Page 53 - Contributed Paper Session (CPS) - Volume 2

CPS1419 Jinheum K. et al.
It is well known that standard survival models such as the Cox proportional
hazards (PH) model can be used for analyzing exact or right-censored data.
However, this model is not applicable to DIC data, so some ad hoc approaches
have used midpoints or right points (Law and Brookmeyer, 1992) in the
censored interval as a heuristic. Since then, many researchers have developed
various statistical methods for analyzing DIC data that make procedures more
feasible and provide more powerful results. Among them, Kim et al. (1993)
considered the method of discretizing the distribution into prespecified mass
points and estimating parameters with the Cox PH model. Sun et al. (1999)
extended the result of Kim et al. (1993) to circumvent the non-identifiability
problem and have simpler forms based on an estimating equation method. In
addition, Goggins et al. (1999) focused on the technique of imputing survival
times based on the joint likelihood and estimated effects of covariates using
a Monte Carlo EM algorithm. Meanwhile, Pan (2000) and Pan (2001) extended
multiple imputation methods to assess more accurate survival times from
interval-censored data and to estimate the regression coefficients with the Cox
regression model. Dejardin and Lesaffre (2013) described a stochastic EM
algorithm for evaluating the impact of covariates in the presence of DIC
survival times through a semi-parametric Cox's model.
In clinical trials or cohort studies, subjects are often at risk for one terminal
event. However, in some applications, subjects do not fail from only a single
type of event, but instead are at risk of failing from two or more mutually
exclusive types of events. When an individual is at risk of failing from two
different types of events, these events are called competing risks. Thus, in this
framework, one of the events censors the other and vice versa (Tsiatis, 2005;
Andersen et al., 2012; Putter et al., 2007). On the other hand, many clinical
trials reveal that a subject can experience both an intermediate event, such as
a disease or relapse, and a terminal event, such as death, in which the terminal
event censors the non-terminal event but not vice versa. These types of data
are known as semi-competing risks data (Fine et al., 2001; Xu, et al., 2010;
Barrett, et al., 2010). There has been some work on analysis of semi-competing
risks data by assuming interval censoring only on the intermediate events. For
instance, Siannis et al. (2007) considered a multi-state model to estimate the
regression coefficients in the presence of informative lost-to-follow-up and
interval-censoring. Later, this model was extended by Barret et al. (2011) who
incorporated an illness-death model to deal with a more complicated model
structure. Kim and Kim (2016) employed an extension of the illness-death
model to estimate transition intensities when an intermediate event is interval-
censored.
However, to the best of our knowledge, there has not been any published
research for which both intermediate and terminal events are interval-
censored. We extend previous approaches for analyzing semi-competing risks

42 | I S I W S C 2 0 1 9