| Sign In to gain access to subscriptions and/or personal tools. |
Value of Predictive ValuesCenter for Clinical Research and Health Promotion, University of Puerto Rico, School of Dental Medicine The author replies: In our publication (Pitiphat et al., 2008), we used a surrogate exposure with a combined sensitivity and specificity of 117 and combined predictive values of 157. Janket et al. objected, stating that a measure with low sensitivity is unacceptable (regardless of specificity and predictive values). They state that using predictive values to validate accuracy is misleading, since predictive values do not measure accuracy. Although there are some similarities in assessing accuracy of measures for different applications, it should be noted that the article they cited in support (Zweig and Campbell, 1993) discusses accuracy for clinical diagnostic tests and not epidemiologic studies. Moreover, that article (Zweig and Campbell, 1993) did not exclude predictive values from accuracy measures. Predictive values can be calculated from a combination of sensitivity, specificity, and prevalence. Our epidemiologic study (Pitiphat et al., 2008) quantified the misclassification in our measure, and calculated the hypothetically stronger relative risk with a standard measure. The relation between the observed and hypothetical relative risk (for the standard) can be computed using a combination of sensitivity and specificity or of positive and negative predictive values (Flegal et al., 1986). The formulas (Flegal et al., 1986) show that it is the combination of sensitivity and specificity or of the predictive values that is important for assessing accuracy or degree of bias in estimates of association in etiologic studies. We acknowledged that the threshold of 120 (Blicher et al., 2005) for the combination proposed in our article was somewhat arbitrary, and one could propose alternative approaches or thresholds. However, the logic of combining sensitivity and specificity or the predictive values to assess accuracy is well-substantiated (Flegal et al., 1986; Zweig and Campbell, 1993). The ROC curve that is widely accepted also combines measures of sensitivity and specificity (Zweig and Campbell, 1993). REFERENCES
Journal of Dental Research, Vol. 87, No. 10,
e1 (2008)
|
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
 |
|
|
 |
|
Sign In to gain access to subscriptions and/or personal tools.
