I was wondering if it would be possible to compute a mean ROC curve, with corresponding mean AUC, for multiple plotted ROC curves.
I have a dataset of 379 observations, each of which has been assigned a score rainging from 1 to 5 (categorical variable) by 4 different readers, by reference to their corresponding real clinical diagnosis ("Reference Variable/Reference Standard: 0=neg, 1=pos). I have plotted a ROC curve with its corresponding AUC for each reader, obtaining sensitivity and specificity for each cut-point (corresponding score), just as shown below for a single reader:
Code:
Detailed report of sensitivity and specificity ------------------------------------------------------------------------------ Correctly Cutpoint Sensitivity Specificity Classified LR+ LR- ------------------------------------------------------------------------------ ( >= 1 ) 100.00% 0.00% 30.61% 1.0000 ( >= 2 ) 75.86% 54.75% 61.21% 1.6766 0.4409 ( >= 3 ) 70.69% 68.44% 69.13% 2.2399 0.4283 ( >= 4 ) 66.38% 80.61% 76.25% 3.4231 0.4171 ( >= 5 ) 54.31% 86.69% 76.78% 4.0810 0.5270 ( > 5 ) 0.00% 100.00% 69.39% 1.0000 ------------------------------------------------------------------------------ ROC -Asymptotic Normal-- Obs Area Std. Err. [95% Conf. Interval] ------------------------------------------------------------ 379 0.7380 0.0284 0.68229 0.79368
I have tried the .cvauroc command to obtain a cross-validated mean AUC in this way:
Code:
.cvauroc FINALDIAGNOSIS READER1 READER2 READER3 READER4
Thank you in advance for your attention.
Kind regards.
0 Response to Computing Average/Mean ROC Curve and AUC among different observers
Post a Comment