PRECISIÓN DISCRIMINATORIA DE PRUEBAS SEROLÓGICAS PARA Leishmania APLICANDO CURVAS COR Y METODOLOGÍA ESTÁNDAR BAJO DIFERENTES CONDICIONES DE VARIANZAS ::: DISCRIMINATORY ACCURACY OF SEROLOGICAL TESTS FOR Leishmania APPLYING COR CURVES AND STANDARD METHODOLOGY UNDER DIFFERENT VARIANCES CONDITIONS

Las pruebas serologicas permiten identificar agentes infecciosos en muestras biologicas. Su eficiencia es evaluada a traves de la precision discriminatoria, dependiendo esta del umbral de decision. La metodologia estandar y la curva caracteristica operativa del receptor (COR) permiten obtener tal umbral, y ambas metodologias se emplean indistintamente en la practica. Se realizo un estudio comparativo de la precision discriminatoria de ambas metodologias. Se utilizo un conjunto de valores de anticuerpos contra Leishmania de pacientes previamente clasificados como enfermos o sanos, y se realizaron simulaciones computarizadas en condiciones homocedasticas y heterocedasticas. A estos datos se les aplico la metodologia estandar y la curva COR. Se calcularon los parametros sensibilidad, especificidad, %Sensibilidad = 100%, %Especificidad = 100% y %Decision perfecta. Los resultados muestran que la precision discriminatoria de la metodologia estandar favorece la especificidad, mientras que la curva COR muestra un equilibrio entre sensibilidad y especificidad, las cuales aumentan a medida que la media de la poblacion de pacientes enfermos se aleja de la media de la poblacion sana, presentando ademas una mayor proporcion de decision perfecta. Los umbrales de decision fueron afectados por la heterocedasticidad. La curva COR mostro una mejor precision discriminatoria y se recomienda su uso para el calculo de umbrales de decision en pruebas serologicas. Palabras clave : Leishmaniasis, punto de corte, simulacion computarizada, sensibilidad, especificidad, homocedasticidad, heterocedasticidad. ABSTRACT Serological tests can identify infectious agents in biological samples. Their efficiency is evaluated through the discriminatory accuracy, which depends on a decision threshold. The standard methodology and the receiver operating characteristic curve (ROC) allow to obtain such threshold, and both methodologies are used indistinctly in practice. A comparative study of the discriminatory accuracy of both methods was performed. A set of values of antibodies to Leishmania in patients previously classified as sick or healthy was used, and computer simulations under homoscedastic and heteroscedastic conditions were made. On these data the standard methodology and the ROC curve were applied. The parameters sensitivity, specificity, %Sensitivity = 100%, %Specificity = 100% and %Perfect decision were calculated. Results showed that the discriminatory accuracy of the standard methodology favors specificity, while the ROC curve shows a balance between sensitivity and specificity, which increase as the average population of sick patients moves away from the average of the healthy population, in addition to present a greater proportion of perfect decision. The decision thresholds were affected by heteroscedasticity. The ROC curve showed better discriminatory precision and its use for calculating decision thresholds in serological testing is recommended. Key words : Leishmaniasis, cutoff, computer simulation, sensitivity, specificity, homocedasticity, heteroskedasticity.