Introduction

Though a life-saving intervention, invasive mechanical ventilation (i-MV) is prone to side-effects and complications [1, 2]. The process of weaning patient off i-MV should be started promptly to make the time spent on i-MV the shortest possible [3]. Weaning has been recently defined as the time between the first separation attempt and successful extubation that leads to either 7 days of continuous spontaneous breathing or intensive care unit (ICU) discharge, whichever comes first and irrespective of the use of noninvasive ventilation (NIV) in the post extubation period [4].

NIV applied immediately after extubation has been proposed as a measure to prevent post-extubation respiratory failure (i.e., prophylactic NIV in high-risk patients) or as an alternative to i-MV in patients not yet ready to be extubated (i.e., NIV to facilitate weaning) [5, 6].

In patients with acute-on-chronic respiratory failure, particularly those secondary to chronic obstructive pulmonary disease (COPD) exacerbations, compared to standard weaning with the endotracheal tube in place, early extubation followed by immediate NIV application reduces rates of weaning failure and ventilator associated pneumonia, duration of mechanical ventilation, ICU and hospital length of stay (LOS), and improves the rate of survival compared to standard weaning with the endotracheal tube in place [6, 7]. Recent guidelines provide a conditional recommendation in favor of this therapeutic approach in hypercapnic patients with acute-on-chronic respiratory failure. The guideline authors were unable to make recommendation in patients with non-hypercapnic hypoxemic acute respiratory failure (hARF), because of scarcity of available data [6]. After completion of these guidelines, however, two properly powered studies have been published, which included many more patients than previous investigations [8, 9].

Therefore, we designed this systematic review and individual patient data meta-analysis (IPD) to re-assess, in a population of patients recovering from an episode of non-hypercapnic hARF, whether NIV after early extubation would reduce the duration of i-MV (primary endpoint), overall time spent on mechanical ventilation (i-MV + NIV), rate of ventilator associated pneumonia (VAP), time from randomization to ICU and hospital discharges, and time from randomization to ICU death (secondary endpoints), when compared to conventional weaning with the endotracheal tube in place.

Materials and methods

Search strategy and selection criteria

This systematic review with meta-analysis was conducted in accordance with the Preferred Reporting Items for a Systematic Review and Meta-analysis of Individual Participant Data.

We considered eligible for inclusion all randomized controlled trials (RCTs) comparing early extubation + NIV with standard weaning with the endotracheal tube in place in adult patients with non-hypercapnic (as defined by PaCO2 ≤ 50 mmHg and pH ≥ 7.35) hARF and receiving i-MV for more than 48 h. Patients were excluded in the case of (1) ARF secondary to neurological/ neuromuscular disorders, status asthmaticus, chronic obstructive pulmonary disease (COPD), cardiogenic pulmonary edema; (2) body mass index ≥ 30 kg/m2; (3) tracheostomy; (4) obstructive sleep apnoea.

Two authors (FM/AP), independently, searched EMBASE, Pubmed/Medline and Cochrane Central Register of Controlled Trials (CENTRAL) bibliographic databases, without language restriction. Our search encompassed a period from database inception to the 1st October 2020. We supplemented this search by searching review articles and reference lists of trial publications. Collaborators were asked if they knew of any additional RCTs.

Search term combinations are detailed in the Additional file 1.

On search completion and after removal of duplicates, two authors (FM/AP), with the help of a third author (RV) in case of discrepancies, independently assessed for relevance all titles identified by the search strategy. Following title screening, the same independent review procedure was adopted for screening of abstracts and, finally, full texts.

Data analysis

Data were extracted onto a piloted proforma by two authors (RV/FBA) independently. Extracted data included characteristics of the studies, populations, intervention and comparator, and outcomes. Data were checked for sequence generation, data consistency and completeness and baseline imbalance. IPD were obtained from the authors through a process detailed in the Additional file 2.

