bivariate probability integral transformation of univariate residuals derived from the paired response measurements and the fitted model, and use these to confirm the choice of an appropriate frailty distribution. We also identify a suitable adjustment of this residual if either of the original response measurements is right censored. Through simulation studies and graphical displays, we characterize the sampling behaviour of these residuals, and demonstrate how well-suited these diagnostic tools are to cope with questions of model fit.;For fitting time-to-event data from paired designs such as studies involving two eyes or organs, we introduce a, bivariate accelerated life regression model that uses shared frailties, and describe a flexible computational framework for fitting this model. As implemented in R, this framework enables the user to combine various choices of frailty distributions with different options for the baseline survivor functions of the times to the event of interest within a pair, given the frailty. To illustrate the flexibility of this framework, we describe results that we have obtained for various model combinations via examples drawn from the statistical and medical literature. |
