In ICUs, fluid administration is frequently used to treat hypovolaemia to enhance cardiac function by increasing preload. Many studies have demonstrated that fluid responsiveness can be predicted by using respiratory derivative indices (pulse pressure variation (ΔrespPP), stroke volume (SV) variation (ΔrespSV) and aortic velocity-time integral variation (ΔrespVTIAo)) 1 2 3 4 5 . From a clinical perspective, owing to altered alveolar capillary membrane permeability, fluid management is critical to the outcomes of ARDS patients 6 7 . In ARDS, dynamic indices predictive of fluid responsiveness present limitations related to the effects of the cardiopulmonary disease on heart-lung interactions (right-sided heart failure, pulmonary hypertension and protective ventilation) 8 9 10 11 12 . In addition, as fluid overload can be harmful, indices using passive leg raising (PLR) have been validated 13 14 15 16 17 . By shifting blood from the lower limbs and splanchnic compartment, PLR is a safe, reversible manoeuvre that mimics fluid expansion 16 . In adult patients with refractory ARDS, despite ventilatory optimisation by means of routine therapies (protective mechanical ventilation, prone position and nitric oxide) 18 19 , the use of respiratory assistance such as venovenous extracorporeal membrane oxygenation (ECMO) ensures oxygenation and decarboxylation 20 21 22 . Venovenous ECMO consists of a circuit supplied by a centrifugal pump without a venous reservoir. Venous return and pump venous injection are preload- and postload-dependent processes that run in parallel to the human right-sided circulation and may interfere with it 23 24 25 . In hypovolaemic patients, PLR prediction of fluid responsiveness is unclear. Blood transfer induced by PLR may be modified by the preload dependence of the ECMO. Because patients supported by ECMO are the frailest ARDS patients, because they present with several interrelated diseases that limit the use of respiratory dynamic criteria, and because fluid therapy can affect outcomes, it is necessary to validate additional manoeuvres such as PLR that may better discriminate responders from nonresponders.

The main goal of this study was to answer the following question: Can PLR be used to predict fluid responsiveness in ARDS patients placed on venovenous ECMO? As ECMO pump flow (PO) is a preload-dependent process, we assumed that changes in PO between baseline and PLR (ΔPLRPO) could reflect a preload-dependent condition. Thus, we analysed the predictive value of ΔPLRPO.

Twenty-five measurements were obtained from seventeen patients. PLR manoeuvres and VE were performed twice on eight patients on different days and under different respiratory and haemodynamic conditions. The patients had acute circulatory failure associated with septic shock and primary or secondary ARDS. The most common cause of heart failure was acute cor pulmonale, defined on the basis of echocardiographic criteria 27 . The patient characteristics are reported in Table 1. All VEs were performed on the basis of the following criteria: arterial hypotension (SAP < 90 mmHg and/or MAP < 70 mmHg) (n = 12), oligoanuria (urine output < 0.5 mL/kg/hour or < 20 mL/hour) (n = 9), skin mottling and/or leg coldness (n = 4).

Of the 25 patients who underwent VE, 13 (52%) were responders: SV increased by > 15%. At baseline, the haemodynamic and echocardiographic parameters were similar in both groups. None of these parameters were predictive of fluid responsiveness. Tables 2 and 3 present the haemodynamic, echocardiographic and PO parameters. SV and CO increases during PLR and after VE were correlated (respectively, r ² = 0.72, P = 0.0001 (Figure 1); r ² = 0.70, P = 0.0001). This correlation was not observed for PO (r ² = 0.07, P = 0.79).

In the whole population, the AUCs were 0.88 ± 0.07 for ΔPLRSV (95% confidence interval (CI₉₅): 0.69 to 0.97; P < 0.0001) and 0.87 ± 0.07 for ΔPLRCO (CI₉₅: 0.67 to 0.97; P < 0.0001) (Figure 2). The ΔPLRCO and ΔPLRSV AUCs were not different (P = 0.66). ΔPLRPO had poor sensitivity and specificity, as well as a poor AUC: 0.53 ± 0.12 (CI₉₅: 0.32 to 0.73; P = 0.79) (Figure 2). The ΔrespPP AUC was 0.56 ± 0.12 (CI₉₅: 0.35 to 0.77; P = 0.59) (Figure 2). The ΔPLRPP AUC was 0.62 ± 0.11 (CI₉₅: 0.4 to 0.8; P = 0.31) (Figure 2). In ten patients, CVP increased by ≥ 2 mmHg (six responders and four nonresponders). Among these patients, the AUCs were 0.83 ± 0.13 for ΔPLRSV (CI₉₅: 0.48 to 0.98; P = 0.012) and 0.83 ± 0.13 for ΔPLRCO (CI₉₅: 0.48 to 0.98; P = 0.012). Table 4 reports the different threshold values of ΔPLRSV and ΔPLRCO.

