Our session includes novel methodological work in two exciting research areas in diagnostic testing, namely covariate-adjusted ROC surfaces for three-class classification problems and use of the length of the ROC curve and its properties in two-class classification problems.
Accurate diagnosis of disease is of great importance in clinical practice and medical research. The receiver operating characteristic (ROC) surface is a popular tool for evaluating the discriminatory ability of continuous diagnostic test outcomes when there are three ordered disease classes. The first speaker will present approaches for incorporating covariates in ROC surface analysis in order to potentially enhance information gathered from the diagnostic test as its discriminatory ability may depend on these, developing on her previous research (Rodriguez-Alvarez, Inacio, arXiv:2003.13111, 2020). A Bayesian distributional regression approach for covariate-specific ROC surface estimation will be presented where, in the model specification, the covariate-specific ROC surface is indirectly modelled using probabilistic distributional models capturing location, scale, shape, and possibly other aspects of the diagnostic test's distribution in each of the three groups. Covariate effects are modelled flexibly through penalised splines. Changes with age and gender in the capacity of several Alzheimer's disease biomarkers for discriminating between subjects showing no clinical symptoms, subjects with mild disease impairment, and subjects suffering from dementia, will be shown.
The length of the ROC curve has recently been proposed as an alternative index for assessing the diagnostic performance of markers by the second speaker (Franco-Pereira, Nakas, Pardo, AStA Adv Stat Anal 104, 625–647, 2020). Two estimation procedures for this summary measure based on (1) normal assumptions; (2) transformations to normality will be presented. These are compared in terms of bias and root mean square error in an extensive simulation study. Testing procedures for the assessment of a single marker and for the comparison of biomarkers will be shown. Furthermore, cases in which the length of the ROC curve outperforms the AUC and the Youden index are illustrated. Finally, an illustration through a real-world application will be provided.
The third speaker further expounds on the use of the length of the ROC curve. His recent publication revolves around the theoretical foundations and utility of the length index (Bantis et al, Stat Med, 2021, https://doi.org/10.1002/sim.8869). During the early stage of biomarker discovery, high throughput technologies allow for simultaneous input of thousands of biomarkers that attempt to discriminate between healthy and diseased subjects. In such cases, proper ranking of biomarkers is highly important. Common measures, such as the area under the receiver operating characteristic (ROC) curve (AUC), as well as affordable sensitivity and specificity levels, are often taken into consideration. Strictly speaking, such measures are appropriate under a stochastic ordering assumption, which implies that higher (or lower) measurements are more indicative for the disease. Such an assumption is not always plausible and may lead to rejection of extremely useful biomarkers at this early discovery stage. The length of a smooth ROC curve as a measure for biomarker ranking is not subject to a single directionality. The length corresponds to a φ divergence, is identical to the corresponding length of the optimal (likelihood ratio) ROC curve, and is an appropriate measure for ranking biomarkers. A complete framework for the evaluation of a biomarker in terms of sensitivity and specificity through a proposed ROC analogue for use in improper settings will be considered. Applications on real data that relate to pancreatic and
esophageal cancer will be shown.
Overall, recent advances in ROC analysis pertaining to the development of an active area of research with an extremely large number of applications will be presented. ROC related topics are useful in a very wide range of applied research problems and contribute to the interdisciplinarity and usefulness of biometry in the advancement of science overall.
We have an international set of speakers from three countries and two continents (UK, Spain, USA). The discussant and the organizer further represent another three countries from the Eastern Mediterranean (Israel, Greece) and Central Europe (Switzerland), while the panel is well balanced both with respect to gender and academic seniority.
The whole panel is comprised of researchers renowned in this field of research with a large number of methodological and applied contributions alike. A nice balance regarding the experience of the panel members also exists.
Proposed Speakers & Discussant
, School of Mathematics, University of Edinburgh (UK)
Alba Maria Franco Pereira
- Distributional ROC surface regression
, Department of Statistics and Operational Research, Faculty of Mathematics,
Complutense University of Madrid (Spain)
- The length of the ROC curve for assessing diagnostic markers
Department of Biostatistics and Data Science, University of Kansas Medical Center (USA)
Discussant: Benjamin Reiser,
- The length of the ROC curve and the two cutoff Youden index within a board framework for discovery, evaluation, and cutoff estimation in biomarker studies involving improper ROCs
Department of Statistics, University of Haifa (Israel)