Change in cardiac output during Trendelenburg maneuver is a reliable predictor of fluid responsiveness in patients with acute respiratory distress syndrome in the prone position under protective ventilation

The primary aim of this study was to evaluate the diagnostic performance of cardiac index variation during a Trendelenburg maneuver to predict fluid responsiveness in patients with ARDS under protective ventilation in the PP. Secondary objectives were to evaluate the diagnostic performance of cardiac index variation during an EEO, and both PPV and change in PPV from baseline during a VT challenge from 6 to 8 ml.kg^-1 PBW.

This study is a prospective single-center study, performed between October 2013 and January 2017 in a 15-bed medical ICU and registered at ClinicalTrials.gov (NCT01965574). The study protocol (see Additional file 1) was approved by the local ethics committee (Comité de Protection des Personnes Sud-Est IV, ID-RCB-2013-A00526-39). Written consent from the patients’ closest relatives was required for inclusion, and eventually confirmed by the patient after ARDS resolution.

The subjects had to fulfill all the following inclusion criteria: ARDS according to the Berlin definition [18], ongoing session of PP under invasive mechanical ventilation, ongoing monitoring with the PiCCO® device (Pulsion Medical Systems, Feldkirchen, Germany), and decision by the attending physician to administer fluids with at least one criterion of acute circulatory failure among the following: arterial lactate >2 mmol.L^-1, mean arterial pressure <65 mm Hg, cardiac output decrease, urine output <0.5 ml.kg.h^-1, heart rate >100 min^-1 and skin mottling.

Non-inclusion criteria were the following: age <18 years, contra-indication to the Trendelenburg position, pregnancy, lower limbs amputation, known obstruction of inferior vena cava, previous inclusion in current study, and patient under a legal protection measure as required by French regulation. Patients exhibiting respiratory effort detected on the pressure-time curve displayed on the ventilator during a 15-s EEO were excluded from the study.

Patients were deeply sedated with a combination of morphine and midazolam targeting a Ramsay score of 6 [19], and remained in PP with a 13° upward bed angulation throughout the study, except during the Trendelenburg maneuver. They were ventilated in volume-controlled mode with a VT 6 ml.kg^-1 PBW. Patients were studied at baseline (baseline-1), during a 1-min postural change to the Trendelenburg position with a −13° downward bed angulation (Fig. 1), during a 1-min VT challenge at 8 ml.kg^-1 PBW, during a 15-s EEO maneuver, and after intravenous infusion (IV) of 500 ml crystalloids over 15 min. Patients were returned to baseline settings for 1 min after each intervention (see Additional file 2). The following adverse events were prospectively collected throughout the protocol: drop of systolic arterial pressure >30 mm Hg, increase in heart rate >10%, decrease in peripheral oxygen saturation <88%, new onset of cardiac arrhythmia, or any other adverse event considered relevant by investigators.

Jugular central venous and femoral arterial lines were connected to an Intellivue MP40 monitor equipped with a PiCCO® module (Philips Healthcare, Andover, MA, USA). Pressure transducers were taped on the thorax at the phlebostatic reference point (Fig. 1). The following hemodynamic variables were measured throughout the study: arterial pressure, central venous pressure (CVP), pulse contour-derived cardiac index (CCI), heart rate, and PPV.

Transpulmonary thermodilution measurements were performed at study onset (baseline-1) and after volume expansion using the PiCCO® device. Values were computed as the mean of four consecutive measurements, using a 15-ml bolus of cold saline serum. ΔCCI_TREND was computed as the difference between the maximal value of CCI during Trendelenburg and baseline-1 CCI, normalized by baseline-1 CCI. ΔPPV_6-8 was computed as the difference between the maximal value of PPV during ventilation with VT 8 ml.kg^-1 and baseline-2 PPV, normalized by baseline-2 PPV. ΔCCI_EEO was computed as the difference between the maximal value of CCI during the EEO maneuver and baseline-3 CCI, normalized by baseline-3 CCI. Fluid responsiveness was deemed present if cardiac index assessed by transpulmonary thermodilution (CI_TPTD) increased by at least 15% after volume expansion, as compared to baseline-1 [20].

The study primary endpoint was the diagnostic performance of ΔCCI_TREND to predict fluid responsiveness. Secondary endpoints were diagnostic performance of ΔCCI_EEO to predict fluid responsiveness, and both PPV and ΔPPV_6-8.

Statistical analyses were performed using R [21]. Median (1st quartile to 3rd quartile) and counts with percentages are reported for quantitative and categorical variables, respectively. A p value below 0.05 was chosen for statistical significance.

