Trendelenburg maneuver predicts fluid responsiveness in patients on veno-arterial extracorporeal membrane oxygenation

Background Evaluation of fluid responsiveness during veno-arterial extracorporeal membrane oxygenation (VA-ECMO) support is crucial. The aim of this study was to investigate whether changes in left ventricular outflow tract velocity–time integral (ΔVTI), induced by a Trendelenburg maneuver, could predict fluid responsiveness during VA-ECMO. Methods This prospective study was conducted in patients with VA-ECMO support. The protocol included four sequential steps: (1) baseline-1, a supine position with a 15° upward bed angulation; (2) Trendelenburg maneuver, 15° downward bed angulation; (3) baseline-2, the same position as baseline-1, and (4) fluid challenge, administration of 500 mL gelatin over 15 min without postural change. Hemodynamic parameters were recorded at each step. Fluid responsiveness was defined as ΔVTI of 15% or more, after volume expansion. Results From June 2018 to December 2019, 22 patients with VA-ECMO were included, and a total of 39 measurements were performed. Of these, 22 measurements (56%) met fluid responsiveness. The R2 of the linear regression was 0.76, between ΔVTIs induced by Trendelenburg maneuver and the fluid challenge. The area under the receiver operating characteristic curve of ΔVTI induced by Trendelenburg maneuver to predict fluid responsiveness was 0.93 [95% confidence interval (CI) 0.81–0.98], with a sensitivity of 82% (95% CI 60–95%), and specificity of 88% (95% CI 64–99%), at a best threshold of 10% (95% CI 6–12%). Conclusions Changes in VTI induced by the Trendelenburg maneuver could effectively predict fluid responsiveness in VA-ECMO patients. Trial registration ClinicalTrials.gov, NCT 03553459 (the TEMPLE study). Registered on May 30, 2018

Several methods are currently available to evaluate fluid responsiveness in critically ill patients [7][8][9]. Mechanical ventilation generates a cyclical change in intra-thoracic pressure and venous return. Based on this interaction, stroke volume variation (SVV) and pulse pressure variation (PPV) are often used as dynamic parameters to predict fluid responsiveness [8]. However, VA-ECMO patients frequently present with low tidal volume, cardiac arrhythmia and especially pulselessness, thus making these parameters less reliable [8,10,11]. Respiratory variations in the inferior vena cava diameter (ΔIVCD) have previously been used as accurate predictors for fluid responsiveness in mechanically ventilated patients [12]. However, drainage cannulation placed in the inferior vena cava impedes the application of ΔIVCD. Simulation of a revisable "autotransfusion" is another feasible approach to assess fluid responsiveness. Passive leg raising (PLR) test transiently increases venous return by postural maneuver [8]. Unfortunately, the risk of cannula distortion or displacement precluded the application of this maneuver to VA-ECMO patients. Overall, the physiological features of VA-ECMO patients restrict the use of conventional methods to assess fluid responsiveness [11].
The Trendelenburg maneuver is a "self " volumetric loading maneuver [13], and has demonstrated good accuracy in predicting responsiveness for acute respiratory distress syndrome [14] and surgical patients [15], and can be performed for the majority of VA-ECMO patients. Transthoracic echocardiography (TTE) is now routinely used to evaluate cardiac function recovery, and allows the measurement of stroke volume and cardiac output in VA-ECMO patients [16]. Thus, the combination of the Trendelenburg maneuver and TTE facilitates the evaluation of fluid responsiveness during VA-ECMO support.
This study was designed to investigate whether a change in velocity-time integral (ΔVTI) measured by TTE induced by the Trendelenburg maneuver, could predict fluid responsiveness in patients on VA-ECMO.

Patients
This study was approved by the Institutional Review Board (Zhongshan Hospital, Fudan University, Shanghai, China: Number B2018-074), and conducted in a 40-bed cardiac surgical intensive care unit (ICU). During the study period (June 2018-December 2019), 22 postoperative patients were prospectively enrolled after informed consent was received from the patient's next of kin. We included ventilated patients with relatively low VA-ECMO pump flow (2-3 L/min) for whom the decision to perform volume expansion was made by the attending physician (hypotension, hypoperfusion [oliguria or skin mottling], or attempt to reduce vasopressor dose [17,18]). Exclusion criteria were patients < 18 years old, pregnant, pulselessness (pulse pressure (PP) < 15 mmHg [19]), contraindication to the Trendelenburg position (cerebral edema, intra-abdominal hypertension and gastric retention), or unsatisfactory cardiac echogenicity (an inability to correctly align the Doppler beam to generate reliable VTI measurements at the left ventricular outflow tract [LVOT]). Patients with evidence of significant hypovolemia, such as kicking drainage cannula [20] (suggesting a transient venous or atrial collapse) and persistent hemorrhage, were also excluded.

