DF/HCC Program Affiliation

Biostatistics

Research Abstract

Dr. Li's current research interests are in developing methodologies for analyzing correlated discrete and continuous outcome data and correlated censored failure time data. These types of data arise frequently in many fields of biomedical research, such as in spatial studies, survey sampling, clinical trials and familial studies. The correlation among observations can be modeled using random effects (frailty) models and generalized estimating equations. Recently, Dr Li and his colleague propose a new class of Cox proportional hazards model for right censored data that are spatially correlated. Specifically, they assume that marginally each observation follows a Cox proportional hazards model, while specifying the underlying spatial dependence by a random field. A class of semiparametric estimating equations are being developed for drawing inference.

Another area that Dr. Li has been working on is the covariate measurement error problem in the framework of survival analysis. Such problems often occur in observational studies. Dr Li and his colleague have considered an imputation-based method for drawing inference. They have developed a new estimating equation by averaging the score equation for the Cox partial likelihood with respect to the conditional distribution of unobserved true covariates, given observed data. Dr Li and his colleague have also developed a general analytic framework for handling heterogeneous covariate measurement error in the context of survival analysis. Their analytical results have led to a simple bias correcting estimator and a simple sample size formula for designing an observation study, while accounting for the contamination of covariates.

The third area Dr Li is interested in lies in clinical trial designs among patients with unrecognized heterogeneity, such as is conferred by genetic subtype. It is well known such unrecognized heterogeneity among patients, such as is conferred by genetic subtype, can undermine the power of a randomized trial, designed under the assumption of homogeneity, to detect a truly beneficial treatment. Dr Li and his colleagues are developing a conditional power approach for the weighted log rank test to allow for recovery of power under unexplained heterogeneity, while controlling the type I error.

Dr Li is also interested in cure rate modeling. He and his colleague are investigating the use of a modified Cramer-von Mises statistic for two-sample survival comparisons with nonnegligible cure fractions. The developed techniques have in-depth statistical applications, for example, computing Loeve principal component decomposition of a complex stochastic process without resorting to resampling schemes.

Dr. Li is actively involved in collaborative research in cancer clinical trials and community-based research with researchers from Harvard School of Public Health and Dana-Farber Cancer Institute.

Publications

No representative citations available.