Weaning from mechanical ventilation (MV) is a complex step during the recovery from critical illness, and defined as either the failure of spontaneous breathing trial (SBT) or, in the case of extubation, the need for reintubation, the institution of non invasive ventilation (NIV) or the occurrence of death during the first 48 h after extubation.1 Weaning failure is associated with significant mortality and morbidity in patients admitted to intensive care unit (ICU),1–4 and with substantial financial and economic burden.1,5 When evaluating the readiness for extubation, clinicians integrates several parameters; among others, patient age and comorbidities, indices of respiratory support and gas exchange, laboratory findings, and echocardiographic parameters seems significant predictors.6–11

The most common mechanisms of weaning failure are identified in the heart and lung interaction,12 even if more evidence is available on the contribution of diaphragmatic dysfunction,13 though it also may be worth checking for extra-diaphragmatic respiratory muscle activation/athropy and for abdominal causes of failure (i.e. ascites).14 Nonetheless, in the largest study on the topic, weaning-induced pulmonary edema, accounted for almost 60% of failures; in this cohort, chronic obstructive pulmonary disease, previous “structural” cardiopathy and obesity were independent risk factors.

Echocardiography provides a real-time assessment of cardiac dysfunction. Its use is continuously expanding in the ICU, and it can be a valuable assessment tool during the process of weaning.16,17 A previous metanalysis18 showed that parameters referring to the left ventricular (LV) diastolic dysfunction (DD) were associated with a greater risk of weaning failure. Conversely, the contribution of systolic dysfunction evaluated by LV ejection fraction (LVEF) was more unclear (standardized mean difference, SMD: −0.86 [−1.92;0.20]; p = 0.11, high heterogeneity). Such results were possibly influenced by the sample size. Considering the uncertainty of evidence, we conducted an updated systematic review and meta-analysis to re-evaluate the association between LV systolic dysfunction and the risk of failing the weaning from MV.

Our systematic review and meta-analysis was registered in PROSPERO, the international prospective register of systematic reviews (CRD42019117832). We regularly recorded the decision to proceed with an update. The metanalysis is reported in accordance with the PRISMA guidelines.19

The update literature search identified a further 1810 titles between MEDLINE/PubMed and EMBASE. A modified PRISMA flowchart and the PRISMA checklist are reported as Supplementary material (Appendix B). After removing articles deemed not pertinent, 12 full-text were assessed, with eight excluded as not reporting data and one as the approach for SBT was not reported. Hence, in the present metanalysis we pooled together the previously included eleven studies26–36 with the addition of data from three new findings.37–39 Of note, regarding the study by Goudelin et al., we gathered the echocardiography data from a letter that the authors subsequently published.40 The previous metanalysis could not include a study from Lamia et al. as the authors did not explicitly report echocardiographic findings according to weaning failure or success, and it was not possible to retrieve such data of interest.41

All studies assessed heart function with transthoracic echocardiography, and as shown in Table 1, data extracted from each study included: setting of critically ill patients enrolled, number of patients examined, mode of SBT and criteria for SBT failure, echocardiography parameters evaluated and severity scores. The most commonly reported echocardiographic parameter was the E/e’ ratio (n = 13), followed by LVEF and E/A ratio (n = 12 for both), E wave (n = 11) and TDI e’ wave (n = 10). Regarding the LV diastolic function, only two studies reported measure of left atrial area,36,37 and two reported tricuspid regurgitant jet velocity.37,38 In the previous metanalysis,18 we found not enough data to analyse the impact of RV dysfunction in the setting of weaning from MV. In this update metanalysis, we pooled data from three studies reporting the tricuspid annular plane systolic excursion (TAPSE),30,38,39 but not other data on RV function were ameanable of meaningful analysis.

