¹Fundación Caubet-Cimera Illes Balears, Mallorca, España. Programa de epidemiología e investigación clínica. Email: martinez@caubet-cimera.es

Se aborda el problema de comparar el poder de clasificación de métodos diferentes a partir de la curva ROC. Por un lado, se propone un método de comparación basado en la medida del supremo y, por otro, una solución al problema de comparar más de dos pruebas diagnósticas a través del área bajo la curva ROC (AUC) a partir de sus propiedades asintóticas. También se comprueba la validez de los estimadores propuestos para muestras pequeñas a partir del método bootstrap. Finalmente, se aplican los métodos propuestos en la predicción de diagnósticos sépticos (infecciosos) en pacientes admitidos en la Unidad de Cuidados Intensivos Pediátricos (UCIP) del Hospital Central de Asturias.

Palabras clave: curvas ROC, sensibilidad, especificidad, AUC, bootstrap.

We study the problem of comparing the power of classification of different methods from the ROC Curve. On one hand, we propose a method based on the supremum measure and, on the other hand, we study the problem of comparing two or more ROC curves from the asymptotic properties of area under ROC curves (AUC). We study the performance of proposed estimators to small samples problems with Bootstrap method and we apply them to differentiate two classes of patients of the Pediatric Intensive Care Unit (PICU) of the Hospital Central de Asturias.

Key words: ROC curves, Sensitivity, Specificity, AUC, Bootstrap.

Texto completo disponible en PDF

Referencias

1. Bamber, D. (1975), `The Area Above the Ordinal Dominance Graph and the Area Below the Receiver Characteristic Graph´, Journal of Mathematical Psychology 12, 387-415.

2. Cai, T. & Pepe, M. S. (2002), `Semiparametric Receiver Operating Characteristic Analysis to Evaluate Biomarkers for Disease´, Journal of the American Statistical Association 97(460), 1099-1107.

3. Dorfman, D. D. & Alf, E. (1969), `Maximum Likelihood Estimation of Parameters of Signal Detection Theory and Determination of Confidence Interval-Rating Method Data´, Journal of Mathatematical Psychology 6, 487-496.

4. Hall, P. G. & Hyndman, R. J. (2003), `An Improved Method for Bandwidth Selection When Estimating ROC Curves´, Statistics and Probability Letters 64(2), 181-189.

5. Hall, P. G., Hyndman, R. J. & Fan, Y. (2004), `Nonparametric Confidence Interval for Receiver Operating Characteristic Curve´, Biometrika 91(3), 743-750.

6. Hsieh, F. & Turnbull, B. W. (1996), `Nonparametric and Semiparametric Estimation of the Receiver Operating Characteristic Curve´, Annals of Statistics 24(1), 25-40.

7. Lloyd, C. J. (1998), `Using Smoothed Receiver Operating Characteristic Curves to Summarize and Compare Diagnostic System´, Journal of the American Statistical Association 93(444), 1356-1365.

8. Lloyd, C. & Yong, Z. (1999), `Kernel Estimators of the ROC Curve are Better than Empirical´, Statistics Probability Letters 44(3), 221-228.

9. Nadaraya, E. A. (1964), `Some New Estimates for Distributions Functions´, Theory Probability Application 9, 497-500.

10. Rey, C., Los Arcos, M., Concha, A., Medina, A., Prieto, S., Martínez-Camblor, P. & Prieto, B. (2007), `Procalcitonin and C-reactive Protein as Markers of Systemic Inflammatory Response Syndrome Severity in Critically ill Children´, Intensive Care Medicine 33(3), 477-484.

11. Silverman, B. W. (1986), Density Estimation for Statistics and Data Analysis, Chapman and Hall, London, England.

12. Stute, W. (1982), `A Law of the Logarithn for Kernel Density Estimatiors´, Annals of Probability 10(2), 414-422.

13. Wand, M. P. & Jones, M. C. (1995), Kernel Smoothing, Chapman and Hall.

14. Zhou, X. & Castelluccio, P. (2003), `Nonparametric Analysis for the ROC Areas of two Diagnostic Test in the Presence of Nonignorable Verification Bias´, Journal of Statistics Planning and Inference 115, 193-213.

Este artículo se puede citar en LaTeX utilizando la siguiente referencia bibliográfica de BibTeX: