Introduction

Fluid administration is one of the first-line therapy interventions used to reverse tissue hypoperfusion during acute circulatory failure. Nevertheless, fluid administration is not free of adverse effects, especially when fluids are excessively administered. Dynamic assessment of preload responsiveness appraising heart–lung interactions is commonly used during the resuscitation of mechanically ventilated patients with acute circulatory failure. In this scenario, assessment of fluid responsiveness might limit fluid administration, potentially reducing the risk of fluid overload, avoiding complications derived from tissue oedema and increasing mechanical ventilation-free days, among others [1].

Several predictors of fluid responsiveness have been described in the medical literature [2]. Dynamic indices evaluating the response of the cardio-circulatory system to reversible preload variations might be grouped based on the way in which preload variation is assessed [3]: (a) first, indices based on mechanical ventilation-induced variations of stroke volume and stroke volume-derived/related parameters, such as pulse pressure variation (PPV), stroke volume variation (SVV), tidal volume challenge (VtC); (b) second, indices based on mechanical ventilation-induced variations of non-stroke volume-derived parameters such as the inferior vena cava respiratory variability (Δ-IVC); (c) third, indices based on preload-redistributing manoeuvers different from standard mechanical ventilation such as passive leg raising (PLR), end-expiratory occlusion test (EEOT), and mini-fluid challenge (m-FC). Indices from the first and second groups are, in principle, limited by the use of low tidal volumes [4, 5], high respiratory rates [6], low pulmonary compliance [7], and low driving pressures [8]. Conversely, indices from the third group could theoretically have better operative performances in most situations commonly observed in critically ill patients [7].

Several meta-analyses evaluating the operative performance of fluid responsiveness predictors in different clinical settings have led to variable results [9,10,11,12,13,14,15,16,17,18,19,20]. These meta-analyses, however, did not evaluate specific subgroups, and there are no meta-regressions assessing the reliability of methods to evaluate fluid responsiveness. Consequently, we sought to conduct a meta-analysis in order to analyse the operative performance of dynamic predictors of fluid responsiveness in critically ill adults mechanically ventilated at Vt ≤ 8 ml kg−1 without arrhythmias and increased respiratory effort. Additionally, we aim to identify clinical variables or methods affecting the operative performance of dynamic predictors of fluid responsiveness under such particular conditions.

Methodology

Protocol

This systematic review and meta-analysis was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) guidelines [21] and was recorded at PROSPERO (registration number CRD42019138147 (https://www.crd.york.ac.uk/prospero/display_record.php?ID=CRD42019138147)) on August 12, 2020.

Study selection and inclusion criteria

Studies prospectively evaluating the operative performance of PPV, SVV, VtC PLR, EEOT, m-FC, and Δ-IVC as predictors of fluid responsiveness in critically ill ventilated patients at Vt ≤ 8 ml kg1 and without respiratory effort and arrhythmias were selected for full-text reading. In addition, studies including subgroups of patients fulfilling our inclusion criteria were also selected and included for the analysis. No language restriction was applied. Only studies recording data about the operative performance of any fluid responsiveness test and including an explicit definition of fluid responsiveness after fluid loading were finally incorporated for the analysis. Studies conducted in the operating room, case reports, and studies including patients < 18 years old, pregnant women were excluded.

Search strategy, data extraction and quality appraisal

A comprehensive search was conducted in the MEDLINE and Embase databases, between January 1999 and May December 2019. Moreover, reference lists of each initially selected manuscript were manually reviewed searching for potential studies not retrieved by the original search. The complete search strategy and the terms used are available in the protocol recorded at PROSPERO. Two reviewers (J.I.A.S. and J.D.C.R.) independently assessed search results for inclusion and undertook data extraction and quality appraisal.

Data items

Data extracted from each clinical trial included: authors, year of publication, number of patients enrolled, type of critically ill patient, age, height; norepinephrine dose, dobutamine, epinephrine, and vasopressin doses; main diagnosis; APACHE (Acute Physiology And Chronic Health Evaluation) II score; SOFA (Sequential Organ Failure Assessment) score; method used to evaluate fluid responsiveness; amount and type of fluids used during the fluid challenge; diagnostic test or fluid responsiveness predictor assessed; definition of fluid responsiveness used; % of response (i.e. cardiac output, VTI, etc.); cut-off point or threshold used to determine fluid responsiveness; tidal volume (Vt); respiratory system compliance; positive end-expiratory pressure (PEEP) level; airway driving pressure; presence of acute respiratory distress syndrome (ARDS); and finally, the sensitivity and specificity, and the area under the ROC curve (AUC) of the diagnostic test used.