We calculated that with a sample size of 33 patients, the study would provide at worst ± 0.15 precision for the 95% confidence interval (CI_95%) of the area under the receiver operating characteristic (ROC) curve (AUC), assuming a prevalence of fluid responsiveness of 50% [16, 22‐24] and an AUC of at least 0.8 [25] (i.e. a lower bound for the AUC CI_95% amounting to at least 0.65).

Comparisons between groups of patients were performed with the Fisher’s exact test for categorical variables, and with the t test, Mann-Whitney test or analysis of variance (ANOVA) for continuous and ordinal variables when appropriate. Hemodynamic parameters were compared using a linear mixed effects model [26, 27]. Multiple comparisons between experimental conditions and baseline-1 were performed using Dunnett’s test [28].

Diagnostic performance of tests under investigation was assessed by computation of the AUC [29]. The CI_95% for the AUC was computed using the Delong method. The optimal cutoffs were computed by maximizing the Youden index. The CIs for optimal cutoffs were computed using the gray zone approach (area of uncertainty of optimal cutoffs) [23]. Response to each test below the lower or above the higher border of the gray zone were considered negative and positive, respectively. Responses to the test within the gray zone were considered inconclusive. The CI_95% for sensitivity, specificity and medians were computed using bootstrapping and 10000 replicates [30, 31].

During the study period, 55 patients presented with inclusion criteria (see Additional file 3) and 33 were included, whose general characteristics, cardiovascular and respiratory parameters at inclusion are reported in Tables 1 and 2. There were 14 patients (42%) who exhibited cardiac arrhythmia at inclusion and were excluded in the analyses pertaining to PPV: 15 patients (45%) were classified as fluid responsive after fluid administration.

The total duration of the study amounted to 26 (24–30) min. None of the variables measured after return to baseline settings (baseline-2 to baseline-4) were significantly different from baseline-1 (Table 3). Mean arterial pressure, CVP, CI_TPTD, global end-diastolic volume index and global ejection fraction increased significantly after fluid administration. No adverse event was identified throughout the protocol.

CVP increased significantly during Trendelenburg, while heart rate remained unchanged (Table 3). Median ΔCCI_TREND amounted to 6% (CI_95%, 3–10%) and was significantly greater in responders than in non-responders (13% vs. 3%, p < 0.001, Fig. 2). ΔCCI_TREND was significantly correlated with change in CI_TPTD related to volume expansion (R ² = 0.41, Fig. 3). ΔCCI_TREND predicted fluid responsiveness with an AUC of 0.90 (CI_95%, 0.80–1.00, p < 0.001), with sensitivity of 87% and specificity of 89% at a threshold of 8% (gray zone, 5–12%) (Table 4, Figs. 4 and 5). Cardiac index response to volume expansion increased stepwise in patients with a negative response, those with an inconclusive response, and those with a positive response to the test (see Additional file 4). Four patients were misclassified (Fig. 2), and none of their hemodynamic and respiratory parameters were significantly different from those of the 29 correctly classified patients (data not shown).

PPV_BASELINE-1, PPV_VT8 and ΔPPV_6-8 did not significantly differ between fluid responders and non-responders (Fig. 2). None of the three PPV-derived diagnostic tests were statistically significant for AUC (Table 4, Fig. 4). ΔPPV_6-8 exhibited the greatest sensitivity (100% (CI_95%, 100–100%)) at a threshold of 29%, but with a very low specificity (40% (CI_95%, 10–70%)).

False positive patients with the PPV_BASELINE-1 test had significantly greater driving pressure while true negative had significantly lower PaO₂/FiO₂ ratio (data not shown). CI_TPTD at inclusion was significantly lower in the 5 false positive patients as assessed by PPV_VT8.

CVP decreased slightly but significantly from 7 (5–11) to 6 (4–10) mm Hg during EEO, while CCI remained unchanged (Table 3). ΔCCI_EEO was not significantly different in responders and non-responders (Fig. 2). The AUC of ΔCCI_EEO to predict fluid responsiveness amounted to 0.65 (CI_95%, 0.46–0.84), and was not significantly different from 0.5 (Table 4, Fig. 4). ΔCCI_EEO had sensitivity of 33% (CI_95%, 13–60%) and specificity of 100% (CI_95%, 100–100%) at a threshold of 10% (gray zone, –4% to 11%) to predict fluid responsiveness (Table 4, Fig. 5).

In the 14 patients with change in CVP (ΔCVP) ≥0 mm Hg, the AUC of ΔCCI_EEO amounted to 0.89 (CI_95%, 0.70–1.00) (p < 0.05), while it was not statistically different from 0.5 in the 19 patients with ΔCVP <0 mm Hg (see Additional files 5 and 6).

