Dongsheng Tu, Queen’s University
Mathematics and Statistics
Professor
Kingston, Ontario
dtu@ctg.queensu.ca
Office:
(613) 533-6000 ext. 77830
Expert In
Bio/Research
I am one of the Senior Biostaticians in the National Cancer Institute of Canada Clinical Trials Group cross-appointed to the Department of Mathematics and Statistics. I spend most of my time in the design, management and analysis of the clinical trials sponsored by National Cancer Institute of Ca...
Click to Expand >>
Click to Expand >>
Bio/Research
I am one of the Senior Biostaticians in the National Cancer Institute of Canada Clinical Trials Group cross-appointed to the Department of Mathematics and Statistics. I spend most of my time in the design, management and analysis of the clinical trials sponsored by National Cancer Institute of Canada, US National Cancer Institute, and pharmaceutical companies. My research in statistical methodology is driven by the need to solve some interesting mathematical and statistical problems in the design and analysis of clinical trials. For example, one of the trials I analyzed was selected as a pivotal study for the final US Food and Drug Administration (FDA) approval of marketing Epirubicin in US as an adjuvant treatment for early breast cancer. During the analysis of this trial and the presentation of the results to the FDA, an interesting question on whether statistical tests of treatment effect should be adjusted for baseline patient characteristics has come out. In a recent paper coauthored with two physicians, we first discussed some statistical principles and regulatory aspects around this issue. A student has worked with me to explore this question further to identify what is the best method to perform the covariates adjustments when an adjustment is necessary. The results were presented recently at the annual meeting of the Society for Clinical Trials.
Another example is my research on statistical issues in the design and analysis of equivalence clinical trials. The objective of these trials is to show that a new treatment has the same efficacy as a standard treatment, which is very different from that of conventional trials to show one treatment is different from another. Traditional methods of statistical analysis cannot be applied to the design and analysis of these trials. I have published several papers that derived mathematical formulas for sample size determination and developed some statistical procedures for statistical analysis when the primary endpoint of the trials is binary or ordinal categorical. I recently supervised a student to study statistical procedures for the analysis of equivalence clinical trials with survival endpoint.
Recently, the investigation of prognostic factors for patients with ovarian and breast cancers stimulated my interests in studying some general issues on the applications of Cox proportional hazards regression models in cancer clinical trials. I first supervised a student to study the applications of some resampling methods to variable selection and model validation of the Cox models. Collaborating with a visiting scholar from Beijing University, I proposed a Bartlett type adjustment to Rao.s score test in Cox proportional hazard models. Recently I worked with a student to explore the applications of some nonparametric regression methods as alternatives to the Cox model in the identification of prognostic factors for cancer patients.
Click to Shrink <<
Another example is my research on statistical issues in the design and analysis of equivalence clinical trials. The objective of these trials is to show that a new treatment has the same efficacy as a standard treatment, which is very different from that of conventional trials to show one treatment is different from another. Traditional methods of statistical analysis cannot be applied to the design and analysis of these trials. I have published several papers that derived mathematical formulas for sample size determination and developed some statistical procedures for statistical analysis when the primary endpoint of the trials is binary or ordinal categorical. I recently supervised a student to study statistical procedures for the analysis of equivalence clinical trials with survival endpoint.
Recently, the investigation of prognostic factors for patients with ovarian and breast cancers stimulated my interests in studying some general issues on the applications of Cox proportional hazards regression models in cancer clinical trials. I first supervised a student to study the applications of some resampling methods to variable selection and model validation of the Cox models. Collaborating with a visiting scholar from Beijing University, I proposed a Bartlett type adjustment to Rao.s score test in Cox proportional hazard models. Recently I worked with a student to explore the applications of some nonparametric regression methods as alternatives to the Cox model in the identification of prognostic factors for cancer patients.
Click to Shrink <<