RCTs included in quantitative synthesis were evaluated using the Cochrane Risk of Bias assessment tool [10]. The following variables were assessed: sequence generation; allocation concealment; blinding of participants, personnel, and outcome assessors; completeness of outcome data; evidence of selective outcome reporting; and other potential threats to validity. We assessed selectivity of reporting either by comparing study protocols against study reports or by specifically asking study authors whether all prespecified outcomes were reported. Two investigators (FM and AP) independently assessed study quality. Details of the assessment are reported in the Additional files 3 and 4.

Our primary endpoint was to determine whether, in adults receiving i-MV due to non-hypercapnic hARF (population), early extubation followed by immediate NIV application (intervention) compared to standard weaning (comparator), reduces the time spent on i-MV, i.e., days spent on i-MV from randomization to ICU discharge (outcome). Secondary endpoints are summarized in the Additional file 4.

Statistical analysis

We conducted a meta-analysis with one-step and two-step approach, incorporating all available IPD. Only complete case data were included for all trials in the main analyses. Continuous variables were presented in descriptive analyses as mean ± standard deviation (SD), while categorical and binary variables were presented as frequencies (n) and percentages (%), as indicated. Data were analyzed on an intention-to-treat basis. Mixed-effects linear regression models were used to model total days of ventilation and the other continuous outcome variables. Time-to-event outcomes were analyzed through parametric survival models, including random effects considering the cluster effect deriving from different studies. Heterogeneity was assessed within 2-stage models using the I2 statistic. We also performed a leave-one-out sensitivity analysis, alternatively removing one study at a time, to measure how each study affected the overall estimate and to identify studies that potentially drove the results.

All tests were two-sides and performed at the 5% level of statistical significance. Statistical analyses were done using STATA software version 15 (StataCorp).

The protocol was registered to PROSPERO database on 12/06/2019 and available at PROSPERO website inserting the study code i.e., CRD42019133837.

Results

Our search identified 1605 records (486 citations in PubMed/Medline, 591 in EMBASE and 528 in the Cochrane Controlled Register of Trials). Following removal of duplicates (n = 460), 1076 records were excluded for title and 56 in abstract form. Thirteen full text articles were assessed for eligibility. Seven studies were excluded in full text: 2 for PICO reasons i.e., 1 for intervention and 1 for population and 5 as IPD were not available [11,12,13,14,15,16,17]. Six studies were included in the quantitative synthesis [8, 9, 18,19,20,21]. Excluded studies and reasons for exclusion are reported in the Additional file 5. The selection process is summarized in the PRISMA-IPD flow diagram (Fig. 1).

Fig. 1
figure 1

PRISMA-IPD flow diagram. The following flow diagram summarizes the selection process of the randomized control trials included in the systematic review, from the identification to the final phase of data analysis. IPD individual patient data, PICO patient intervention comparison outcome

Full size image

Patients meeting all formal inclusion criteria were available for two RCTs [9, 21], while for the remaining 4 studies only selected patients fulfilling inclusion criteria were included (Table 1) [8, 18,19,20].

Table 1 Characteristics of the randomized control trials included in qualitative synthesis
Full size table

We conducted the quality assessment only for studies contributing to IPD meta-analysis. All the studies were rated as being at low risk of bias for randomization process, allocation concealment and incomplete outcome data (attrition bias). The inability to blind caregivers to treatment allocation meant that all the studies were at high risk of performance bias. The risk of detection bias was overall low; in 3 studies the strategies to blind outcome assessors from group allocation were described [8, 9, 21], in 2 studies we received description after contacting the authors [18, 20] and for one study the risk remained unclear [19]. One study was not registered in advance [19]. In 2 studies the predefined outcomes were not properly reported [18, 20], encompassing the risk of reporting bias (Additional file 3).