The methodological quality of the included studies performed with the NOS showed that seven studies had the maximum score (9 points),26–29,37–39 six scored 8 points30–35 and one scored 7 points36; thus all studies were judged at low risk of bias. Also the two studies using a longer time-frame for reintubation and used for sensitivity analysis had low risk of bias.15,42 The assessment of the risk of bias for each study is provided separately as Supplementary material (Appendix B). In Table 2 we provide a comparison of the results between the previous and the updated metanalysis, reporting also the sample size and the weaning failure rate.

Parameters describing LV systolic function

Among the included studies, we found LVEF data on 745 patients from 12 studies.26–29,31–37,39 Four studies approached the SBT in PSV and eight in T-tube trial. The overall weaning failure was 33.4% (n = 249/745). The weaning failure rate was 36.7% (n = 194/528) with T-tube trial and 25.3% (n = 55/217) with PSV. Weaning failure was significantly associated with lower LVEF: MD −4.71 95%CI [−9.18, −0.23]; p = 0.04, Fig. 1), with high heterogeneity (I2 = 90%, p < 0.0001). There were no subgroup differences according to the type of SBT (p = 0.92, I2 = 0%). Two further subgroup analyses were conducted (Figures in Supplementary material (Appendix B)). When dividing the twelve studies according to the mean LVEF in the upper and lower half, we found no subgroup differences (p = 0.91, I2 = 0%); similarly, when separating subgroups according to studies with a mean LVEF > or <50% (n = 8 and n = 4, respectively), we found no subgroup differences (p = 0.55, I2 = 0%).

Figure 1.

Forest plot comparing values of left ventricular ejection fraction between critically ill patients experiencing weaning failures vs successes. Studies are analysed in sub-groups according to the modality of spontaneous breathing trial.

CI: confidence interval; df: degrees of freedom; IV: inverse variance, SD: standard deviation.

The TSA performed to explore the robustness of the analysis regarding the impact of LVEF on the outcome of weaning from MV and computed with a MD of −4.71% showed that more research is needed. In fact, the Z-curve did not cross the adjusted boundaries nor reached the information size required (n = 745/1378, Fig. 2). Grade of evidence according to the current recommendations resulted in a low certainty of evidence (Supplementary material (Appendix B)).

Figure 2.

Trial sequential analysis on the impact of left ventricular ejection fraction (LVEF) on the outcome of weaning from mechanical ventilation. The Z-curve describes the results after pooling data on 745 patients. The curve crosses the 1.96 × standard deviation, suggesting a significant result; however it also lays in the area between the adjusted boundaries (curved line on the top) and the futility (triangular) area, meaning that results are not robust.

MD, mean difference; RE, random effect.

Parameters for the evaluation of LV diastolic function

Data on E/e’ ratio were gathered from 857 patients included in 13 studies,26–34,36–39 six in the PSV and seven in the T-tube trial subgroups. One study38 reported the values of E/e’ for both septal and lateral annulus; the analysis is presented with data from the average of their mean and SD. As shown in Fig. 3, weaning failure was confirmed as significantly associated with higher E/e’ ratio: MD 1.90, 95%CI [0.48, 3.32]; p = 0.009), with high heterogeneity (I2 = 84%, p < 0.0001), and no subgroup differences according to the type of SBT (p = 0.68, I2 = 0%). The subgroup analysis performed according to the regional criteria of TDI sampling showed significant differences between subgroups (p = 0.02, I2 = 75.3%; Supplementary material (Appendix B)), with greater association for the average E/e’ values and weaning failure. Analyses performed with either the septal or the lateral E/e’ values from Song et al.38 did not produce statistical changes.

Figure 3.

Forest plot comparing values of E/e’ ratio between critically ill patients experiencing weaning failures vs successes. Studies are analysed in sub-groups according to the modality of spontaneous breathing trial.

CI: confidence interval; df: degrees of freedom; IV: inverse variance; SD: standard deviation.