Protocol
Throughout the study, patients were sedated with a combination of remifentanil and midazolam, with the aim of achieving a Richmond Agitation-Sedation Scale [21] of − 5. The protocol included four sequential steps ( Fig. 1): 1. Baseline-1: a supine position with a 15° upward bed angulation, 2. Trendelenburg maneuver: 15° downward bed angulation [22], 3. Baseline-2: recover to the same position as baseline-1, and 4. Fluid challenge: administration of 500 mL gelatin over 15 min without postural change.
After 1 min stabilization for each step, VTI, heart rate (HR), systolic blood pressure (SBP), diastolic blood pressure (DBP), central venous pressure (CVP) and pulse pressure (PP) were recorded. In our center, the tip of the drainage cannula was placed in the right atrium to guarantee adequate blood collection [11]. The pressure transducer was fixed at the level of the patient's right atrium, which was located at the intersection of the mid-axillary line and the fourth interspace. During the study, the pump was maintained at the same rotation speed. Other therapies such as sedation, vasopressors and ventilation also remained unchanged. To acquire as much data as feasible, some patients underwent the test more than once (on different days) if required, therefore 39 measurements were conducted and recorded.

VTI measurements
Transthoracic echocardiography was performed on a CX50 instrument (Philips Healthcare, Eindhoven, The Netherlands) by the same experienced operator (LL Dong) who was blinded to patient clinical data. VTI was measured at the level of the LVOT, using the 5-chamber apical view. Five consecutive measurements were recorded to calculate a mean VTI value. For TTE measurement, some indices were defined to estimate the reproducibility [17,23]: (1) coefficient of variation (CV) = standard deviation (SD)/mean; (2) coefficient of error (CE) = CV/√n (the number of replicate TTE); (3) precision = 2 × CE; (4) least significant change (LSC) = CE × 1.96 × √2, indicting the minimal observed change value can be considered as real. These assessments were performed in eight patients at baseline position by obtaining twice by the same operator (LL Dong; intra-observer reproducibility) and a second operator (Y Liu; inter-observer reproducibility). The CV was 4% for intra-observer variability, and 5% for inter-observer variability. The intra-observer LSC was 8% for VTI measurements.

Data collection
Upon patient inclusion, demographic information, surgical procedures, Acute Physiologic and Chronic Health Evaluation (APACHE) II scores, left ventricular ejection fraction (LVEF), laboratory examinations and blood gas analyses were collected. Supportive therapies included VA-ECMO and mechanical ventilation (MV) indices, vasoactive drug doses (transferred to equivalent inotropic scores) [24], and the concomitant use of intraaortic balloon pumps (IABP), renal replacement therapy (RRT) or inspired nitric oxide (iNO). We recorded all data related to these parameters. All study patients were followed up until hospital discharge or death, to record clinical outcomes, such as length of MV, tracheotomy rate, length of ICU stay, length of hospital stay and hospital mortality.
Continuous variables were expressed as medians (with interquartile ranges [IQR]) and compared between groups using the Wilcoxon or Friedman rank sum tests. For pairwise multiple comparisons, we used the Nemenyi post hoc test. Categorical variables were expressed as numbers (and percentages) and compared using the Fisher's exact test. Linear regression analysis was used to demonstrate relationships between percent change of VTI (ΔVTI) induced by the Trendelenburg position, and fluid challenge. Fluid responsiveness was defined as a VTI increase of 15% or more after volume expansion [17].
Receiver operating characteristic (ROC) curves were generated to evaluate percent changes in VTI and arterial pressure parameters (ΔSBP, ΔDBP and ΔPP), induced by the Trendelenburg maneuver to predicts fluid responsiveness. The area under ROC curves (AUROC) were compared using the DeLong test [25]. Sensitivity, specificity, positive and negative predictive values (PPV and NPV), and associated 95% confidence intervals (CI) were calculated based on the cutoff value as determined by the Youden Index (specificity + sensitivity − 1) [26]. To evaluate the variation of best threshold, we conducted a gray-zone analysis, as described by Georges et al. [17]. To visualize whether a predictive test can recognize the positive events fully and accurately, the recall (i.e., true positive rate)-precision (i.e., PPV) curves (PRC) were generated and area under precision-recall curve (AUPRC) were calculated [27].
All statistical tests were two-tailed, and a value of p < 0.05 indicated statistical significance. Statistical analyses were performed using R, version 3.6.2 (R Foundation for Statistical Computing, Vienna, Austria).

The effects of the Trendelenburg maneuver and fluid challenge
Fluid responsiveness was observed in 22 of the 39 measurements (56%). Hemodynamic parameters from each step are shown (Table 3 and Fig. 3a, b) Fig. 2a). In comparison, the R 2 of ΔSBP, ΔDBP and ΔPP were 0.2167, 0.0715 and 0.2187, respectively ( Fig. 2b-d).

