Introduction

Initial fluid resuscitation remains the first treatment step for most children experiencing circulatory failure and/or systemic hypotension. Repetition of volume expansion (VE) in critically ill children under mechanical ventilation may be ill-advised, however, as inappropriate fluid challenge could worsen or result in pulmonary edema. Studies using traditional non-clinical hypovolemia parameters showed poorly predicted fluid responsiveness in adults or provided conflicting results [1]. These parameters were based on static cardiac preload estimations, such as cardiac filling pressure – e. g., central venous pressure (CVP) or pulmonary artery occlusion pressure (PAPO) – and LV end-diastolic diameter or area [2]. To our knowledge, only one pediatric study, performed by Tibby et al., has evaluated the predictive value of fluid responsiveness (i. e. more than 10% increase of stroke volume after 10 ml.kg of human-derived blood product) of two static indices: CVP and esophageal Doppler LVETi [3]. Regrettably, neither index had threshold values for the receiver operating characteristics (ROC) curve (higher than 0.8) and they therefore could not be considered reliable indicators of fluid responsiveness. Recent clinical and experimental studies in adults have emphasized that respiratory variations in aortic blood flow are reliable indicators of cardiac preload reserve [4]. It must be emphasized that volume responsiveness assessment by ultrasound dynamic parameters (i. e. stroke volume variation or its surrogates, superior vena cava collapsibility or inferior vena cava distensibility) is still included in the specific post graduate training program for adult intensivists in France [5, 6]. Concerns have been raised, however, about the usefulness of bedside dynamic parameters using heart–lung interaction in patients with increased chest compliance or those ventilated with small tidal volume [4, 7]. Moreover, it must bear in mind that during either spontaneous breathing or pressure support, pulse pressure or stroke volume variation failed to predict fluid responsiveness [8]. Being aware that children differ from adults in, especially, arterial compliance and chest wall-to-lung elastance ratio, we set out to investigate whether three dynamic parameters validated in adults (systolic pressure variation, pulse pressure variation and aortic peak flow velocity variation) are also reliable indicators of volume responsiveness in sedated children under controlled-mode ventilation [4, 9].

Materials and methods

This prospective and observational study included 26 mechanically ventilated children who required VE between January 2003 and December 2005.

For each patient included in the study, the attending physician (junior or senior) confirmed the need of VE on the basis of tachycardia, hypotension, oliguria, delayed capillary refilling or hemodynamic instability despite vasopressor drugs. Included children had been previously monitored by a radial or femoral artery catheter and were sedated and ventilated on controlled-mode (BiPAP-assist or volume-assist, Evita 2 Dura, Lübeck, Germany). The absence of spontaneous respiratory movements was carefully verified. Availability of the main investigator (P.D.) was required.

Exclusion criteria were arrythmia, congenital cardiac disease with left-to-right shunt, patent ductus arteriosus, severe left ventricular systolic dysfunction (defined as a LV short axis shortening fraction or LV ejection fraction by Simpson rule less than 28% and 50% respectively) and infants under 3 months of age. A few patients were excluded for want of a transthoracic window.

All patients received VE with artificial colloids (Plasmion, Fresenius Kabi, France) or crystalloids (normal saline) in accordance with our unit policy (i. e. volume around 20 ml.kg−1 up to 500 ml infused over 15–30 min). In the 11 patients treated with norepinephrine, the dose was kept unchanged over the entire study period (from before VE until the hemodynamic measurements after fluid challenge).

Hemodynamic measurements were performed just before and at the end of VE and included heart rate (HR) as well as mean, systolic and diastolic arterial pressures. Arterial pressures were recorded through an M1092A bedside monitor (Hewlett-Packard, Les Ullis, France). Pressure waveform analysis was performed online with the soft tracing provided by the monitor and included the whole of the respiratory cycle. ΔPS and ΔPP were calculated as previously described, using maximal and minimal systolic or pulse pressure recorded at end-inspiration and during the expiratory cycle (ΔPS (%) = PAS max–PAS min)/[(PAS max+PAS min)/2] × 100; ΔPP (%) = PP max–PP min)/[(PP max+PP min)/2] × 100) [4].

