More information about text formats
To the editor:
Olaciregui et al. reported an interesting article and concluded that
the diagnostic value of procalcitonin (PCT) is greater than C reactive
protein (CRP) in predicting infants with more invasive bacterial diseases
(sepsis, bacteraemia). However, due to the following reasons, the
conclusion should be more conservative.
First, the authors claimed that the area under curve (AUC) is greater...
First, the authors claimed that the area under curve (AUC) is greater
when comparing PCT using a cut-off point of 0.5 ng/ml and CRP using a cut-
off point of 30 mg/l (Table 2). I think it is a misuse of receiver
operating characteristic (ROC) curves. Because ROC curves are plotted by
sensitivity and 1-specificity using different cut-off points, AUC will not
change when shifting cut-off points.(1) Obviously, it is not sufficient to
support any specific cut-off point by comparing AUC of ROC curves.
Second, the authors strengthened the conclusion by larger odds ratio
of PCT in the multivariate logistic regression model in subgroup analysis
(Table 3 and Table 4). However, because CRP and PCT are both good
predictors of serious bacterial infections, they must be highly correlated
with each other. It brings a very serious problem of collinearity.(2)
Thus, I am afraid that the model is an unstable model and the result might
greatly change when adding some more cases. It should be very conservative
when explaining the results of this model.
Finally, because CRP is almost routinely performed for febrile
infants under 3 months of age in clinical practice, it is meaningless to
argue PCT is better than CRP or not. It will be more interesting to see
how much the increment in AUC is between CRP alone and PCT plus CRP. Some
new methods of measuring quantify the improvement such as net
reclassification improvement and integrated discrimination improvement can
be applied for this purpose.(3)
1.Bewick V, Cheek L, Ball J. Statistics review 13: receiver operating
characteristic curves. Crit Care 2004;8(6):508-12.
2.Nathanson BH, Higgins TL. An introduction to statistical methods
used in binary outcome modeling. Semin Cardiothorac Vasc Anesth
3.Pencina MJ, D'Agostino RB, Sr., D'Agostino RB, Jr., Vasan RS.
Evaluating the added predictive ability of a new marker: from area under
the ROC curve to reclassification and beyond. Stat Med 2008;27(2):157-72;