Major research activities are underway to develop and improve technologies for disease screening, for diagnosing disease, for predicting its course and for monitoring its treatment. New biomarkers, imaging modalities, clinical prediction scores, staging systems and genomic/proteomic profiles are amongst candidate prediction technologies for which we use the generic terms 'predictive markers' and 'medical tests' here. Before being implemented in practice, however, the diagnostic and prognostic capacities of tests (or markers) must be evaluated rigorously, particularly in view of potentially grave or costly consequences associated with errors. ? ? This grant will continue to develop statistical methods for the evaluation of studies that seek to investigate predictive markers. The methodology we propose to develop in Aims 1-4 will examine the extent to which a marker reflects the outcome variable of interest (e.g., disease, disease outcome or treatment failure) as is of primary interest in the early stages of marker development. It will allow identification of conditions/populations that maximize the discriminatory capacity of a marker (Aim 1) and of optimal algorithms for combining multiple predictive markers (Aim 4).
Aims 2 and 3 seek to extend statistical methods for characterizing discrimination to accommodate data on outcome variables that are time dependent and censored (Aim 2) or that are subject to verification biased sampling (Aim 3), which is common in practice. Finally, in Aim 5 we will develop statistical methods for characterizing and comparing the predictive values of markers when applied prospectively in defined populations, as is typically of interest in the later stages of marker development.
Showing the most recent 10 out of 74 publications
