Preparació defensa de Tesis doctoral de Núria Porta
Seminari GRASS - 13 de juliol de 2010
PROGRAM:
10:30 – 12:30 PREPARACIÓ DEFENSA TESIS DE NÚRIA PORTA
TÍTOL: Interval-censored semi-competing risks data: a novel approach for modelling bladder cancer.
Interval-censored semi-competing risks data: a novel approach for modelling bladder cancer.
This PhD thesis is concerned with survival analysis in the presence of multiple endpoints and complex censoring patterns. In this context, we propose a new methodology for interval-censored semi-competing risks data. This work is motivated by the Spanish Bladder Cancer/EPICURO Study, the most important study on bladder cancer ever done in Spain. Our contribution in this study was focused on modelling the course of the disease and on identifying prognostic factors of its evolution.
The course of complex diseases such as cancer or HIV-infection is characterized by the occurrence of multiple events on the same patient, for instance, the relapse of the disease, or death. These events can be terminating events, if the follow-up of the individual is stopped by their occurrence, or intermediate events, if the individual continues under observation after their occurrence. The presence of terminating events complicates the analysis since their occurrence prevents from observing other events later, inducing a possibly dependent censoring.
Appropriate methods are required in this context and, specifically, in this thesis we will focus on competing risks, multi-state models and semi-competing risks. These methodologies will be useful to describe important aspects of bladder cancer course. In particular, two novel contributions to the understanding of bladder cancer result from an appropriate use of competing risks and multi-state models: (1) the characterization of those patients with a high risk of progressing as a first event after diagnosis, and (2) the proposal of a dynamical prognostic model for progression.
Competing risks arises when we model the time until the first of K possible events, together with the indicator of the type of event observed. In the Spanish Bladder Cancer/EPICURO Study we are interested in the time until the first event is observed, distinguishing between recurrence, progression or death. The characterization of this first event is of paramount clinical importance in order to better target the adequate treatment for each patient.
Multi-state modelling is handled by describing all possible paths that the course of disease could follow, and establishing relationships between the events of interest. In the Spanish Bladder Cancer/EPICURO Study, for instance, a patient after is diagnosed could experience a recurrence, and then die, or he/she could die in remission (before any disease-related event is observed). One interesting feature of multi-state models is the possibility to make updated predictions according to the occurrence of intermediate events along time. For the bladder cancer course, we will be able to assess the influence that the occurrence of a recurrence has on the posterior risk of progression.
A special kind of multi-state model is one with an intermediate event, E1, and a terminating event, E2. Denote by T1 and T2 their corresponding time-to-event endpoints. The study of the marginal law for T1 is not addressed neither by the competing risks approach nor by the multi-state modelling. While the competing risks approach allows us to analyse the time T to the first between E1 and E2, that is, T=min(T1,T2), multi-state modelling focus on the conditional law of T2|T1, that is, in how the occurrence of E1 modifies the risk of E2. The distribution of T1 is unidentifiable based only on observed data. The above situation is known as semicompeting risks data (Fine et al. 2001), where the occurrence of the terminating event prevents the observation of the intermediate event, and thus T2 dependently censors T1. The strategy of Fine and colleagues to solve this problem is to assume a joint model for (T1, T2), and then recover the distribution for T1 derived from the assumed joint model.
Our contribution is focused towards the development of new methods in this area of survival analysis. Specifically, we propose a new methodology to deal with interval-censored semi-competing risks data, which arises when the time to the intermediate event, T1, is interval-censored. In many longitudinal studies the occurrence of the intermediate event is evaluated at periodic visits, so T1 is only known to lie between the times of two specific visits. Methods for right-censored semi-competing risks data are no longer valid in this scenario and a new approach is necessary. We extend the semi-parametric method proposed by Fine et al. (2001), which assumes a Clayton's copula model to describe the association between T1 and T2. Our methodology consists of an iterative estimation algorithm which jointly estimates the association structure of the model and the distribution of the intermediate event.
Share: