Statistical methods for pediatric research

点击翻译按钮获取中文摘要

My research centres on three areas: statistical methods for meta-analysis, causal inference, and methods for pregnancy outcomes. With Khajak Ishak (PhD graduate 2006) I have done extensive work on meta-analysis of multivariate data. We demonstrated that meta-analysis of multivariate data does not always provide benefit, in particular when correlation between the multivariate outcomes comes from sources other than correlation among treatment effects (eg study heterogeneity). I will continue to work on a new model for multivariate outcomes based on a Bayesian approach to a multinomial distribution, and individual patient data (rather than summary data). Iwill implement our model using frequentist and Bayesian analytic techniques, both in simulation and in real data. As I and others have noted, pregnancy outcomes pose many statistical challenges. In particular, I am interested in the role of gestational age. Gestational age is an important perinatal predictor, but is subject to both systematic and random misclassification. I will extend my work on measurement error in gestational age to multivariate data; when information on birth length, head circumference and other characteristics are available, they should provide added information on the likelihood of errors in gestational age. In addition, I will develop a more general model for this measurement error, of which my original model represents a special case. Typical analyses of gestational age have treated the measurement error and subsequent analyses as separate problems. Bayesian approaches to survival analysis. Finally, I intend to apply the theory of causal inference based on potential outcomes (Rubin 1976, Robins 1987 and subsequent papers) in several settings. I will develop and compare methods for model selection for inverse probability weighted estimation procedures, and develop principal-stratification based methods for randomized trials with noncompliance.