Our study demonstrates that a 10% increase ΔPLRSV predicts a > 15% increase in SV after VE in ARDS patients placed on venovenous ECMO (Figure 3). ECMO is an efficient treatment for refractory ARDS 22 . No studies have assessed the usual dynamic parameters of preload reserve on ECMO-assisted patients, probably because of the complex interactions between the protective mechanical ventilation and ECMO system on heart-lung interaction. As previously shown in the context of ARDS without ECMO support, neither ΔrespPP nor ΔPLRPP predicts fluid responsiveness 8 9 10 12 . In two prospective studies, the poor predictive performance of ΔrespPP has been attributed to insufficient changes in transpulmonary and pleural pressure 10 12 , which are related to protective ventilation and altered pulmonary compliance. In addition, ΔrespPP could reflect postload variation on right ventricular dysfunction and cannot be used as a predictor even in the presence of hypovolaemia 8 9 11 . Such clinical situations are frequent in the treatment of ECMO-assisted patients, which limit the use of such indices. ECMO patients with late ARDS were ventilated with 'ultraprotective' ventilation because of altered compliance and a high incidence of acute cor pulmonale (Table 1). As Δresp indices must be avoided because they fail to predict fluid responsiveness and fluid overload, fluid management may rely on a reversible and safe fluid challenge. Thus, we assessed predictive values of ΔPLR indices.

ΔPLRPP cannot predict fluid responsiveness, even among patients with a minimal increase in CVP of 2 mmHg. On the basis of physiological knowledge, the use of PP as a substitute for SV would assume constant arterial compliance. By increasing intrathoracic blood volume, PLR may also induce sympathetic activation; however, our patients were deeply sedated, and their HRs remained unchanged during PLR. In clinical practice, SV, PP and vascular tone can vary with the patient's haemodynamic conditions and can be altered by PLR 28 , which may have been the case in our present study.

In this context, echocardiographic measurement of ΔPLRSV and ΔPLRCO may predict fluid responsiveness. The best threshold was 5% for ΔPLRSV, with 92% sensitivity (CI₉₅: 64 to 100) and 83% specificity (CI₉₅: 52 to 98), and 5% for ΔPLRCO, with 85% sensitivity (CI₉₅: 46 to 95) and 83% specificity (CI₉₅: 52 to 98) (Table 4). The lower sensitivity of ΔPLRCO may be explained by the fact that CO is the product of SV and HR. In some responders, the VE-induced SV increase was associated with decreased HR (nonsignificantly). Whereas sensitivity differed, the ΔPLRCO and ΔPLRSV AUCs were not statistically different. Taking into account echocardiography, interindividual reproducibility and the fact that we wanted to avoid volume overloading, a threshold of 10% for ΔPLRSV was proposed, with 92% specificity and 62% sensitivity (Table 4 and Figure 3).

Although ECMO modifies the right-sided heart load condition and interferes with pulmonary artery circulation 20 23 24 , it does not prevent an increase in venous return, as also reflected by increased CVP and persistent fluid responsiveness during PLR and after VE with sensitivity and specificity percentages comparable to values usually recorded in the ICU without ECMO 13 14 15 17 . One assumption might be that, in hypovolaemic patients, fluid expansion may load the ECMO pump first and alter VE (and ΔPLR predictive values), as patients' SV may not increase. In four nonresponders, PO increased but SV also increased by > 10%. Moreover, ΔPLRSV were correlated with VE-induced SV variations, with values close to those recorded in the ICU among patients not being treated with ECMO 14 . Thus, in nonresponders, an increase in CVP without a change in CO and PO confirms an increase in right-sided preload and the fact that fluid expansion should not refill the ECMO system. Unlike Lakhal et al. 15 , we found that the CVP measurement did not improve the accuracy of ΔPLR indices. CVP values should be analysed with caution, even if the mean CVP values increased with PLR and VE. Since the distal extremities of the ECMO canula and CVC are close, the CVP mean value may vary depending on measurement variations. In addition, this subgroup analysis was performed in only 10 patients.

