Biostatistical methods in epidemiology

点击翻译按钮获取中文摘要

The applicant's work involves collaborative research with clinical and population epidemiologists, and development of related biostatistical methodology. The proposed work will be in three main areas: (1) evaluation of diagnostic and screening tests; (2) properties of effect measures commonly used in epidemiology; and (3) design and analysis of clinical trials. Area 1: Specific issues to be considered include: test evaluation with an imperfect reference standard; meta-analysis of diagnostic and screening studies; properties of the summary ROC curve; and extension of latent class models to cover uncertainty in diagnostic markers. Area 2: Two effect measures (NNT and PAR) will be investigated. The NNT  (Number Needed to Treat) index reflects the relative number of patients who benefit from a therapeutic intervention. Among various statistical issues to be investigated is the estimation of a threshold value for NNT, taking into account the values of both harms and benefits of the intervention.  A related focus will be the behaviour of PAR (population attributable risk) estimates in the presence of imperfect data on exposure and/or disease status. Area 3: The applicant will work on basic properties of two new methods - the expertise-based clinical trial, and preference-based analysis of trial data. Research on the verification of treatment blinding procedures, and bias from measurement error in baseline variables will also continue.