Quality assessment

Two authors (JIAS and JDCR) independently assessed the quality of each study by using the QUADAS-2 tool (Quality Assessment of Diagnostic Accuracy Studies) [22]. Disagreements were planned to be solved by consensus between these authors, with the possibility to consult a third author if discrepancies were maintained.

Statistical analysis

Analysis of individual studies

Data regarding sensitivity, specificity, and diagnostic odds ratio (DOR) were calculated by using a contingency table. In some trials, prediction of fluid responsiveness was assessed by using different ventilation parameters or different thresholds, which resulted in multiple data about operative performances; in such cases, all data regarding operative performances were included for analysis.

Analysis of summary measures

Fitted sensitivity, specificity, and AUC data were assessed through bivariate and hierarchical analyses. The summary of receiver operating characteristic (ROC) curves was assessed by using the method of Rutter and Gatsonis [23]. Operative performance quality was graduated according to Fisher et al. [24]. Heterogeneity among trials was assessed using the Cochran’s Q tests and its effect was quantified by calculating the inconsistency (I2). An I2 > 50% was considered significant [25].

Analysis of risk of bias across studies

Asymmetry was assessed by the Thompson and Sharp test. Nevertheless, this was not applicable for PLR, Δ-IVC, VtC, and m-FC because the low number of studies addressing these predictors impedes the application of such test. Publication bias was fitted using the trim-and-fill method.

Additional analysis

Subgroup and meta-regression analyses were performed for all the clinical and physiological variables potentially influencing the operative performance of fluid responsiveness predictors: tidal volume, PEEP, driving pressure, compliance of the respiratory system, type of patient, method used to calculate the index, threshold used to predict fluid response, volume of fluid finally administered. This analysis was also used to determine the source of heterogeneity among studies.

A sensitivity analysis was carried out by performing a meta-regression based on the methodological quality of included studies (QUADAS-2). The threshold effect was assessed using Spearman´s rank correlation coefficient and the Moses–Shapiro–Littenberg method. Data analysis was performed using R software, version 3.4.3, together with the mada and meta packages. Data are expressed as a value (95% confidence interval (CI)), and p < 0.05 was considered statistically significant.

Results

A total of 644 studies were retrieved, including 612 from the MEDLINE and Embase databases, and 32 obtained from the reference lists of the studies retrieved from the original search. Finally, 33 studies fulfilling all the inclusion criteria were included for the quantitative analysis (Fig. 1).

Fig. 1
figure 1

Study selection

Full size image

General characteristics of the studies included

A total of 33 studies involving 1352 patients were included for analysis. General characteristics of studies included are summarized in Tables 1 and 2. A total of 1413 fluid challenges were performed with an average fluid responsiveness of 53.06%.

Table 1 General characteristics of selected studies
Full size table
Table 2 Operative performance of predictors of fluid responsiveness in mechanically ventilated patients at Vt ≤ 8 ml/kg without arrhythmia and respiratory effort
Full size table

Risk of bias

The risk of bias of the included studies is summarized in Additional file 1: Table S1.

Syntheses of results

Operative performance of fluid responsiveness predictors is shown in Table 3. Receiving operator (ROC) curves for the three groups of predictors are presented in Figs. 2, 3 and 4. Moderate heterogeneity was found among studies assessing PPV (see Additional file 2: Figure S1), SVV (Additional file 3: Figure S2), PLR (Additional file 4: Figure S3, and EEOT (Additional file 5: Figure S4). Conversely, heterogeneity was not found among studies that assessed the other predictors (see Additional file 6: Figures S5, Additional file 7: Figure S6 and Additional file 8: Figure S7).