Since venous return and pump venous injection are preload- and afterload-dependent processes, we assumed that ECMO parameters could demonstrate preload-dependent conditions. At baseline, the PO/RPM ratio, which may reflect PO adequacy to preload status, was not different between the two groups. PLR and VE increase PO and PO/RPM ratio and decreased PI only when the patient was preload-dependent, that is, only in responders (Table 3). Nevertheless, ΔPLRPO was not predictive of fluid responsiveness (Figure 2). Different mechanisms may explain this result. First, low variations in PO may have exposed ΔPLRPO to an insufficient signal-to-noise ratio that was distorted by our method of calculating PO. Second, the afterload dependency of ECMO may alter PO. As we did not investigate ECMO afterload, we cannot link an increase in PO to an increase in ECMO preload and/or a decrease in ECMO afterload. Indeed, the present results confirm the cardiac dependence of the venovenous ECMO circuit. The venovenous ECMO circuit acts as a cardiac preload sensor (more than a volume sensor). Nonetheless, for the reasons mentioned above, we have not demonstrated that ΔPLRPO is predictive of fluid responsiveness.

This study has a number of limitations. The small number of measurements may limit the interpretation of the results, but this number is comparable to the sample sizes of other published studies and the statistical significance is sufficient 17 . The ΔPLRSV cutoff is close to ultrasonic interindividual reproducibility. This threshold is comparable to values usually recorded in the ICU 17 29 and is more than twice the values of inter- and intraindividual reproducibility. We assessed SV using TTE, which has been validated against the thermodilution technique 30 . Thermodilution monitoring cannot be used because of recirculation phenomena 31 , and preload reserve haemodynamic indices are not currently validated in this setting for the reasons described above. There were some concerns about the safety of PLR among ECMO patients. In our cohort, all patients underwent PLR without any adverse impact on the ECMO system. Before PLR, however, precautions were taken regarding the length of inflow and outflow cannulas.

In this study, a > 10% increase in ΔPLRSV was predictive of fluid responsiveness in patients placed on venovenous ECMO respiratory assistance. This diagnostic procedure is easy to perform, reversible, familiar to intensive care physicians and easily reproducible, and it may be helpful in reliably identifying patients who will benefit from fluid loading. In contrast, we have not demonstrated that ΔPLRPO and ΔrespPP can be used to predict volume responsiveness. Further studies of a larger sample of patients placed on various types of ECMO are necessary to assess these results.

• Derivative pulse pressure indices (ΔrespPP and ΔPLRPP) failed to predict fluid responsiveness in ARDS patients placed on venovenous ECMO.

• A > 10% increase of ΔPLRSV may predict fluid responsiveness in patients treated with venovenous ECMO.

• ΔPLRPO cannot be used to predict fluid responsiveness.

ΔPLRPO: passive leg raising pump outflow change; ΔPLRPP: passive leg raising pulse pressure change; ΔPLRSV: passive leg raising stroke volume change; ΔrespPP: respiratory pulse pressure variation; ΔrespSV: respiratory stroke volume variation; ARDS: acute respiratory distress syndrome; CI₉₅: 95% confidence interval; CO: cardiac output; CVC: central venous catheter; CVP: central venous pressure; DAP: diastolic arterial pressure; ECMO: extracorporeal membrane oxygenation; IRB: Institutional Review Board; LVEF: left ventricular ejection fraction; MAP: median arterial pressure; PI: pulse index; PO: pump outflow; PLR: passive leg raising; ROC: receiver operating characteristic; RPM: rotation per minute; SAP: systolic arterial pressure; SD: standard deviation; SV: stroke volume; TTE: transthoracic echocardiography; VE: volume expansion; VTIAo: aortic velocity-time integral ratio.

The authors declare that they have no competing interests.

PGG conceived, designed and coordinated the study and drafted the manuscript. EZ, MD and TC participated in the coordination of the study. MM, VH, LB and EB participated in the coordination of the study and helped in drafting the manuscript. HD performed the statistical analysis and helped in drafting the manuscript.