Echocardiographic measurements were performed with a commercial bedside ultrasound system (EnVisor C HD Philips Ultrasounds, Andover, USA) and an 8-MHz transthoracic transducer. All measurements were made online with the EnVisor software system by a single investigator (P.D.) to avoid interobserver variability. Aortic diameter (D) was measured at the level of the aortic annulus in a two-dimensional view from the parasternal long axis window and was assumed to be unchanged over the time of the study. Pulsed Doppler aortic flow was recorded at the exact level of the annulus from the apical five-chamber view. The aortic velocity–time integral (VTIao) was measured as the mean of three to five consecutive measurements over a single respiratory cycle. LV stroke volume (ml) and cardiac output (l.min−1) were determined as follows using the EnVisor software: CO = SV[VTIao × π (D2)/4] × HR. Maximal and minimal values of aortic peak velocity (Vpeak ao) were determined beat-to-beat over a single respiratory cycle at a speed of 35 mm/s. ΔVpeak ao was calculated as follows: ΔVpeak ao (%) = (Vpeak ao max–Vpeak ao min)/[(Vpeak ao max+Vpeak ao min)/2] × 100 [9].

Statistical analysis

As data were not normally distributed, results were expressed as median and interquartile range (25th–75th percentiles). The effects of volume expansion on hemodynamic parameters were assessed using a nonparametric Wilcoxon rank sum test. Patients showing an increase in LV stroke volume of 15% or more, after volume expansion, were classified as responders (R) (n = 18). Patients whose LV stroke volume increased by less than 15% were classified as non-responders (NR) (n = 8). This cutoff value of 15% was chosen because of the finding in adult studies that this difference is clinically significant for stroke volume or cardiac output after therapeutic intervention [9, 10, 13]. Differences in dynamic parameters between responders and nonresponders prior to VE were established by a non-parametric Mann–Whitney test. ROC curves were plotted for ΔVpeak ao, ΔPS and ΔPP to evaluate the capacity to predict fluid responsiveness. Linear correlation was tested using the Spearman rank method. A p-value < 0.05 was considered to be significant. The intraobserver reproducibility was assessed for LV stroke volume measurement and ΔVpeak ao with Bland and Altman test analysis in eight patients over a 3-min period. The mean bias was 0.12 ± 4 ml for stroke volume and 0.2 ± 3.9% for ΔVpeak ao determination.

The study was approved by the Bicêtre Hospital's ethical review board (no. 05–65). The requirement for informed consent was waived as no specific intervention was required and all hemodynamic measurements were routinely obtained in children.

Results

Table 1 presents the characteristics of the 26 patients, including 18 fluid responders (R; 69%) and 8 fluid non-responders (NR; 31%). Six (23%) of the 26 patients were over 60 months old. Exhaled tidal volume, maximal inspiratory pressure (Bipap assist mode) or plateau pressure (volume-controlled mode) and respiratory rate did not differ between R and NR. Eleven patients (42%) received a vasopressor drug (norepinephrine) for septic and/or distributive shock (8/18 in the R group; 3/8 in the NR group). Hemodynamic parameters before and after volume expansion are shown in Table 2. Crystalloids (normal saline) were administered in only three patients (mean volume 24 ml.kg). Mean arterial pressure, cardiac index and stroke volume increased significantly, with a median VE-induced gain of 12.2% (4.6–21.2), 24% (1.4–51), and 33.6% (12.5–49) respectively. Before VE, the R group showed higher ΔVpeak ao than did the NR group [19% (12.1–26.3) vs. 9% (7.3–11.8), p = 0.001], whereas ΔPP and ΔPS did not significantly differ between the two groups (Fig. 1). The prediction of fluid responsiveness was higher with ΔVpeak ao [ROC curve area 0.85 (95% CI 0.99–1.8), p = 0.001] than with ΔPS [0.64 (95% CI 0.9–1.33), p = ns] and ΔPP [0.59 (95% CI 0.93–1.1), p = ns] (Fig. 2). The best cut-off for ΔVpeak ao as defined by the ROC curve analysis was 12%, for which sensitivity, specificity, and positive and negative predictive values were 81.2%, 85.7%, 93% and 66.6%, respectively. The degree of increase in stroke volume could be significantly predicted by the ΔVpeak ao value. A positive linear correlation was found between the ΔVpeak ao before VE and the volume expansion-induced gain in stroke volume (Rho = 0.68, p 0.001) (Fig. 3). VE induced a significant decrease in ΔVpeak ao, ΔPP and ΔPS (Table 2). An illustrative example of one of the responder patients is shown in Fig. 4. For both R and NR children, Spearman rank test did not reveal any significant correlation between the inspiratory tidal volume (ml/kg) and ΔV peak ao.