Data on TDI e’ wave were available from 585 patients includeed in 10 studies,26,28–30,33,34,36–39 five in the PSV and five in the T-tube trial subgroups. As above, Song et al.38 reported both septal and lateral e’ values; the analysis is presented with data from average of their mean and SD. Weaning failure was not confirmed as associated with lower e’ wave values: MD −0.59, 95%CI [−1.51, 0.32]; p = 0.20, Fig. 4), with moderate heterogeneity (I2 = 64%, p = 0.003). There were no subgroup differences according to the type of SBT (p = 0.67), with no heterogeneity (I2 = 0%). The subgroup analysis performed according to the regional criteria of TDI sampling showed significant differences between subgroups (p = 0.03, I2 = 72.2%), with greater association for the average e’ wave values and weaning failure (Supplementary material (Appendix B)). Analyses performed with the septal or the lateral e’ values from Song et al.38 did not affect the above-mentioned statistical findings.

Figure 4.

Forest plot comparing values of e’ wave between critically ill patients experiencing weaning failures vs successes. Analysis is performed in sub-groups according to the modality of spontaneous breathing trial.

CI: confidence interval; df: degrees of freedom; IV: inverse variance; SD: standard deviation.

The E wave data were available on 696 patients from 11 studies,26–30,33,34,36–39 five in the PSV and six in the T-tube trial subgroups. Weaning failure was significantly associated with higher E wave velocity: MD 8.32, 95%CI [2.59, 14.05]; p = 0.004), with moderate heterogeneity (I2 = 54%, p = 0.02). There were no subgroup differences according to the type of SBT (p = 0.29, I2 = 11.5%; Supplementary material (Appendix B)).

Data on E/A ratio were reported by 12 studies including 829 patients,26,29–37,39 five in the PSV and seven in the T-tube trial subgroups. Weaning failure was confirmed as not significantly associated with E/A ratio: MD −0.01, 95%CI [−0.13, 0.11]; p = 0.85), with mild heterogeneity (I2 = 47%, p = 0.04). There were no subgroup differences according to the type of SBT (p = 0.70, I2 = 0%; Supplementary material (Appendix B)).

The data on deceleration time of the E wave (DTE) were available from 661 patients from 9 studies,26,28–31,33,35–37 three in the PSV and six in the T-tube trial subgroups. Weaning failure was confirmed as significantly associated with lower DTE: MD −12.01, 95%CI [−21.13,−2.89]; p = 0.01), with no heterogeneity (I2 = 0%, p = 0.46). There were no subgroup differences according to the type of SBT (p = 0.99, I2 = 0%; Supplementary material (Appendix B)).

We conducted this pre-planned updated metanalysis primarily to explore the association of weaning failure from MV with LVEF, the only parameter describing LV systolic dysfunction consistently investigated and reported in the previous metanalysis.18 That analysis resulted in an unclear link between LVEF and weaning failure (p = 0.11), with no evaluation of its robustness with TSA. Moreover, the effect was statistically described by standardized means (instead of MD), with no signal on the clinical relevance and magnitude of the difference. The present update was pre-planned (registration in PROSPERO database); despite we retrieved only three further studies in the updated search (with only two providing usable data on LVEF), it increased the sample size of data ranging between 20% and 30%. We found that lower LVEF was significantly associated with weaning failure; however, this result should be interpreted cautiously for at least three reasons. First, in term of magnitude of effect, the MD in LVEF between groups was lower than 5%, a value well-below the known intra- and inter-observer variability.43,44 The second reason is linked to the statistical fragility of our finding as in half of the sensitivity analyses conducted removing one study at time the association with weaning failure was lost; moreover, the TSA suggested that the result is not robust and more research would be needed before drawing firm conclusions. A third reason for suggesting caution is that lower LVEF could partially act as a surrogate for altered diastolic function and increased filling pressure. Indeed, according to the most recent guidelines,45 any patient with depressed LVEF is diagnosed by definition with LVDD. Moreover, it is becoming increasingly apparent the “U-shaped” effects of LVEF on clinical outcomes in the ICU setting,46 with negative impact on the outcome of critically ill patients also from hyperdynamic LVEF,47,48 a finding confirmed by a recent metanalysis.49 However, whether a hyperdynamic LVEF has negative effects at the time of weaning from MV has not been adequately investigated.

