|
Research Projects Supported by HKU's High Performance Computing Facilities |
|
¡@ |
| Researcher: |
| Dr Kwok-fai Lam, Department of Statistics and Actuarial Science |
| ¡@ |
|
Project Title: |
| Semiparametric Regression Analysis with Current Status Data |
| Project Description: |
|
In a randomized clinical trial
study where the response variable of interest is the time to occurrence of a
certain event, it is always too expensive or even impossible to observe the
exact time to event. However, the current status of the subject at a random
time of inspection is much more natural, practical, and feasible in terms of
cost effectiveness. This type of data is called current status data that
arise frequently and naturally in medical research. For example, in a
bioassay study concerning the time to occurrence of tumors induced by
exposure to a toxic agent, only the current status of the subjects can be
observed at the time of examination at a random time. Semiparametric
regression analyses of current status data have been considered by various
authors using different models such as the semiparametric proportional
hazards and odds models that all covariates are assumed to have a linear
relationship with some function of the failure time T, Q(t). In the cases
where a certain covariate may have a nonlinear relation with Q(t), the
methods available in the literature may no longer be appropriate. In the
first part of this project, we aim to propose a methodology to analyze
current status data where some clinical factors are assumed to have a linear
relationship with Q(t) while a particular clinical or risk factor W has a
nonlinear relationship with Q(t). The nonlinear relation will be estimated nonparametrically. This project aims to help answering some clinical questions by exploring the nonlinear relationship between Q(t) and some potential explanatory variable W arise from current status data in various situations. |
| Project Duration: |
| December 2003-December 2005 |
| Project Significance: |
| The first situation is a typical setup in survival analysis that all subjects in the population are susceptible to the event of interest, but a certain clinical factor has a nonlinear relationship with the failure time T. For example, the effect of the amount of dosage of certain medication on the time to reaction may attain its maximum at some dosage level W0 and then decreases with W or retain at the maximum level for W > W0. The main objective of this part is to study the estimation of the model parameter in the presence of the nonlinear relationship. The second situation aims at modeling the current status data that hypothesize subpopulation of individuals highly susceptible to some types of adverse events while others are assumed to be at much lower risk, say recurrence of breast cancer tumors. There has been a recurring interest in the estimation of cured proportion from a sample subject to right censoring. With the current status data, a mixture model is suggested that combines a logistic regression model for the probability of cure with the semiparametric regression model proposed in the first part of the project is suggested. The cured fraction using different treatment combinations, as well as the linear and nonlinear relationships between various clinical factors and the failure time T can be estimated. The asymptotic properties of the proposed estimators will be developed and the performance of the estimators will be studied empirically. With these powerful and flexible methods for analyzing current status data, randomized controlled clinical trials can be carried out at a much cheaper cost and some nonlinear relationships between the failure time and some clinical factors can be explored easily. |
| Results Achieved: |
|
Expected Publications: 1. Sieve MLE for Semiparametric Regression Models with Current Status Data. H. Xue, K.F. Lam & G. Li. Journal of the American Statistical Association 2004. 2. A Semiparametric Regression Cure Model with Current Status Data. K.F. Lam & H. Xue. In Progress. 3. Zero-Infated Semiparametric Poisson Regression Model. K.F. Lam & H. Xue, In Progress. |
| Remarks on the Use of High Performance Computing Cluster: |
| The project is highly computational demanding and would not have been so smooth without the HPC Cluster in carrying out the simulations and computations. |