Dynamic arterial elastance as a predictor of arterial pressure response to fluid administration: a validation study

We prospectively included all patients equipped with an indwelling arterial catheter and evaluated by esophageal Doppler monitoring who were receiving a fluid challenge for the presence of clinical signs of acute circulatory failure, including hypotension (defined as a MAP ≤65 mmHg or a systolic arterial pressure (SAP) ≤90 mmHg); requirement for vasopressor drugs, presence of lactic acidosis, urine output ≤0.5 ml∙kg⁻¹∙hr⁻¹ during at least 2 hours, heart rate >100 beats/min and/or the presence of skin mottling. Preload dependency was assessed according to our institutional protocol for hemodynamic resuscitation and defined as a CO increase ≥10% after a 2-minute leg-raising maneuver [17]. In all cases, the final decision to start or continue fluid administration was made by the treating physician.

Patients were under controlled mechanical ventilation with no spontaneous respiratory efforts, as assessed by visual inspection of the airway pressure curve. Patients with unstable cardiac rhythm were excluded, although this condition did not affect the decision to administer fluids.

Patients were monitored with an esophageal Doppler monitoring system (CardioQ-Combi™; Deltex Medical, Chichester, UK). This system combines a standard Doppler monitor with arterial pressure analysis capability. The probe was inserted into the esophagus, preferably using the nasal route, and advanced until the maximal peak velocity of the aortic blood flow was reached. The gain setting was adjusted to obtain the optimum outline of the Doppler waveform. In order to reduce signal noise from heart valves and wall thump artifact, a built-in filter function was activated in some patients and kept unchanged throughout the study.

An arterial pressure transducer (TruWave; Edwards Lifesciences, Irvine, CA, USA) was zeroed to atmospheric pressure, and optimal damping of the arterial waveform was carefully checked by fast-flushing the line. The arterial pressure signal was transferred from the patient’s bedside monitor to the Doppler system using a serial cable and automatically synchronized with the aortic blood flow waveform for analysis (Additional file 1: Figure S1).

Our arterial load assessment was grounded on a basic physiological framework based on a two-element Windkessel model of arterial circulation with static and dynamic components. The static component represents a unique pressure–volume (P–V) relationship and can be described by a resistive element or the systemic vascular resistance (SVR = MAP/CO*80) and a pulsatile element or the net arterial compliance (C = SV/arterial pulse pressure) [18]. This distinction is a product of the oscillatory nature of the arterial flow and the mechanical properties of the arterial system [7]. The effective arterial elastance (Ea = 0.9*SAP/SV) is considered an integrative variable of arterial load that incorporates both steady and pulsatile elements [15],[19],[20]. Dynamic assessment of arterial load was performed by calculating Ea_dyn, which we defined as the ratio between PPV and SVV during a respiratory cycle. Rather than being a steady-state variable, Ea_dyn depicts the slope of the P–V relationship and provides a functional assessment of arterial load [21]. PPV obtained from the arterial line and SVV from the aortic blood flow were simultaneously calculated by using the Doppler monitor software with the standard formulae [22] and averaged over three respiratory cycles (Figure 1). In order to do so, respiratory rate was manually introduced on the Doppler monitor.

All hemodynamic variables were measured prior to and just after VE, consisting of 500 ml of normal saline given within 30 minutes. No changes in ventilatory settings or vasoactive drugs were made during volume administration. No paralyzing agents were used for the study. All hemodynamic variables were automatically recorded every 10 seconds and averaged over 1 minute for statistical purposes.

The normality of the data distribution was tested using the Kolmogorov-Smirnov test. The results are expressed as mean ± standard deviation or as median (25th to 75th interquartile range), as appropriate. Patients were classified as hypotensive if they had a MAP ≤65 mmHg and/or SAP <90 mmHg. Sepsis was defined according to standard criteria [5]. As we selected an increase in CO ≥10% for defining fluid responsiveness, we chose a similar cutoff for MAP changes for defining a positive pressure response. This threshold was selected assuming a matched ventriculoarterial coupling and optimal hydraulic efficiency (maximal stroke work) [23]. The relationships between variables were analyzed using a linear regression analysis. Because changes in arterial pressure depend not only on Ea_dyn but also on the magnitude of CO changes, we also calculated the weighted least-squares regression, taking into consideration the contribution of CO changes in the relationship between arterial pressure increase and preinfusion Ea_dyn. Differences at baseline between pressure responders and non-responders were compared by means of an independent sample t-test, and their evolution over time was assessed by one-way analysis of variance with repeated measurements, using group (pressure responders vs. non-responders) as the between-subjects factor and time (preinfusion vs. postinfusion) as the within-subjects factor. Differences between groups were compared using an independent samples t-test. Comparison between preinfusion and postinfusion periods was tested using a t-test for repeated measurements. Categorical variables were compared using the χ² test. For each arterial load variable, a receiver operating characteristic (ROC) curve was created to test the ability of predicting a positive MAP response after VE. Areas under the ROC curve (AUC) with 95% confidence intervals (95% CIs) were compared by using the method described by DeLong et al. [24]. Optimal cutoff values were calculated by maximizing the Youden index (J = sensitivity + specificity −1). The diagnostic performance of Ea_dyn was also assessed on the basis of its positive and negative likelihood ratios (LHRs). A diagnostic variable with potential to influence clinical decisions is usually considered when its positive LHR is >10, its negative LHR is <0.1 and the AUC is >0.9 [25]. A gray zone for Ea_dyn cutoff was created using a resampling method [26]. In summary, the Youden index for each bootstrapped sample from 1,000 replications of the original study population was calculated, then the median value and the 95% CI of these 1,000 optimal cut-offs were obtained. This bootstrapped 95% CI defines a gray zone around the optimum criterion in which formal conclusions about prediction of MAP response cannot be obtained [25],[26].

