Predicting fluid responsiveness using esophagus Doppler monitoring and pulse oximetry derived pleth variability index; retrospective analysis of a hemodynamic study

Fluid therapy during major surgery can be managed by providing repeated bolus infusions until stroke volume no longer increases by ≥ 10%. However, the final bolus in an optimization round increases stroke volume by < 10% and is not necessary. We studied how different cut‐off values for the hemodynamic indications given by esophagus Doppler monitoring, as well as augmentation by pulse oximetry, are associated with a higher or smaller chance that stroke volume increases by ≥ 10% (fluid responsiveness) before fluid is infused.


Editorial Comment
There is always a response when an intravenous fluid bolus is administered. This study examines different thresholds for defining "responsiveness" using a secondary analysis of trial material from major abdominal surgery cases. Associations between results with different stroke volume increase thresholds after the controlled fluid bolus and pre-infusion circulatory conditions, together with other measured circulatory variables including for the pleth variability index, are presented. The findings demonstrated the complexity and potential utility of additional context in interpreting single-stroke volume change measurements after a fluid bolus.

| INTRODUCTION
A recommended means of managing fluid therapy during major surgery is to increase cardiac preload via repeated intravenous infusions of fluid boluses if the cardiac stroke volume (SV) responds by increasing by ≥ 10%. 1,2 One problem with such "optimization rounds" is that with the algorithms used today, the last fluid bolus in each series is never necessary and may contribute to fluid overload and reduced oxygen delivery.
Several direct and indirect methods are used to determine whether a fluid bolus increases SV (i.e., fluid responsiveness).
Esophageal Doppler monitoring (EDM) is one of the best-established of these methods. [3][4][5][6] In EDM, a flow probe is placed in the esophagus to measure blood flow velocity in the descending thoracic aorta, from which SV is derived with a nomogram. Another measurement given by the EDM is flow time corrected (FTc), which is the width of the waveform or, specifically, the duration of the blood flow per heartbeat in milliseconds corrected to a heart rate of 60. However, it remains ambiguous as to whether FTc reflects preload or afterload. [7][8][9] Analysis of perfusion variation via pulse oximetry is an alternative noninvasive approach to assess fluid responsiveness. In this method, the maximum and minimum perfusion amplitudes measured in a finger during the respiratory cycle are compared, and the relative difference expressed as a pleth variability index (PVI) value. 10,11 The view that evaluating multiple parameters instead of a single one may offer benefits when making decisions about whether to administer fluid is growing and is even reflected in clinical guidelines. 12 For this purpose, we have examined how the likelihood of fluid responsiveness changes by considering different cut-offs for the hemodynamic values delivered by both EDM and pulse oximetry-derived PVI before giving a fluid bolus.
The data were derived from major abdominal operations in which both methods were applied. We hypothesized that pre-infusion hemodynamic data delivered by EDM and pulse oximetry combined could refine fluid responsiveness predictions, as compared to each method by itself. This report presents a secondary analysis of the data but with a different focus; the original case study randomized 150 patients to goaldirected fluid therapy (GDFT) optimization protocol based on either EDM or PVI. In the first 75 patients, measurements were made simultaneously with both methods, but only one was available to the anesthetist in charge of each patient. During the second half of the study, the patients were usually monitored with EDM or PVI only. EDM was used in 108 patients and all of them are included in the present analysis.

| METHODS
However, EDM and PVI data were only available in 77 of them.  EDM was performed using a CardioQ apparatus (Deltex Medical) equipped with a DP12 probe. 3 The signal was averaged over 20 cardiac cycles. The measurements were performed by clinicians with long-term experience with the Doppler technique and delivered both SV and FTc. The SV index was derived offline from SV with correction for the body surface area according to the Du Bois formula. 16 PVI values were monitored using a Radical 7 Pulse CO-oximeter (Version SET V7.8.0.1), a re-usable sensor (R2-25r), and a disposable adhesive (R2-25a; Masimo Corp.). The sensor was covered to avoid room light interference and placed on the middle or index finger of either hand. PVI is based on the dynamic change of the pulse oximetry-derived perfusion index during one respiratory cycle. 11 The first fluid bolus was given after the induction of anesthesia.

| Interventions and measurements
EDM was applied just before and 5 min after the bolus was given. The bolus infusions were considered warranted if SV increased by ≥ 10% or PVI decreased by ≥ 10%. Additional fluid boluses were then given until the SV no longer increased by 10% or PVI was <10% or not further decreasing. This is called an optimization round.
Additional optimization rounds were performed later during the surgery if the SV decreased by ≥ 10% or PVI increased to ≥ 10%, as detailed by the original publication. 13 We considered fluid responsiveness to be present if the EDM showed that SV increased ≥ 10%, which is a commonly used cut-off value for EDM based on measurement characteristics of the device. 17 The pre-defined endpoint for our evaluation was to establish hemodynamic cut-offs that most clearly increased or decreased the likelihood of a fluid bolus infusion being warranted in patients undergoing major surgery.

