The "grey zone" or how to avoid the binary constraint of decision-making.

Abstract

In this issue of the Journal, Vos et al. report on the accuracy of pulse pressure variation (PPV), stroke volume variation (SVV), and stroke volume index (SVI) to predict fluid responsiveness at multiple thresholds. In their study, they show that the threshold used to define fluid responsiveness impacts the predictive value of these indices and the range of their zones for decision-making (i.e., the ‘‘grey zone’’)—the lower the threshold for the definition of fluid responsiveness, the lower the predictive value and the wider the grey zone. Although these results are not completely surprising (i.e., how one defines an outcome generally impacts the accuracy of the tools used to predict that outcome) or entirely new (i.e., the concept of using a PPV grey zone though perhaps not these authors’ precise definition for understanding fluid responsiveness had already been reported), they nevertheless remind us of the importance of using the grey zone methodology when approaching any diagnostic tool (such as PPV and SVV). Accordingly, this editorial serves as a reminder of the relevance, impact, and global meaning of the grey zone for diagnostic tools. The predictive value of continuous diagnostic indices such as PPV are best evaluated using the receiver operating characteristic (ROC) curve approach. Very briefly, this approach determines an optimal threshold that provides the highest combination of sensitivity and specificity for a given diagnostic tool. In addition, the accuracy of a test depends on the ability of the test to separate the group being tested into those with and those without the disease, and it is quantified by the area under the ROC curve which ranges from 0 to 1. An area of 1 represents an ideal test, while an area of 0.5 represents a worthless test (i.e., same predictive value as flipping a coin). While this approach has been used for years to assess the accuracy of diagnostic tools (such as the ability of PPV to separate responders from non-responders to fluid administration), its main limitation is that it transforms the biological nature of a continuous variable into an artificially dichotomous (i.e., ‘‘black or white’’) statistical index that does not always accurately reflect the decision-making process applied to clinical management. Indeed, the very reason we originally proposed using the grey zone approach to evaluate PPV was to avoid this type of binary constraint. The grey zone technique proposes two numerical cutoffs that constitute its borders. The first cutoff is used to exclude the diagnosis (e.g., a 9% PPV where fluid responsiveness is not present) with near certainty (i.e., privilege sensitivity and negative predictive value), whereas the second cutoff is chosen to include the diagnosis (e.g., a 13% PPV where fluid responsiveness is present) with similar near certainty (i.e., privilege specificity and positive predictive value). Intermediate values representing the grey zone correspond to a prediction that is too imprecise for a diagnostic decision, referring to what Feinstein called ‘‘the inadequacy of binary models for the clinical reality of three-zone diagnostic decisions’’. Indeed, for PPV, the first study published on the topic found a threshold of 13% to predict fluid responsiveness, an associated sensitivity of 94%, a specificity of 96%, and an area (standard deviation) under the ROC curve of 0.98 (0.03). Most subsequently published studies found similar results. Nevertheless, a few years later when we studied the predictive value of PPV using a grey zone approach, we found a similar area under the curve, but the grey zone ranged from 9-13% and M. Cannesson, MD, PhD (&) Department of Anesthesiology & Perioperative Care, University of California, Irvine, CA, USA e-mail: mcanness@uci.edu