Other parameters describing the LV systolic function (i.e. the s’ TDI wave) could be valuable during the weaning process; differently from the LVEF, s’ TDI wave has a more linear association with outcome,48 although its predictive value remains controversial.50 In our metanalysis the s’ TDI wave was described by one study only.29 The mitral annular plane systolic excursion (MAPSE) is an interesting echocardiographic parameter in the critically ill patients, at least for its simplicity and reproducibility.51 MAPSE describes longitudinal LV systolic function but it was reported only by one study in the context of weaning38 and therefore analysis was not feasible. Similarly, strain echocardiography, which seems associated with prognosis in septic patients,52 has not been yet adequately explored in the setting of weaning.

The shift to negative pressure ventilation during weaning produces both an increase in venous return (which may cause issues in patients with LVDD and higher LV filling pressures),53 and an increase in LV afterload, especially if significant inspiratory efforts are made.54 In this update, we confirmed the association of higher E/e’ ratio with weaning failure. Apart from being physiologically sound, a MD between groups of almost 2 points could be a clinically relevant difference. Moreover, the inter- and intra-individual variability for this parameter is relatively low.55,56 Notably, E/e’ ratio is the most used surrogate marker of increased LV filling pressure. Cut-off values of E/e’ suggested by the guidelines refer to the outpatient setting,57 and much lower values (7.5–9) are associated with increased LV filling pressures in patients supported by MV.58,59 Conversely, we found no association of e’ velocity values and outcome of weaning in this updated meta-analysis, contraddicting our previous findings.18 Nonetheless, previous literature suggests the importance of e’ velcoity on the outcome of critically ill patients with sepsis.60 With regard to weaning, Moschietto et al. showed that improvements in e’ during the course of SBT were positively associated with weaning success, suggesting the importance of a LV relaxation reserve during weaning. Further, a significant association of lower e’ values and weaning failure was seen in our sub-group analyses, and particularly according to sampling site (average e’) and SBT mode (PSV). In light of these arguments and considering the association of lower values of e’ velocity with outcome in septic patients,60 the present metanalysis suggests the fragility of results not only regarding LVEF but also of e’ velocity. The link between LVDD and worse outcomes in weaning remains highly likely, considering also the results of E wave velocity and DTE, both pointing towards an association of weaning failure and worse LV relaxation.53 The increase of E wave velocity during the SBT reflects the increased venous return with the shift to negative pressure breathing. From a clinical standpoint, the management of these patients during weaning shuold be tailored with particular attentions on prevention of arrhythmias and tachycardia, reduction of afterload, avoidance of positive fluid balance whenever possible and early initiation of fluid removal.61

Apart lack of enough data on s’ TDI wave and on strain echocardiography, we found inadequate reporting in the context of weaning of left atrial size and tricuspid regurgitation jet velocity (variables currently recommended for the evaluation of LV diastolic function57). Regarding the influence of RV function on weaning outcome we found another gap in the literature, though we were able to pool data from a very limited number of studies regardig the TAPSE and the RV-LVEDA. In both cases the results suggested not differences between groups.