Discussion
To the best of our knowledge, this is the first study to explore the predictive value of ΔVTI, induced by the Trendelenburg position for fluid responsiveness in patients with VA-ECMO support. Our work demonstrated that ΔVTI monitoring during the Trendelenburg position was a reliable parameter in predicting fluid responsiveness in this population. VA-ECMO could provide extracorporeal life support, as well as reducing the workload of the injured heart, thereby creating favorable conditions for myocardial recovery [1,2]. During this support, clinicians usually increase ECMO flow to correct hypotension, which would lead to a higher left ventricular afterload and a downward shift of the Frank-Starling curve [29]. Apart from device flow, native cardiac output also plays an important role in maintaining systemic circulation, especially when the heart is in recovery, and device flow decreases correspondingly. Moreover, the volume status of the heart varies at any time due to hemorrhage, capillary leak, fluid therapy and changes in cardiac function, and hence should be dynamically evaluated.
Fluid responsiveness, as a cardiac response parameter to additional preload, is typically used for septic shock patients [8]; however, it may also be useful for patients with VA-ECMO support. On one hand, during VA-ECMO support, traditional parameters such as ejection  [30]. Fluid responsiveness may indicate whether the heart works at the steeper part of the Frank-Starling curve, thus guiding volume administration for hypotension, thereby providing indications for myocardial recovery and weaning feasibility from VA-ECMO. On the other hand, fluid unresponsiveness could also be used to guide fluid removal in ventilated patients with fluid overload [31], which may induce heart congestion, pulmonary edema, and acute kidney injury [32,33], potentially increasing mortality [34]. Hence, assessment of fluid responsiveness could help to optimize preload and also evaluate cardiac function.
The Trendelenburg maneuver is a method that facilitates "autotransfusion". Geerts et al. concluded that the final changes in cardiac output induced by the Trendelenburg position were similar to PLR [13]. In this study, the Trendelenburg position induced a 58% change in VTI, induced by a 500-mL fluid challenge, indicating a similar physiological effect to PLR. The high correlation between changes in VTI induced by the Trendelenburg position or fluid challenge allowed us to accurately predict fluid responsiveness in VA-ECMO patients.
Arterial pulsatility, coming from left ventricular ejection, is often considered as a marker of cardiac recovery during VA-ECMO [4,35]. PP, as a parameter of pulsatile-flow, has been used as a predictor of successful weaning from VA-ECMO [36,37]. However, in this study, changes in PP had a poorer predictive performance for fluid responsiveness than VTI. Although both variables could reflect the stroke volume, the PP is just a difference between SBP and DBP, rather than the area under the pulse contour per se. In fact, the relationship between SV and peripheral PP changes were not straightforward because they depended on arterial compliance, and pulse wave amplification from the aorta to the periphery. Previous studies have indicated that changes in PP exerted significant heterogeneity towards predictive accuracy for fluid responsiveness [38,39]. Similarly, neither ΔSBP nor ΔDBP, induced by the Trendelenburg maneuver, demonstrated an acceptable predictive accuracy in this study.

Study limitations
Our study had several limitations. First, it was conducted in a single center, which may have limited generalizability across different clinical settings. Second, a larger angle of Trendelenburg positioning may have introduced more "autotransfusion" to the central circulation [13], and was not evaluated in this study. Third, measurement of LVOT VTI is not the gold standard for evaluating cardiac output, because of its dependence on patient echogenicity and operator expertise. However, techniques based on thermodilution have proven unreliable in VA-ECMO patients [3], although they were accurate in estimating cardiac output under most circumstances [40]. Fourth, all patients in this study were under deep sedation, thus the Trendelenburg maneuver may be less reliable in predicting fluid responsiveness in non-sedated patients. Finally, because the study was conducted in a low VA-ECMO pump flow setting, our data may not be extrapolated to the acute phase of heart failure requiring full mechanical circulatory support.

Conclusions
Our study suggested that an increase in VTI of at least 10%, induced by the Trendelenburg maneuver, was reliable in predicting fluid responsiveness in patients with VA-ECMO.

Table 4 Predictive parameters of receiver operating characteristic (ROC) curves of variable percent changes induced by Trendelenburg position
Values are median true value (95% confidence interval). Best threshold value was determined using the Youden index The AUROC of ΔVTI was higher than ΔPP (p = 0.004), ΔSBP (p = 0.025) and ΔDBP (p = 0.018). There were no statistical differences among AUROCs of ΔPP, ΔSBP and ΔDBP (ΔPP vs. ΔSBP, p = 0.285; ΔPP vs. ΔDBP, p = 0.760; ΔSBP vs. ΔDBP, p = 0.645) AUROC area under the receiver operating characteristic curve, PPV positive predictive value, NPV negative predictive value, SBP systolic blood pressure, DBP diastolic blood pressure, PP pulse pressure, VTI velocity-time integral, Δ value indicate percent change of each variable