This multicentre prospective, superiority single-blind randomized, clinical trial in three parallel groups was conducted in two medical ICUs at Bordeaux University Hospital and Lyon University Hospital, France between October 2016 and January 2020.

Characteristics of participants

Inclusion criteria were:

patients older than 18 years

receipt of at least 18 h of controlled MV

first failure of a spontaneous breathing trial (SBT) or failed extubation. SBT failure was defined by the presence of at least one of the following outcome: (1) change in state of consciousness (anxiety, agitation, drowsiness), (2) impaired oxygenation (cyanosis, PaO2 < 50 mmHg or SpO2 < 90% with FiO2 > 50%), (3) dyspnea and signs of respiratory distress (drawing, use of accessory respiratory muscles, tachypnea > 35/min), (4) hemodynamic changes (tachycardia heart rhythm > 140 beat/min, (5) arterial hypertension > 180 mmHg or an increase of more than 20%, (6) arterial hypotension < 90 mmHg, presence of arrhythmias). Failed extubation was defined by the need of reintubation within 48 h after extubation.

presence of weaning criteria as defined in the European Consensus Conference in 2007, including sedation reduction, spontaneous breathing cycles, PaO2/FiO2 ≥ 150, absence of inotropes or vasopressors at high doses or increasing doses (< 1 mg/h), oxyhaemoglobin saturation (SaO2) > 90% with FiO2 ≤ 50%, PEEP ≤ 8 cmH2O, temperature between 36 and 39 °C, and Glasgow Coma scale ≥ 8 [5].

Non-inclusion criteria were hemodynamic or respiratory instability, severe ventricular arrhythmias, poor short-term vital prognosis, cardiac arrest with a guarded neurological prognosis, proven neurodegenerative pathology, tracheotomy, current pregnancy, a curatorship measure, a do-not-resuscitate order, or a lack of coherence to follow verbal commands with a best motor response item of the Glasgow Coma scale lower than 6.

Randomization

An unbalanced randomisation ratio favoured the innovative MI group (2:1:1), based on a randomised, computer-generated list with stratification according to study centre and permuted blocks of varying sizes (4 and 8). Given that a greater benefit was expected in participants benefiting from the MI strategy, and that we wished to collect as much data as possible on this new strategy, we decided to implement an unbalanced randomization scheme at design stage.

Randomisation was performed using a web-based centralised system after confirmation of participant eligibility. For reasons of staffing and feasibility, assessors were not independent and blinded for the allocation. Both patients and physiotherapists were not blinded to treatment allocation, but physicians responsible for the extubation decision were blinded to the IMT protocol.

The study protocol was approved by an independent ethics committee (Comité de Protection des Personnes Sud-Ouest et Outre-Mer III; DC 2016/03; NCT02855619). The study was performed in accordance with the CONSORT statement and the Declaration of Helsinki.

Adverse effects were notified by the investigator to the security and vigilance unit who determined the attributability of the event under study and the need for a declaration to the competent authorities.

Sample size

A statistical analysis plan was developed and validated by the trial steering committee prior to the final database lock and analyses. Anticipated increases in MIP were projected to be 12 cmH2O in the MI group, 9.7 cmH2O in the HI group, and 9.9 cmH2O in the LI group, as previously reported [23, 24], with a common standard deviation of 2.5 cmH2O. An increase of 2 cmH2O in the MI group was considered a minimum threshold for additional gain compared with the other groups, representing a 20% increase in inspiratory force measurement. The study was thus designed to detect a difference equal to or greater than this threshold of 2 cmH2O between groups. As the MI group was compared with each of the two other groups for the primary objective, the two-sided type I error rate was set at 0.025 using Bonferroni correction. With 80% power and an unbalanced 2:1:1 ratio, 88 participants were required (46, 23, and 23 for MI, HI, and LI groups, respectively).

Interventions

Study design is represented in Fig. 1.

