Receiver operating characteristic (roc) curve and covariates: new contributions in statistical inference, software development and biomedical applications
- Autores: María José Rodríguez Álvarez
- Directores de la Tesis: Javier Roca Pardiñas (dir. tes.), Carmen María Cadarso Suárez (dir. tes.)
- Lectura: En la Universidade de Santiago de Compostela ( España ) en 2011
- Idioma: inglés
- Tribunal Calificador de la Tesis: José Manuel Prada Sánchez (presid.), Francisco Gude Sampedro (secret.), Luís Meira Machado (voc.), Thomas Kneib (voc.), Alberto Rodríguez Casal (voc.)
- Materias:
- Texto completo no disponible (Saber más ...)
- Resumen
The Receiver Operating Characteristic (ROC) curve is the most widely used tool for evaluating the discrimination capacity of continuous marker between two alternative states. In certain circumstances, however, the discrimination capacity of a marker can be affected by covariates. So far, different methodologies have been proposed in the literature to evaluate the effect of covariates on the ROC curve. This thesis is focused on those within the general regression framework, namely the direct and the induced methodologies. The main objectives of this thesis are Given the gaps in the literature on this subject, we ha conducted a thorough and in-depth simulation study to compare statistically the different existing methodologies.
We have extended the induced methodology to the nonparametric framework, by allowing for a nonparametric effect of a continuous covariate in the discrimination capacity of a continuous marker. Also, we have developed hypotheses tests about the effect of the covariate on the ROC curve.
We have extended the direct methodology to the nonparametric framework. As for the induced methodology, we have also developed hypotheses tests about the effect of continuous covariates on the ROC curve, and the presence of interaction terms between continuous and categorical covariates.