Finally, results were consistent across SBT appraoches performed with either T-tube or PSV. Some authors suggested that T-tube trial could be a more challenging test for the cardiorespiratory function,62 but a meta-analysis showed that both approaches are comparable in predicting the outcomes of weaning.63 Moreover, a recent large randomized controlled trial suggested that, in patients at high risk of extubation failure, SBT performed with PSV do not improve outcomes as compared to T-piece.64

Our meta-analysis has several limitations. The major one is exploring just associations between one echocardiographic variable and the weaning outcome, with no adjustment for confounders by regression/multivariate analyses. Hence, we are not able to explore the clinical value of combining echocardiographic parameters as in the pioneer study by Lamia et al. where the combination of E/A and E/e’ ratio allowed an accurate noninvasive detection of weaning-induced pulmonary oedema in patients with difficult weaning.41 Further, we could not explore the prognostic value of a more holistic evaluation, where echocardiography is added to other ultrasound imaging (i.e. lung and/or diaphragm) or labaraturistic findings (i.e. troponin). In this regard, when exploring the outcome of weaning with echocardiography only, a more adequate approach would be probably to focus on the occurrence of weaning-induced pulmonary oedema (as studied by Liu et al.),12 as compared to using as outcome non-specific weaning failure, where non-cardiac factors may act as confounders. However, in this study the authors did not find an association between this phenomenon and concomitant myocardial ischemia.12 Similarly, Bedet et al. found no association of weaning-induced cardiac ischemia with weaning-induced pulmonary edema nor with weaning outcomes.15

From a clinical standpoint, our results suggest particular attention to the weaning in patients with increased LV filling pressures and possibly in those with abnormal LVEF, though a proper interpretation could be made only with an individual pateints data metanalysis. Such issue is common to the other meta-analyses conducted in the field of critically care echocardiography,60,65,66 mainly due to large heterogeneity in data reporting.67 Second, we evaluated the association between echocardiography data performed before the SBT is started with the weaning outcome. However, it could be very valuable to explore changes of echocardiography parameter from their baseline during the course of weaning attempt. In this regards, Moschietto et al. showed that increase in e’ during SBT could be a positive predictor of weaning success.28 Goudelin et al. showed also that an echocardiography-guided treatment enabled to successfully extubate all the patients that previously failed the SBT.37 A third limitation is that the included studies enrolled critically ill patients with different pathologies with significant clinical heterogeneity, spamming from those with heart failure and patients with normal premorbid heart performance. Forth, the included studies did not consistently report hemodynamic data at the time of echocardiography, nor described if any patient had developed LV outflow tract obstruction or experienced worsening of mitral regurgitation. Clearly, patients developing any of these two, would have experienced an increase in LA pressures facilitating the occurrence of pulmonary oedema, and further reducing the cardiac output. All these factors are confounders that add complexity to the topic. Moreover, another confounding effect may be introduced by “non-cardiac” causes of weaning failure (i.e. respiratory and/or diaphragmatic), though the cardiac function is the commonest cause of weaning failure12. Finally, statistical heterogeneity was moderate to high in most of our analyses.

The association between failure to wean from MV and worse LV systolic performance as evaluated by LVEF is possible, but such result seems very fragile for several reasons. The difference in LVEF between failures and successes is well-below a clinically relevant inter- and intra-observer variability, and results are not robust. Weaning failure confirms its association with elevated LV filling pressures, and in general with parameters indicating worse LV diastolic function, though the role of e’ is now questioned. There is still a gap in the literature regarding weaning failure and other echocardiographic parameters describing LV performance (i.e. s’ TDI wave, MAPSE and longitudinal strain), as well as with those evaluating RV function.

• •

  Conception and design: CS, AN, EB, SS, AM, FS

• •

  Screening of automated search: FS, AP, CS

• •

  Consensus on discrepancies: AN, AM

• •

  Independent manual search: CS, SM, MZ, FS

• •

  Acquisition of data and/or data cross-check: FS, AP, AN, CS, MZ

• •

  Analysis: CS, SM, AN, FS

• •

  Risk of Bias Assessment: CS, AP, MZ, FS

• •

  Interpretation of data: FS, AN, CS, AM, EB, SS, MZ

• •

  Drafting the article: CS, MZ, FS

• •

  Critically Revising draft for important intellectual content: AM, EB, SS, AN, SM, AP

All Authors.

No funding was received for this study.

AI program Resub.app was used to check adherence to journal formatting.