| Statistics
The data are presented as mean and SD. We used linear regression analysis to illustrate the relationships between hemodynamic variables measured just before a bolus infusion and the change in SV induced by the subsequent infusion. We then used one-way analysis of variance to compare demographic and hemodynamic data associated with unwarranted and warranted bolus infusions. The percentage of warranted boluses for an increasing number of optimization rounds was studied by contingency table analysis.
Receiver operating characteristic (ROC) curves expressed the ability of the hemodynamic variables to predict whether a bolus infusion was warranted (i.e., the patient shows fluid responsiveness) before giving the infusion. ROC curves are probability curves in which sensitivity is plotted versus (1-specificity). The calculated area under the curve for this relationship reflected how well ranges of fluid intake could be separated. The prediction given by the ROC curve was statistically significant if the 95% confidence interval did not include 0.5.
The optimal cut-off when dichotomizing a hemodynamic variable to indicate the absence or presence of fluid responsiveness is given by the point situated at the longest distance from the diagonal reference line. p < .05 was accepted as statistically significant. Corrections for multiple comparisons were not made.

| Basic data
Data were available from 108 patients, from which we studied the first bolus infusion of the 5 first optimization rounds in each patient. Later optimization rounds were rare and thought to involve a selection bias.

| Hemodynamic measurements
SV increased by ≥ 10% in 117 of all infusions (44%), which showed that the patient was fluid-responsive. Hence, there was a 44% overall likelihood that these boluses, as indicated by conventional GDFT algorithms, would be warranted. Table 1 shows select demographic variables and the hemodynamic measurements performed just before and after the bolus infusions depending on whether they were warranted or not. SV, FTc, PVI, and ΔSV correlated significantly with fluid-induced changes in SV (Figure 1). The ROC curves suggested that SV, FTc, and PVI had almost the same ability to predict whether a bolus infusion would be warranted, but ΔSV was the strongest predictor of fluid responsiveness ( Table 2).
Optimal cut-offs were difficult to determine for SV and FTc based on our ROC curves shown in Figure 2A, B; the optimal cut-off should be the point at the longest distance from the diagonal reference line.
However, we agreed that the most appropriate ones were on 80 mL for SV and 360 ms for FTc. By contrast, the optimal cut-offs for PVI were more clearly 10% and the ΔSV À8% ( Figure 2C, D). We still varied the cut-offs in our further analysis to explore how combinations of them changed the likelihood that a fluid bolus would be warranted. The reported cut-offs and combinations of variables were the most useful of the ones we tested.

| Decreased likelihood of fluid responsiveness
Certain cut-offs were associated with a lower likelihood of fluid responsiveness than the overall average of 44%.
SV > 80 mL was followed by fluid responsiveness in 36% of the optimizations, 30% of the infusions with FTc > 360 ms, and 38% of the infusions where PVI was <10%. Stroke volume index (SVI) could not distinguish between those who would be responders and nonresponders.
T A B L E 1 Demographic data and hemodynamic indices depending on whether a bolus infusion was warranted or not. Mean (SD). F I G U R E 1 Relationships between hemodynamic parameters measured before fluid bolus infusion and the subsequent fluid-induced stroke volume (SV) response.
The likelihood of fluid responsiveness was 21% when ΔSV > À8% (i.e., had increased or not decreased by as much as 8%).
Raising the SV cut-off point to 100 mL reduced the likelihood of fluid responsiveness to 28% while raising FTc to >390 ms yielded 25% responsiveness. Lowering the PVI limit had negligible influence on its ability to indicate fluid responsiveness.

| Increased likelihood of fluid responsiveness
Hemodynamic cut-offs could also be associated with higher likelihoods of fluid responsiveness than the overall average of 44%.
SV ≤80 mL was followed by fluid responsiveness in 50% of the

| Key findings
The present results show that the likelihood of a patient being fluid-responsive was reduced from the overall likelihood of 44% to 30%-35% by considering SV and FTc values above the cut-offs suggested by ROC curves. The likelihood of fluid responsiveness could be further reduced by 10% by applying even tougher cut-off values. A limited decrease in SV since the previous optimization round had the same negative predictive power as the raised cut-off for FTc (> 390 ms). By considering these cut-offs, the likelihood of fluid responsiveness decreased from 44% to 21%-25%. Further strengthening of this prediction required that both ΔSV and the raised cut-off for SV were also considered. Hence, anesthetists could greatly reduce F I G U R E 2 Receiver operating characteristic (ROC) curves showing the sensitivity and (1-specificity) for ranges of hemodynamic parameters used to indicate whether a subsequent fluid bolus was warranted, that is, increased stroke volume (SV) by ≥ 10%. The curves are arranged to display increased sensitivity to predict fluid responsiveness. One of the given variable values is the optimal cut-off, and the others indicate the scale and direction of the continuous hemodynamic measurement.
the risk of providing a bolus infusion that later would prove not to be warranted.
Applying cut-offs on the opposite side of the hemodynamic spectrum increased the chance of providing warranted bolus infusions to 65%-70%, which is approximately 50% above the overall average. In general, the results indicating increased likelihood of fluid responsiveness mirrored those that reduced the likelihood, but combining hemodynamic variables was somewhat more useful for the former.
A surprising finding was that the SVI was the same in nonresponders and responders before an infusion was given ( Table 1).
The modest difference in SV between these groups might then only be due to the body size of the patients. Hence, cardiac capacity rather than hypovolemia could be the major determinant of who would be fluid-responsive. More precise indications of fluid responsiveness by the hemodynamic parameters would probably be given if young patients with blood loss had been studied.

