Lung cancer is the leading cause of cancer related death in the United States. Early detection of patients with lung cancer by imaging screening tests, such as chest X-rays, CT or MRI, among the high risk population is of great interest. In large validation studies for the utility of imaging screening tests, definitive lung cancer diagnosis procedures, such as biopsy, are too invasive and expensive to be undertaken on all study subjects. It is often more ethical or cost effective to ascertain the true disease status using targeted sampling schemes. Specifically, one may oversample subjects with high screening score and undersample or exclude subjects with low screening score. However, this kind of targeted sampling scheme would introduce verification bias for assessing the accuracy of screening tests. To fully accommodate the sampling schemes, we develop semi parametric methods to estimate the parameters of the covariate-specific ROC regression model for continuous screening tests.
The specific aims are (1) to develop a semi parametric empirical likelihood method to estimate the covariate-specific ROC curve of continuous tests when the selection probability of observing the true disease status is unknown or inestimable; (2) to develop a semi parametric empirical likelihood method and an augmented inverse probability weighting method (AIPWCC) to estimate the covariate-specific ROC curves when the selection probability for observing the true disease status is known or estimable; and (3) to develop open source software to implement these methods. The project involves both theoretical and empirical work, drawing on collaborative opportunities. Though lung cancer screening through imaging modality is used as a motivating example, with a growing interest in rigorous validation of screening tests for cancer and other diseases, the proposed methods may find broad usage. In cancer screening studies, a fraction of subjects is often selected from the study cohort to ascertain the true disease condition due to ethical or economical reasons. When subject selection is dependent on screening test scores, standard statistical methods will yield incorrect assessment on the accuracy of the screening tests. This is known as verification bias in the literature of diagnostic medicine. This study will develop efficient and consistent methods to estimate the covariate-specific ROC curve of continuous screening tests. It involves both theoretical and empirical work, drawing on existing collaborative opportunities. With a growing interest in rigorous validation of screening tests for cancer and other diseases, the proposed methods may end broad usage. ? ? ?
| Wang, Xiaofei; Ma, Junling; George, Stephen L (2013) ROC curve estimation under test-result-dependent sampling. Biostatistics 14:160-72 |
| Wang, Xiaofei; Ma, Junling; George, Stephen et al. (2012) Estimation of AUC or Partial AUC under Test-Result-Dependent Sampling. Stat Biopharm Res 4:313-323 |
| Wang, Xiaofei; Zhou, Haibo (2010) Design and inference for cancer biomarker study with an outcome and auxiliary-dependent subsampling. Biometrics 66:502-11 |
| Wang, Xiaofei; Wu, Yougui; Zhou, Haibo (2009) Outcome- and auxiliary-dependent subsampling and its statistical inference. J Biopharm Stat 19:1132-50 |