Table 3 Operative performance of predictors of fluid responsiveness
Full size table
Fig. 2
figure 2

Summary ROC curve for the first group of predictors of fluid responsiveness. SVV, stroke volume variation; PPV, pulse pressure variation; VtC, tidal volume challenge. Closed curve: 95% confidence region

Full size image
Fig. 3
figure 3

Summary ROC curve for the second group of predictors of fluid responsiveness. IVC, inferior vena cava respiratory variability. Closed curve: 95% confidence region

Full size image
Fig. 4
figure 4

Summary ROC curve for the third group of predictors of fluid responsiveness. EEOT, end-expiratory occlusion test; m-FC, mini-fluid challenge; PLR, passive leg raising. Closed curve: 95% confidence region

Full size image

Risk of bias across studies

Asymmetry was present among studies assessing PPV (p = 0.02), SVV (p = 0.04), and EEOT (p < 0.03), and it was caused by publication bias (see Additional file 9: Figures S8, Additional file 10: Figure S9 and Additional file 11: Figure S10). Meanwhile, asymmetry was not performed for other predictors due to the low number of studies evaluating them.

Asymmetry among studies on PPV was fitted by using the trim-and-fill method, improving heterogeneity (I2 = 37.3%; p = 0.02), and the DOR obtained using the random effects model was decreased (DOR = 6.68; 95% CI 3.85–11.58). On the other hand, when the asymmetry of studies that assessed SVV was fitted, DOR by random effects also decreased (DOR = 11.3; 95% CI 4.34–29.66), but there were no changes in the heterogeneity (I2 = 73.1%; p < 0.001). Finally, when asymmetry among studies that assessed EEOT was fitted, both DOR by random effects (DOR = 12.93; 95% CI 5.31–31.50) and heterogeneity decreased (I2 = 29%; p = 0.13).

Additional analysis

Subgroup and meta-regression analyses attaining statistical significance are shown in Table 4. Operative performance of PPV was affected by the method used to calculate cardiac output (p = 0.02) and by the compliance of the respiratory system (p = 0.05) (Fig. 5). Additionally, these variables were a source of heterogeneity (p < 0.05).

Table 4 Subgroups and meta-regression analysis
Full size table
Fig. 5
figure 5

Bubble plot for meta-regression of pulse pressure variation with lung compliance pulmonary as a continuous covariate. The number of the point is the reference number of each study

Full size image

Operative performance of SVV was affected by the method to calculate cardiac output (p = 0.01), the threshold selected to define positive fluid responsiveness (p = 0.05), the type of critically ill patient (p < 0.001), and the volume of fluid finally used during the fluid challenge (p = 0.01). These subgroups were a source of heterogeneity since they disappeared among studies (I2 < 25%, p > 0.05).

Additionally, subgroup analysis of studies assessing PLR showed that volume of fluids infused to determine variation in cardiac output, significantly affected its operative performance (p < 0.01), and it was a source of heterogeneity since it disappeared among studies (p = 0.93). Subgroup and meta-regression analyses of the remaining predictors did not show any change in their operative performance or heterogeneity (Additional file 12: Table S2).

According to the sensitivity analysis based on the methodological quality of the included studies (QUADAS-2), there were no changes in the operative performance of PPV (p = 0.39), SVV (p = 0.23) and EEOT (p = 0.15) (see Additional file 12: Table S2). It should be noted that this analysis was not performed for other predictors due to the low number of studies evaluating them. According to the rho correlation coefficient or the Moses–Shapiro–Littenberg test, there was no threshold effect for any of the predictors (p > 0.05).

Discussion

This systematic review and meta-analysis reveal that VtC, EEOT, and SVV have excellent operative performance, while ∆-IVC, PLR, m-FC, and PPV had good operative performance as predictors of fluid responsiveness in critically ill ventilated patients at Vt ≤ 8 ml kg−1 and without respiratory effort and arrhythmias. Methods to calculate cardiac output was important sources of heterogeneity. In addition, as expected, compliance of the respiratory system and type of patient affected the performance of SVV, while the volume of fluids infused to determine variation in cardiac output, significantly affected the performance of SVV and PLR.

Several meta-analyses have evaluated the operative performance of these predictors in different clinical settings [9,10,11,12,13,14,15,16,17,18,19,20]. Differently from this current metanalysis, patients included received Vt from 4.9 to 12 ml kg−1 [9, 10, 17] and evaluated other types of populations [14, 18]. Even though, our data suggest that most of fluid responsiveness predictors have good reliability even in conditions in which such prediction could be assumed that it would not be good.