Patient characteristics, stratified by randomization group, are summarized in Table 2. We overall included 459 participants, 233 and 226 in the intervention and control group, respectively, mean (SD) age 62 (15) years, 269 (59%) males. The principal causes for instituting i-MV were post-operative ARF and acute respiratory distress syndrome (ARDS). Surgical and medical patients were 203 (44%) and 256 (56%), respectively. Mean risk of predicted in-hospital mortality based on APACHE [22] or SAPSII [23] scores, varied from 12 to 35% for surgical and medical patients.

Table 2 Patient characteristics at ICU admission and ventilator settings and gas exchange at randomization
Full size table

Criteria for readiness to wean and spontaneous breathing trial before randomization are summarized in the Additional file 6. Ventilator settings and arterial blood gas values at randomization and prior to spontaneous breathing trial (SBT) are also displayed in Table 2. Mean positive end-expiratory pressure (PEEP) and pressure support levels were 7 and 11 cmH2O, respectively, in both groups. Noteworthy, PaO2/FiO2 was slightly though significantly different between intervention 242 (58) mmHg and control group 258 (77) mmHg, (p = 0.014).

The primary outcome of the study, i.e., length of i-MV, was available for all 459 patients. The two-stage IPD meta-analysis (Fig. 2a) showed a shorter time of i-MV in the treatment group, compared to the control group (mean difference: − 4.16 days; 95% CI − 7.09 to − 1.22; p = 0.006). The sensitivity analysis based on the leave-one-out method did not substantially modify the results, which remained statistically significant after exclusion of each study, with point estimates ranging between − 2.70 and − 5.06 days.

Fig. 2
figure 2

Results of 2-stage IPD-MA. a Mean i-MV time (p value = 0.006); b mean total ventilation time (p value = 0.13); c occurrence of VAP (p value = 0.03); d time to ICU discharge (p value = 0.09); e time to ICU death (p value = 0.222), f time to hospital discharge (p value = 0.009). CI confidence interval, ICU intensive care unit, i-MV invasive mechanical ventilation, IPD-MA individual patient data meta-analysis, VAP ventilator associated pneumonia

Full size image

The one-stage IPD analysis confirmed the previous results. After adjusting for demographic (age, gender) and severity-related variables (severity scores at ICU admission and PaO2/FiO2 at randomization), the duration of i-MV remained significantly different between treatment and control group (mean difference: − 3.43 days; 95% CI − 5.17 to − 1.69; p < 0.001). Results of random and fixed-effects models did not substantially differ (Table 3).

Table 3 Results from 1-stage IPD-MA, according to different models
Full size table

Results of two-stage IPD meta-analysis for each secondary outcome are reported in Fig. 2 (panels b–f). The overall duration of mechanical ventilation was similar between intervention group and controls (mean difference: − 0.88 days; 95% CI − 2.01 to 0.25; p = 0.130). Time to ICU discharge (time ratio: 0.79; 95% CI 0.60 to 1.04; p = 0.09) and mortality (time ratio of 0.63; 95% CI 0.30 to 1.32; p = 0.222) were also not significantly different between groups, while the time to hospital discharge (time ratio: 0.82; 95% CI 0.71 to 0.95; p = 0.009) and the risk for VAP, (odds ratio: 0.29; 95% CI 0.09 to 0.90; p = 0.03) were reduced in the intervention group, as opposed to controls.

Analyses of secondary outcomes based on one-stage approach are reported in Table 3. After adjusting for demographic (age, gender) and severity-related variables (ratio between partial pressure of oxygen and inspired oxygen fraction at randomization and severity scores at ICU admission), all the results became significantly different, except for time to ICU mortality.

As further analysis, reintubation occurrence resulted similar in the two groups as reported in the Additional file 7.

Discussion

The present IPD meta-analysis shows that, in selected patients recovering from an episode of hypoxemic ARF, early extubation followed by immediate NIV application reduces the duration of i-MV, as opposed to conventional weaning and extubation. Furthermore, compared to standard weaning, early extubation + NIV decreases overall duration of mechanical ventilation, risk of VAP, and time to ICU and hospital discharge.

