To the Editor:
We read with great interest Biais et al.’s study1 investigating the mini-fluid challenge during neurosurgery. In line with previous mini-fluid challenge research,2–4 the mini-fluid challenge predicted fluid responsiveness with compelling accuracy.1 Still, we feel there are some very important methodologic aspects to highlight for the existing mini-fluid challenge results: predictor and outcome variables being calculated from the same baseline.
Except for Guinot et al.’s study,4 all existing studies1–3 calculated their predictor and outcome variables as follows: the predictor variable is based on a change from baseline to after the mini-fluid challenge —in the present study, a ΔSVI100 variable was calculated, see the study’s figure 1.1 The outcome variable (defining the fluid response) has been calculated as a change also from baseline (before the mini-fluid challenge) to after the full fluid challenge—in the present study, a ΔSVI250 variable was calculated. Now, ΔSVI100 and ΔSVI250 are mathematically coupled via the baseline value (ΔSVI100 = [SVafterMFC, 100ml– SVbaseline]/SVbaseline and ΔSVI250 = [SVafterFC, 250 ml– SVbaseline]/SVbaseline). Unfortunately, this means that the high predictive power of the mini-fluid challenge approach can be explained by not only one but by three reasons: (1) A true predictive power of the mini-fluid challenge (which we would all love to believe); (2) A statistical phenomenon (see below), or (3) A combination of 1 and 2 (most likely the case). To understand the statistical phenomenon, let’s imagine the case where the mini-fluid challenge itself induces a significant increase in stroke volume, say ΔSVI100 is 10%. If stroke volume stays unaltered when infusing the remaining 150 ml, this would give rise to a ΔSVI250 of still 10%, defining a positive fluid response—even though stroke volume has not changed at all with the second infusion. In other words, ΔSVI100 and ΔSVI250 are likely to agree (even when they don’t) simply because they have been calculated based on the same baseline value, whose random measurement error and/or physiologic variation (which is present in any measurement) is carried over in calculations of both ΔSVI100 and ΔSVI250. The problem is even demonstrated in figure 2,1 where some responders (as defined by ΔSVI250) experience status quo or even reductions in stroke volume during the last part of the 250 ml infusion, i.e.ΔSVI100 is higher than ΔSVI250. To take this down to a clinical everyday level, consider our standard fluid challenge approach in most goal-directed therapy applications: We usually administer a first fluid challenge of 250 ml and evaluate the stroke volume response: Let’s say we encounter a stroke volume increase of 20%. Afterward, we give a second fluid challenge (as merited by our goal-directed therapy protocols) and stroke volume stays the same—we have reached the Frank-Starling curve plateau, and we now consider our patient unresponsive to fluids. According to the mini-fluid challenge design described earlier, however, this second fluid challenge would be considered a positive fluid response, because stroke volume is still 20% higher than the baseline value before the first fluid challenge. This obviously makes no sense. Note that the only difference in this example is that we replaced 100 + 150 ml infusions with 250 + 250 ml infusions, and it should be clear that in future mini-fluid challenge studies, the outcome/response variables must be independent of the predictor variables. It could be suggested to use the stroke volume value after the mini-fluid challenge as a new baseline for the subsequent fluid challenge, but that approach also creates a mathematical coupling, which theoretically reduces the predictive power of the mini-fluid challenge because the outcome, ΔSVI250, is then defined as ΔSVI250 = (SVafterFC, 250ml– SVafterMFC, 100ml)/SVafterMFC, 100ml. The SVafterMFC, 100ml measurement would then be part of both predictor (ΔSVI100, described above) and outcome calculations. Because SVafterMFC, 100ml is a positive term in the ΔSVI100 calculation and a negative term (being subtracted) in the ΔSVI250 calculation, the random variation in SVafterMFC, 100ml would drag ΔSVI100 and ΔSVI250 in “opposite” directions and thus make ΔSVI100 and ΔSVI250 less likely to agree (as opposed to the design with a common baseline value). In that sense, we strongly encourage following the design suggested by Guinot et al.,4 who had a new baseline measurement 5 min after the mini-fluid challenge, or at least keeping baseline variables “separated” as Mallat et al.3 did (i.e. by measuring pulse pressure variation changes after the mini-fluid challenge and relating that to stroke volume changes after the fluid challenge—changes in two different variables). Otherwise, the study design itself may artificially boost the true predictive power of the mini-fluid challenge and result in a self-fulfilling prophecy, and we would have no means to evaluate how big the boost had been. So, even though it might be argued that we are not far from recommending that clinicians start getting familiar with and gathering experience with this simple, low-dose-fluid approach, we believe that the next step is to settle optimal methodology for this otherwise compelling approach.
The authors declare no competing interests.