|
1:00 pm Registration
1:30 pm Invited Speaker: Andrea Rotnitzky
Department of Biostatistics
Harvard School of Public Health
Double-robust estimation for correcting non-ignorable verification bias
A frequently encountered problem in the evaluation of newly available diagnostic tests is that the data available to assess their discriminatory ability is obtained in observational studies. In such studies often not all patients undergo disease verification because the verification test is expensive, invasive or both. Furthermore, the decision to send patients to verification often depends on the new test and on other predictors of true disease status and is not under the control of the investigator. In such case, usual estimators of indicators of the marker ability such as the area under the receiver operating characteristic curve based on verified patients only are biased. In this talk we will describe methods that adjust for selection to verification that may depend on measured patient covariates and diagnostic test results and additionally adjust for an assumed degree of residual selection bias. Such estimators can then be used in a sensitivity analysis to examine how the estimates of marker discriminatory ability change when different plausible degrees of residual association are assumed. As with other missing data problems, due to the curse of dimensionality, a model for disease or a model for selection is needed in order to obtain well behaved estimators when the marker and/or the measured covariates are continuous. We describe a doubly robust estimator that has the attractive feature of being consistent and asymptotically normal if either the disease or the selection model (but not necessarily both) are correct.
2:45 pm Invited Speaker: Betsy Hill
Department of Biostatistics, Bioinformatics and Epidemiology
Medical University of South Carolina
Latent class inter-rater agreement models
An important outcome in oral health research is the pocket probing depth (PPD), which helps in determining the presence and extent of periodontal disease. PPD is recorded as the greatest integer in the range 0-10 mm less than or equal to the observed depth on a manual probe. In a typical periodontal study, multiple examiners record PPD at multiple sites in the mouth of each subject, inducing correlation among measurements from the same patient and from the same examiner. It is critical that the examiners are calibrated for agreement in PPD results among themselves and with an experienced research examiner (the gold standard). We propose a latent class model for PPD measurements that accommodates clustering at both the subject and examiner level, and that postulates two classes of subjects, those for whom examiners' agreement with the gold standard is relatively "easy," and those for whom agreement is more difficult. Estimation and inference is performed within the Bayesian paradigm. We use our model to analyze data from a recent calibration study, illustrating how assessments of inter-rater agreement and agreement with the gold standard derive naturally from the model.
4:00pm Student Paper Competition:
- Vanessa Brown, University of South Carolina
- Lisa Kaltenbach, Clemson University
Do low-quality embryos alter pregnancy rate in an Assisted Reproductive Technology program?
- Wei Lin, Clemson University
Error variance estimation and test for the linear single-index models
- Meng Zhao, Clemson University
Minimax estimation of linear functionals under squared error loss
5:00 pm Election of Officers
| 