To the Editor:
We read with great interest the article of Cannesson et al. 1regarding the accuracy of pulse pressure variation monitoring to predict fluid responsiveness. We applaud the introduction of “misclassification cost” as a novel approach to evaluate the clinical utility of a widely advocated monitoring technique. However, we feel that the authors may have overlooked an important factor in their analysis. It has been clearly shown that the prevalence of a disease significantly impacts the predictive value as well as the calculated costs of a diagnostic test in a specific population.2
From the description in the Materials and Methods section and the illustrations in figures 2 and 4, it appears that Cannesson et al. defined their “explicit cost” for a given Cost Ratio [R = Cost False Positives (FP)/Cost False Negatives (FN)] as follows:
This formula can be rewritten as:
.2
The formula above indicates that there are three primary determinants of explicit cost including the cost ratio used, the discriminatory power of pulse pressure variation, and the prevalence of the responders. The same applies for the determination of an optimal threshold. To our opinion, there are two important implications that should be taken into account when interpreting the results of Canneson et al. :
Using a bootstrap method to determine the confidence intervals of an optimal threshold will cause resampling of 1,000 populations with varying prevalence, as well as different optimal thresholds. The incorporation of a statistical variance to account for uncertainty of prevalence (rather than to use the measured prevalence) falsely elevates the confidence intervals for optimal threshold in all three “misclassification cost” scenarios.
Using the Youden index to determine the threshold when R = 1 is incorrect. Smits3recently showed that the Youden index implicitly changes its cost ratio in function of prevalence of the studied population.
These considerations may not completely invalidate the conclusion of Cannesson et al. but at least question the accuracy of the data. We believe that the confidence intervals for accuracy of pulse pressure variation are being overestimated and the validity of the technique underrated because the authors did not control for the prevalence of responders in their study population(s).