| Esophagus Doppler
The statistical correlation between cardiac output and body surface is limited, and not superior to the correlation between cardiac output and body weight. 18 There are several methods for calculation of the body surface area that may give quite different results. 19 The CardioQ device calculates SVI using the Du Bois equation, 16 but these values were not recorded. In retrospect, we calculated the body surface area according to both de Bois equation and an alternative equation. 20 However, the choice of equation had hardly any influence of the difference in SVI between responders and nonresponders (data not shown).
A 10% decrease in SV is a common trigger for initiating a new optimization round in many GDFT algorithms. However, we and others have described its limited performance in aiding this decisionmaking. 21,22 Contrary to what is usually assumed, maximal or optimal SV may not always be constant throughout a surgical procedure. Such variations may be due to changes in adrenergic tonus due to anesthetic drugs and occasional use of vasopressors.
FTc is a parameter that several authors have found useful as an indicator of fluid responsiveness, 7,23 but it might be a better measure of afterload by being inversely proportional to systemic vascular resistance. 8 In our present evaluation, we found FTc to be a modestly good indicator of fluid responsiveness by increasing the likelihood by almost 10%, which is comparable to SV ≤80 mL. Moreover, high FTc values were more effective in precluding fluid responsiveness than vice versa. However, both FTc and SV are expressed in absolute values and, therefore, may be more susceptible to measurement errors than ΔSV and PVI.
The ROC curves for SV and FTc showed poor discriminative ability to indicate fluid responsiveness before an infusion is given, which is why we tested several cut-offs for these variables. However, the ROC curves expressed how many warranted and not warranted infusions were indicated as hemodynamic parameters varied, while the likelihood calculations compared the incidence of responsiveness between predetermined and quite wide hemodynamic ranges.

| Pulse oximetry
Pulse oximeters that report PVI values may indicate fluid responsiveness as well, 24,25 but this approach has not reached widespread acceptance. Indeed, we previously found poor concordance between EDM and PVI. 13 Our present evaluation confirmed that PVI reduces the likelihood of fluid responsiveness to 38% but increases it up to 53% depending on whether the PVI value is lower or higher than 10%. These predictions only marginally changed by applying other cut-off values (data not shown).
Pulse oximetry could also augment indications given by EDM by increasing the likelihood of a fluid bolus being warranted, which has rarely been reported previously. Deng et al. reported a metaanalysis showing much improved performance by using a combination of dynamic GDFT (e.g., PVI) and CO/CI goals compared to using only dynamic GDFT goals. 26 Taken together, these data support the idea that the evaluation of multiple parameters instead of one may offer benefits when making decisions about whether to administer fluid. 12

| Limitations
Our present study assumes that the clinician wants to follow a GDT protocol that has been recommended for major surgery by many authorities. 1 Other limitations of this report include that it represents an explorative retrospective analysis of a prospective single-center study.
This implies that the findings may need to be confirmed in a separate population. Some limitations mentioned in our previous reports should also be considered regarding the present analysis. 13 For clinical reasons, a tidal volume of 7 mL/kg predicted body weight was used although a tidal volume of at least 8 mL/kg is usually recommended when using dynamic parameters such as PVI. This choice could have limited the performance of the PVI.

| CONCLUSION
Considering one or several advanced hemodynamic variables before infusing fluid during major abdominal surgery correctly predicted fluid responsiveness with a likelihood of between 20% and 70%.
Combinations of several parameters were associated with the greatest

ACKNOWLEDGMENTS
We are indebted to research nurses Susanne Lind and Gunilla Gagnö for assistance with data collection. This work was supported by the Department of Anaesthesiology and Intensive Care, Linköping University Hospital, and the County Council of Östergötland. RGH has received a grant from Grifols for the study of 20% albumin as a plasma volume expander.

FUNDING INFORMATION
This work was supported by the Department of Anaesthesiology and Intensive Care, Linköping University Hospital, and the County Council of Östergötland.

DATA AVAILABILITY STATEMENT
The data used for the mathematical analysis are available as Supplementary File 1.