Post date: Sep 11, 2014 2:32:34 PM
9/5/2014 Short Course "Diagnostic tests and biomarkers I"
Zheyu Wang, Division of Biostatistics and Bioinformatics, SKCCC, JHU
9/5/2014 Short Course "Diagnostic tests and biomarkers II"
Zheyu Wang, Division of Biostatistics and Bioinformatics, SKCCC, JHU
9/19/2014 Short Course "Diagnostic tests and biomarkers III: Time-dependent ROC analysis" (slides)
Speaker: Mei-Cheng Wang , Department of Biostatistics, JHU
[abstract for short course I, II & III] In recent years there has been a renewed interest in diagnostic tests, especially when biomarkers are involved. New technologies produce thousands of biomarkers that are potentially useful for diagnosis or prognosis. Rigorous evaluation of diagnostic tests and biomarkers is a high priority in research. The first unit of this short course will start with a brief review of the basics in diagnostic test studies, a discussion on different measures for assessing test accuracy and a cautionary note against some misusage. The course will then focus on the effect of covariates in diagnostic testing studies, reveal its difference compared to epidemiological studies, and summarize methodologies for covariate adjustment in diagnostic studies. The second unit of this short course will focus on risk prediction, including discussions on aspects for assessing prediction accuracy, which are different from those for assessing diagnostic accuracy, on marker combination for improve predictive performance and on assessing the predictive improvement – the incremental values. The third part of the short course will extend disease outcome from binary variable to time-to-event variable and focus on time-dependent ROC analysis. The role of Neyman-Pearson Lemma is discussed for identification of optimal composite marker. Cumulative/incident sensitivity and dynamic specificity are introduced when failure time is involved in determination of disease status. Parametric, nonparametric and semiparametric approaches are studied subject to different models and different levels of model assumptions.