Table 1 Patient characteristics at inclusion
Full size table
Table 2 Hemodynamic parameters recorded at baseline and after volume expansion
Full size table
Fig. 1
figure 1

ΔPS, ΔPP and ΔVpeak ao at baseline expressed in plot box form (median, interquartile range) in volume expansion-responders (R, n = 18) and non-responders (NR, n = 8); Values expressed in percent; * p < 0.001

Full size image
Fig. 2
figure 2

ROC curves comparing the ability of ΔPS (top panel), ΔPP (middle panel) and ΔVpeak ao (bottom panel) to discriminate between responders and non-responders to volume expansion

Full size image
Fig. 3
figure 3

Relationship between gain in stroke volume after volume expansion (SV gain) and ΔVpeak ao before volume expansion. Values expressed in percent; Rho 0.68, p = 0.0012

Full size image
Fig. 4
figure 4

Recording of aortic blood flow velocity in one illustrative responder patient (no. 26) before and after volume expansion through an apical window approach. Beat-to-beat measurement of aortic blood flow velocity over a single respiratory cycle allowed the calculation of ΔVao peak % as the difference between Vao peak max and Vao peak min divided by the mean of the two values. In our example, the ΔVao peak falls from 20% before VE (top) to 3.6% after VE (down), while LV stroke volume and cardiac index increase by more than 45% of baseline value

Full size image

Discussion

The main finding of this study is that the respiratory variations in aortic blood flow velocity, displayed by pulsed Doppler with apical transthoracic window before volume expansion, accurately predict fluid responsiveness in children receiving mechanical ventilation. Conversely, ΔPS and ΔPP are of little value in predicting the effects of fluid expansion on stroke volume or cardiac index. With a ΔVpeak ao threshold value of 12%, 93% of children (positive predictive value) are likely to respond to VE i. e. show at least a 15% increase in stroke volume.

The good discriminative value of ΔVpeak ao observed here is in agreement with two previous adult studies [9, 10]. Feissel et al. showed that ΔVpeak ao, measured in septic shock patients and displayed at the aortic annulus level via transesophageal echocardiography, was significantly higher in R than in NR. A ΔVpeak ao threshold value of 12% was found to reliably discriminate between R and NR [9]. Similarly, Monnet et al. arrived at a threshold value of 13% (mean area under the ROC curve 0.82) using an esophageal Doppler probe [10]. In our study, however, four patients were misclassified in terms of response to fluid expansion by ΔVpeak ao. One of them exhibited high ΔVpeak ao (13.6%): although he did not respond to VE in terms of stroke volume (+12.5%), he showed a 20% mean arterial pressure augmentation (false positive). The three others showed a significant increase in stroke volume or cardiac output but exhibited ΔVpeak ao lower than 12% (9%, 10% and 6% respectively; false negatives) and did not show significant arterial pressure augmentation.

