Bivariate Conditional ROC Curves via Distributional Regression

Ó. Lado-Baleato, S. Pérez-Fernández

The diagnosis of many diseases relies on the joint evaluation of multiple continuous biomarkers that are often correlated. Traditional ROC analysis assesses each marker separately, ignoring this dependence and the influence of clinical covariates such as age on their joint distribution. This limitation may lead to suboptimal patient classification and misdiagnosis in specific subgroups. We propose a method to estimate bivariate conditional ROC curves using multivariate distributional regression. The model captures covariate effects on marginal distributions and dependence structures, allowing nonlinear relationships. Covariate-specific ROC curves are derived from conditional likelihood ratio. It yields a more accurate assessment of diagnostic performance by accounting for markers correlation and covariate effects. The method is evaluated via simulations under multiple scenarios and illustrated in a diabetes study where age influences two glycemic biomarkers across two patient groups.

Palabras clave: Bivariate ROC curves, bivariate distributional regression, multiple biomarkers.

Bioestadística II
4 de septiembre de 2026  15:30
Aula 24

I. Galván Femenía, J. Saperas Riera, J. Palarea Albaladejo

P. Jurado Bascón, J. A. Villatoro García, P. Carmona Sáez

B. Monteiro, A. H. Tavares, T. Gregório

B. Piñeiro Lamas, A. López-Cheda, R. Cao