A preliminary power analysis determined that a sample size of 48 patients was required for detecting an AUC difference of 0.1 (α = 0.05; β = 0.20; allocation ratio = 1:1).

A P-value <0.05 was considered significant. All statistical analyses were two-tailed and performed using MedCalc statistical software version 14.8.0 [27].

Hemodynamic changes after VE are shown in Table 2. Overall, VE increased CO by 14.7% (13.2% to 19.1%), SV by 20.1% (17.1% to 21.9%) and MAP by 9.8% (4.9 to 10.3%). Thirty-three patients (41%) were classified as pressure responders. Of 32 fluid challenges performed in the 22 hypotensive patients, only 17 showed a MAP increase by ≥10% (53%). The rate of pressure responders was similar between patients with and without sepsis (34% vs. 51%; P = 0.18), or between hypotensive and non-hypotensive patients (53% vs. 33%; P = 0.13). There was a weak relationship between VE-induced changes in MAP and CO (R ² = 0.05; P = 0.04) (Figure 2) and between VE-induced changes in arterial pulse pressure and SV (R ² = 0.13; P <0.001).

Fluid administration decreased Ea from 1.75 (1.54 to 1.91) mmHg/ml to 1.59 (1.45 to 1.81) mmHg/ml (P <0.0001) and SVR from 1,229 ± 554 dyn∙s∙cm⁻⁵ to 1,150 ± 514 dyn∙s∙cm⁻⁵ (P <0.0001). Net arterial compliance did not change. In non-responder patients, VE reduced the static component of arterial load but did not affect Ea_dyn (Table 3).

At baseline, Ea_dyn was higher in pressure responder patients (1.04 ± 0.28 vs. 0.60 + 0.14; P <0.0001) (Figure 3). Pressure responders were also more hypotensive (Table 2). Furthermore, preinfusion Ea_dyn and SVR were higher in hypotensive patients (Ea_dyn: 0.90 ± 0.35 vs. 0.70 ± 0.23, P <0.01; SVR: 1,036 ± 605 dyn∙s∙cm⁻⁵ vs. 1,359 ± 486 dyn∙s∙cm⁻⁵, P <0.05) (Additional file 1: Table S2). No differences were observed in preinfusion arterial load variables between septic and non-septic patients (Additional file 1: Table S3).

Preinfusion Ea_dyn was correlated with VE-induced changes in SAP (R ² = 0.56; P <0.0001), diastolic arterial pressure (DAP; R ² = 0.61; P <0.0001), MAP (R ² = 0.60; P <0.0001) and arterial pulse pressure (R ² = 0.42; P <0.0001) (Figure 4). When adjusted for CO changes, this relationship was higher (SAP: R ² = 0.62; DAP: R ² = 0.70; MAP: R ² = 0.69; pulse pressure: R ² = 0.44; P <0.0001, respectively; Additional file 1: Figures S2 and S3). None of the static components of arterial load were associated with changes in arterial pressure produced by VE. No relationship was observed between Ea_dyn and other variables of arterial load at baseline or after VE.

The AUC for preinfusion Ea_dyn (0.94 ± 0.03; 95% CI: 0.86 to 0.98) was higher than for any other variable of arterial load: Ea (0.53 ± 0.07; 95% CI: 0.42 to 0.65; P <0.0001), C (0.51 ± 0.07; 95% CI: 0.39 to 0.62; P <0.0001); SVR (0.55 ± 0.07; 95% CI: 0.44 to 0.66; P <0.0001) and preinfusion MAP (0.62 ± 0.07; 95% CI: 0.50 to 0.72; P <0.0001) (Figure 5).

At baseline, an Ea_dyn value ≥0.73 predicted an increase ≥10% in MAP after VE with a sensitivity of 90.9% (95% CI: 75.7% to 98.1%) and a specificity of 91.5% (95% CI: 79.6% to 97.6%), with a positive predictive value of 88.2 (95% CI: 72.5% to 96.7%) and a negative predictive value of 93.5 (95% CI: 82.1% to 98.6%). The positive and negative LHRs for Ea_dyn were 10.68 (95% CI: 4.2 to 27.4) and 0.1 (95% CI: 0.03 to 0.3), respectively. The bootstrapped 95% CI defined a gray zone for Ea_dyn ranging between 0.72 and 0.88. Only ten fluid challenges (12.5%) were situated in the inconclusive zone.

When only the first VE per patient was included in the analysis, the predictive performance of Ea_dyn was similar (AUC: 0.92 ± 0.04; 95% CI: 0.81 to 0.98; P = 0.64 vs. including all fluid challenges) (Additional file 1: Figure S4). When we considered an increase of CO and MAP ≥15% for defining a positive preload and pressure response, the AUC of Ea_dyn was also excellent: 0.97 ± 0.03 (95% CI: 0.86 to 0.99; P = 0.37 vs. definition of ≥10% for CO and MAP increases) (Additional file 1: Figure S5). Because the impact of PPV seems to be predominant on the Ea_dyn ratio, we also compared the performance of PPV alone against Ea_dyn (AUC for PPV: 0.76; 95% CI: 0.65 to 0.85; P <0.001 vs. AUC for Ea_dyn) (Additional file 1: Figure S6). The ability of Ea_dyn for predicting pressure responsiveness was similar between hypotensive and non-hypotensive patients, as well as between septic and non-septic patients (Additional file 1: Tables S4 and S5).