In the present study, ΔCCI_TREND amounted to 6% (CI_95%, 3–13%), and was in the range of the 9% increase observed in a recent systematic review [32], although performed in subjects in the supine position, with various degrees of head-down tilt angulation. An important issue in the reliability of the Trendelenburg maneuver to predict fluid responsiveness is related to baroreflex activation in this position, leading to systemic vasodilation, decreased heart rate and myocardial contractility. However, we did not observe a significant change in heart rate in the Trendelenburg position as compared to baseline, in keeping with the results of the aforementioned systematic review [32]. While baroreflex activation is immediately evident after carotid declamping in patients who were awake and undergoing carotid surgery, the maximal effect on heart rate is observed 10 min later [33]. Opposite to this, in healthy volunteers, midazolam has been shown to dose-dependently blunt the fast parasympathetic efferent pathway of the baroreflex [34, 35]. Taken together, these data suggest that the 1-min duration of the maneuver may not have been long enough to significantly activate the baroreflex in deeply sedated patients with ARDS.

The present study confirmed the lack of predictive ability of PPV to predict fluid responsiveness in patients with ARDS under protective ventilation [11‐13]. This finding is not unexpected since cardiopulmonary interactions under mechanical ventilation (the underlying physiological mechanism behind PPV) are dependent on both ventilatory settings and transmission of airway pressure to cardiac filling pressures. This transmission is inversely related to respiratory system elastance [15], and linearly related to the ratio of chest wall to respiratory system elastances [36]. In conditions combining low VT and low respiratory system elastance as observed in patients with ARDS under protective ventilation, a high rate of false negative patients is expected.

Performing a VT challenge did not significantly enhance the reliability of the PPV test, since sensitivity increased at the expense of specificity. While this VT challenge increased the reliability of the PPV test in one study performed on 22 ICU patients in the supine position (9% with ARDS) [13], our data suggest that this finding should not be extrapolated to patients with ARDS on PP. Previous studies have shown that false positive patients for PPV may occur in the context of right ventricular failure [7, 37], and the higher driving pressure in this group of the present study favors this hypothesis [38].

The EEO test has been shown to accurately predict fluid responsiveness in the supine position in four studies [12, 14, 39, 40] including one restricted to patients with ARDS ventilated with VT slightly greater than 6 ml.kg^-1 [14]. However, its predictive performance was poor in a recent study in which 6 ml.kg^-1 VT was first applied, but was restored during a VT challenge at 8 ml.kg^-1 [13]. Our results are in line with this study, and the high rate of false negative patients in our study (30%) suggests that the decrease in the cyclic stress applied to the cardiovascular system during ARDS under protective ventilation (due to both low VT and decreased respiratory system compliance) is not sufficient to generate a detectable effect on cardiac index in some patients. We therefore hypothesized that the PP-induced increase in intra-abdominal pressure [41] could generate an impediment to venous return, promoting a zone-2 condition in the inferior vena cava in some patients [42] (and hence a pressure gradient between the inferior vena cava and the right atrium), and could explain the high false negative rate in our study. The slight decrease in CVP during the EEO in 53% of the patients, combined with the low predictive value of ΔCCI_EEO in this subpopulation favors this hypothesis, although the lack of direct measurement of intra-abdominal pressure and venous return precludes any definite conclusion. Finally, it should be emphasized that the EEO remains a highly specific test in patients with ARDS under protective ventilation in PP.

Some limitations of the present study should be acknowledged. First, the monocentric feature of this study questions the generalizability of its results. Second, amplifying the postural change during the Trendelenburg maneuver (beyond 13° and −13°) could have maximized the blood transfer from the lower body parts towards the central circulation and may have further increased the sensitivity of this test. Third, the high number of patients with cardiac arrhythmia (therefore excluded from PPV analyses) makes the study strongly underpowered for all analyses pertaining to this test. Fourth, the lack of blinding precludes control of a potential evaluation bias. Fifth, cardiac output assessed from pulse contour analysis may be inaccurate in relation to change in resistive and elastic characteristics of the vascular system, although its reliability is acceptable during the hour following calibration [43]. Finally, the lack of randomization between the three maneuvers performed to predict fluid responsiveness (namely Trendelenburg maneuver, VT challenge and EEO) could have hampered a reliable evaluation of the latter two tests.

Risk minimization is an important issue in fluid administration in patients with severe ARDS, given the potential for harm of unnecessary fluid bolus. Using the Trendelenburg test, 70% of the patients could be classified outside the gray zone, meaning that fluid responsiveness was assessed with near certainty in the majority of the patients. Regarding the 30% of patients within the gray zone of the Trendelenburg test, fluid administration may be considered, although response to fluid therapy may be less intense in this group. Finally, whether guiding fluid therapy using indices of fluid responsiveness improve ARDS prognosis remains unknown, although it may help to decrease fluid administration in patients with septic shock [44].