The VtC and EEOT performances for determining fluid responsiveness were superior. Some studies showed that operative performance of EEOT was not good at Vt < 6 ml kg−1 [49, 56]. Meanwhile, a recent meta-analysis reported an adequate reliability of EEOT in mechanically ventilated patients at Vt ≤ 7 ml kg−1 [57], a finding in agreement with our results. Therefore, EEOT could be used for patients ventilated at any Vt. SVV depicted a better performance than PPV, which may be explained by the fact that PPV depends on effective arterial elastance [58], a variable that summarizes the features of arterial vascular load in humans [59]. We assessed studies that included critically ill patients who could have a low arterial load. Therefore, PPV susceptibility to haemodynamic changes may be increased when a low Vt is used.

Prediction of fluid responsiveness of some indices rely on tidal volume and intrathoracic pressure variations [4, 5]. Interestingly, operative performance of predictors analysed in this current metanalysis were apparently not affected by PEEP levels or driving pressures, which differ from other studies [8, 60] (see Additional file 12: Table S2). Nevertheless, respiratory system compliance directly affected the reliability of PPV (p = 0.05) to predict fluid responsiveness, which suggests that effects of respiratory pressure and tidal volume mainly rely on the degree to which these variables are transmitted to the pulmonary circulation and not on their absolute values [7].

Methods used to classify patients as fluid responders or not responders after the final fluid loading significantly affecting the reliability of PPV and SVV to predict fluid responsiveness. In this regard, operative performance was lower when transpulmonary thermodilution was used (through a PiCCO monitoring system) than when using the conventional thermodilution (through a pulmonary artery catheter) (see Table 4). Thus, more than errors implicit to the cardiac output calculations, classification as responder or non-responder derived from the method to estimate cardiac output was apparently a determinant of the reliability of such predictors. In addition, use of different thresholds to classify patients as fluid responders also influence on their operative performance (p = 0.05).

As expected, lower thresholds might increase operative performances in some cases (see Table 4). Importantly, reliability of SVV also varied depending on the type of critically ill patient (p < 0.01): better performance was found in post-cardiovascular surgery patients and in those with septic shock (DOR = 95.67; p = 0.03, and DOR = 21.23; p < 0.01, respectively), than in post high-risk surgery patients (DOR = 6.70; p = 0.13). We hypothesized that this finding represents a higher proportion of abdominal hypertension cases in the last group of patients since this might be a common complication in the postoperative period [61]. The presence of intraabdominal hypertension decreases thoracic compliance, resulting in increased SVV values regardless of preload dependency [62] and reduced operative performance. Finally, volume of a fluid loading with which fluid responsiveness was finally determined, significantly influenced the reliability of SVV and PLR. Nevertheless, these findings should be taken with caution, and we think that they should be considered as a source of heterogeneity.

An important point to retain is that positive fluid responsiveness should not systematically lead to fluid administration. Indeed, only during circulatory failure accompanied by altered tissue perfusion status, fluid administration should be considered aiming to increase cardiac output assuming this will revert tissue hypoperfusion and will restore normal cell respiration. Benefit of increasing cardiac output by volume expansion in positive fluid responders should be always balanced with the risk of fluid overload, which may be harmful.

This meta-analysis had several limitations. First, only adult critically ill ventilated patients with a Vt ≤ 8 ml kg−1 and without respiratory effort and arrhythmias were included, so the findings reported cannot be extrapolated to other clinical settings. Second, some predictors of fluid responsiveness were evaluated by a small number of studies, which limit their analysis. Third, the GRADE system (Grading of Recommendations, Assessment, Development, and Evaluations) was not used to determine or assess the meta-analysis’s quality since it was not established in our protocol. Conversely, we performed a sensitivity analysis based on the methodological quality of the included studies (QUADAS-2).

Fourth, moderate heterogeneity was found for some predictors, so these findings should be interpreted with caution. Nevertheless, other sources conversely decreased heterogeneity, which would allow extrapolation of our findings to clinical practice. Finally, operative performance of fluid responsiveness test was classified according to ROC curve analysis, which does not consider the DOR, a variable that summarizes the relation between sensitivity and specificity; however, in our opinion, DOR should always be considered for measuring operative performance when choosing a predictor of fluid responsiveness.

In conclusion, VtC, EEOT, and SVV have excellent operative performance, while ∆-IVC, PLR, m-FC, and PPV had good operative performance as predictors of fluid responsiveness in our setting. Method to calculate the cardiac output, threshold used to determine fluid responsiveness, volume administered during the fluid loading, and type of patient in which the test has been applied should have in account at moment to use it in clinical practice.