A note on modeling placement values in the analysis of receiver operating characteristic curves

Abstract

Recent advances in receiver operating characteristic (ROC) curve analyses advocate modeling of placement value (PV), a quantity that measures the position of diseased test scores relative to the healthy population. Compared to traditional approaches, this PV-based alternative works directly with ROC curves and is attractive when assessing covariate effects on, or incorporating a priori constraints of, ROC curves. Several distributions can be used to model the PV, yet little guidelines exist in the literature on which to use. Through extensive simulation studies, we investigate several parametric models for PV when data are generated from a variety of mechanisms. We discuss the pros and cons of each of these models and illustrate their applications with data from a study of prenatal ultrasound examinations and large-for-gestational age birth.

Keywords:

• AUC
• ROC
• diagnostic accuracy
• large for gestational age
• beta regression

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This research was supported by the Intramural Research Program of Eunice Kennedy Shriver National Institute of Child Health and Human Development.

Notes on contributors

Zhen Chen

Zhen Chen, Ph.D., is a Senior Investigator in the Biostatistics and Bioinformatics Branch, Division of Intramural Population Health Research, Eunice Kennedy Shriver National Institute of Child Health and Human Development, National Institutes of Health.

Soutik Ghosal

Soutik Ghosal is a postdoctoral fellow in the Biostatistics and Bioinformatics Branch, Division of Intramural Population Health Research, Eunice Kennedy Shriver National Institute of Child Health and Human Development, National Institutes of Health.