In contrast to the main finding, respiratory variations of systolic or pulse pressure (i. e. ΔPS or ΔPP) were unable to discriminate R and NR regarding fluid responsiveness to volume expansion. Yet, the children in our study showed a mean exhaled tidal volume of 7.4 ml.kg. This negative finding conflicts with those from three earlier adult studies [11–13]. It is consistent, however, with two more recent adult reports emphasizing that low tidal volume (i. e. below 8 ml.kg of ideal body weight) through its effect on pulse pressure or left ventricular stroke volume variations negatively influences accurate prediction of fluid responsiveness [14, 15]. Recently, Charron et al. demonstrated that ΔVTIao values are less affected by inspiratory tidal volume than are ΔPP values [16].

As another finding, we demonstrated a strong relationship between ΔVpeak ao value before VE and VE-induced gain in stroke volume. Interestingly, in a hemorrhagic shock rabbit model, Slama et al. demonstrated that respiratory variation in the aortic blood flow, measured by transesophageal Doppler, was a relevant indicator of blood loss volume [17, 18].

We assumed that cyclic changes in left ventricular stroke volume during the respiratory cycle reflected biventricular preload dependency and hence did so when LV was operating on the steep portion of the Frank–Starling curve [4]. We believe that at least two physiological factors may contribute to the low predictive value of ΔPS or ΔPP found in the infants in our study.

First, on the basis of the windkessel model, arterial pulse pressure in adults appears to be proportional to left ventricular stroke volume but inversely related to arterial compliance [19]. Therefore, we may assume that systolic or pulse pressure variation in our responder group may be limited by the higher arterial elastic properties observed in children [20]. Secondly, Papastamelos et al. have shown that chest wall stiffness in the second year of life will increase to the point that the chest wall and lung are nearly equally compliant, as in adulthood [21]. Therefore, in addition to the negative small tidal volume effect, the imbalance between chest wall and lung elastance in children under 2 years may contribute to attenuating the cyclic change of pleural and transpulmonary transmitted pressure. In open-chest conditions during human cardiac surgery, variations in arterial systolic or pulse pressure and in stroke volume were less pronounced for the same tidal volume [22, 23]. Interestingly, in the only dog model of hemorrhagic shock which clearly demonstrated a tight correlation between blood loss and ΔPP, the authors reduced chest wall compliance by means of an inflatable vest in order to mimic human chest–lung elastance ratio [24, 25]. Discrepancies between discriminative values of ΔVpeak ao and ΔPS or PP during the respiratory cycle nevertheless remain unclear. Still, they would seem to suggest a minimal effect of respiratory conditions previously developed – and more specifically of small tidal volume as suggested elsewhere [26]. We speculate that the Doppler study of respiratory variations in pulmonary venous flow may help to distinguish between cyclic left ventricular preload increase (resulting from a boosting of blood from the capillary bed via increased alveolar pressure) and cyclic effect of changes in extramural aortic pressure [27].

Our study has some limitations. First, the beat-to-beat changes in the amplitude of the aortic pulse peak waves were assumed to be equivalent to the beat-to-beat changes in stroke volume shown by children with normal cardiac function. However, ΔVpeak ao is only influenced by changes in preload, when other parameters (i. e. heart rate, annulus aortic diameter) are quite stable in the short term. Secondly, online measurement of pulse pressure and systolic pressure variations without subsequent offline validation may be open to criticism, given that ΔPP calculation could be markedly affected by small errors in manual pressure measurements [14, 26]. However, our methodology is in agreement with clinical practice regarding the need for emergency fluid expansion and routine monitoring condition. Thirdly, technical artifacts in recording aortic blood flow variation with transthoracic Doppler probe during the ventilator cycle (chest wall motion) could not be completely excluded. In this light we would like to emphasize that the transesophageal route is less affected by the respiratory cycle [9, 10].

In conclusion, the respiratory variations in aortic blood flow velocity measured by pulsed Doppler before VE accurately predict the effects of fluid expansion, whereas ΔPS and ΔPP are of little value in ventilated mechanically children.