The study did not identify a significant difference in ICU mortality between the two groups. One possible explanation is that ICU deaths are a relatively rare events (40 cases), leading to an underpowered analysis.

To the best of our knowledge, this is the first IPD meta-analysis of trials investigating the role of NIV in the weaning process of patients recovering from an episode of non-hypercapnic hARF. Previous systematic reviews and meta-analyses addressing the potential of NIV to facilitate weaning [7, 24], considered data on both COPD patients and mixed populations, in the present IPD meta-analysis we analyzed data from 459 patients with non-hypercapnic hARF only, allowing the study to focus on this specific population. By excluding not only hypercapnic patients with COPD or other chronic respiratory disorders, such as neuromuscular disease and obesity-hypoventilation, and those with cardiogenic pulmonary edema, we removed the group of patients who usually show a fast response to NIV.

A recent guideline considers the potential usefulness of NIV in the process of facilitating weaning from i-MV [6]. No recommendation was made for patients with non-hypercapnic hARF due to the paucity of available data. After these guidelines were completed, however, two properly powered studies were published. Both included many more patients than all previous investigations. The first assessed 364 mixed patients, mainly those with non-hypercapnic hARF [8] from 41 ICUs of the UK National Health Service, while the second, 130 non-hypercapnic hypoxemic patients from 9 ICUs, 6 in the Chinese Republic and 3 in Italy [9]. Notably, the results on the time to liberation from i-MV and from any ventilation were largely similar in both cases, showing a shorter duration of i-MV and a similar duration of overall mechanical ventilation, i.e., invasive plus noninvasive. We choose to consider i-MV, rather than the overall duration of mechanical ventilation, as primary endpoint since it has been repeatedly shown to be associated with greater requirement of sedatives, rate of VAP and mortality [7, 25].

Before drawing conclusions, some strengths and limitations of our study require discussion. The major strength is the study design; an IPD meta-analysis is considered to achieve the highest level of evidence and offers several advantages over aggregate patient data meta-analysis [26]. Furthermore, the present work considers only RCTs. If on the one hand our choice excludes observational studies of potential interest, on the other hand it incorporates the studies providing the highest level of evidence. Finally, the amount of missing data was small, and only present for outcomes considered secondary endpoints, in a range from 0 to 3%.

Our meta-analysis has several additional potential limitations. First, we could not include patients from 5 of the identified studies (one of which was available only in abstract form [11]) as in 2 cases we could not reach the authors, while in the other 3 cases datasets were not available [11, 1314, 16, 17]. Second, the study protocols of the included studies were not identical, as NIV after early extubation was applied before readiness for SBT in two studies [9, 21], after failing one SBT [8, 18, 20] in three RCTs, or after failing SBT for three consecutive days [19] in one study. Nevertheless, the sensitivity analysis based on the leave-one-out method indicates no effect on the primary endpoint. Third, despite the overall risk of bias being assessed as low, blinding the caregivers to treatment allocation was not possible in all the original studies. This is partly mitigated by our choice of objective outcomes, where the risk of detection bias is low. We share this limitation with previous meta-analyses on the use of NIV to facilitate weaning [7, 24]; however, the reporting bias affects IPD meta-analysis to a lesser extent than traditional meta-analysis. Fourth, most of the included studies are of limited size. As a result, baseline imbalances between treatment groups, such as PaO2/FiO2 values that was different in the intervention and control group could have occurred by chance. However, the results are not substantially affected when adjusting for possible confounders.

Conclusions

Patients recovering from an episode of hARF may benefit from a weaning strategy based on early extubation followed by immediate NIV application. Compared to conventional weaning, replacing the endotracheal tube with a noninvasive interface reduces the duration of i-MV. Overall time spent on mechanical ventilation, length of ICU and hospital stay, and risk of VAP may also be reduced by this weaning strategy. Future studies are warranted to evaluate whether this approach is also associated with reduced mortality.