Fig. 1

Study design. SBT spontaneous breathing trial, IMT inspiratory muscle training, D day

Patients were enrolled within 48 h of the first failed SBT or failed extubation if they met the aforementioned eligibility criteria and consent had been obtained from the next of kin. Subsequently, they were randomised into one of the three IMT protocol groups (Fig. 2).

Fig. 2

Description of the 3 IMT programs

All participants received two IMT interventions daily (with ≥ 4 h between sessions) 7 days per week, from enrolment until one of two possible endpoints—successful extubation or 30 days of mechanical ventilation (D30), whereas arrived first. Considering potential fluctuations in consciousness or general stability, the inclusion criteria were assessed daily; patient participation in IMT sessions was contingent upon meeting these criteria. Before each training session, patients were positioned in a 45° Fowler’s position, and cardiorespiratory variables were evaluated to ensure hemodynamic stability. Criteria for instability included respiratory rate (RR) > 35 breath/min, SaO2 < 90%, systolic blood pressure (SBP) > 180 mmHg or < 90 mmHg, paradoxical breathing, agitation, and/or tachycardia.

Maximal inspiratory pressure (MIP; cmH2O) was measured daily before the first IMT session to obtain primary outcome data, although this measurement had the potential to induce a training effect and/or fatigue. All patients were disconnected from MV and performed IMT using the Threshold IMT device (Philips Respironics; Murrysville, PE, USA), directly connected to the ETT. Supplementary oxygen was added as needed. All patients were instructed to take large deep breaths from functional residual capacity and to breathe in with more force when it became difficult to open the IMT Threshold valve.

Between each set, or in case of instability as defined earlier, or if patient desire to discontinue the intervention, training sessions were halted and patients were reconnected to MV with previous pressure-support ventilation settings.

HI IMT program [24]: patients completed four sets of six breaths against the highest inspiratory resistance tolerated. Each series was interspersed with a 2-min pause during which patients were reconnected to MV. Resistance titration was initiated at 9 cmH2O (lowest load possible on the Threshold IMT device) on day 1. Subsequent adjustments of + 0, + 2, or + 5 cmH2O were made for each set according to patient tolerance and physiotherapist evaluation. Regardless of the daily MIP measurement, sessions began using the highest resistance from the previous day, with increases of + 0, + 2, or + 5 cmH2O for each set. Each set was interspersed with a 2-min pause to reconnect the patient to MV without parameter changes.

LI IMT program [23]: The physiotherapist applied a single inspiratory resistance equivalent to 30% of the MIP measured on the inclusion day for 5 min. The load was increased daily by 10% of the initial MIP, independent of the daily MIP measurement, if the previous session was completed. According to the American Sport College Medicine [26], for local muscular endurance training, it is recommended that light to moderate loads (40–60% of 1 maximal resistance) be performed for high repetitions (> 15) using short rest periods (< 90 s). To our point of view, this regimen of training performed by Cader et al. correspond to a low intensity program for endurance training.

MI IMT program: In this novel protocol, the physiotherapist set the inspiratory resistance device to 30% of the daily MIP for the first set of 20 breaths. In each set, resistance was increased by 10% of the daily MIP until a resistance equivalent to 60% of the MIP was achieved in the fourth set. Each set was interspersed with a 2-min pause to reconnect the patient to MV without parameter changes. This mixed regimen (both strength and endurance training have been inspired by the works of Bird et al. [27].

Measurements

The primary outcome was the change in strength between the randomisation day (D1) and successful extubation or Day 30, using MIP (cmH2O) based on the method of Caruso et al. [28]. A unidirectional expiratory valve attached to the ETT and an external pneumotachograph (Fluxmed GrH monitor; MBMED, Buenos Aires, AR, USA) were used. MIP was daily assessed before the first session.

The main secondary outcome was endurance using an incremental endurance test to obtain peak (Ppk; cmH2O) [29]. For this test, patients breathed through an IMT Threshold device beginning at 30% of the initial MIP and increasing by 10% every 2 min until the effort was no longer tolerable. The maximum pressure tolerated for 2 min by the patient constituted the Ppk. This incremental endurance test was expressed with Ppk time corresponding to the test duration from beginning to task failure. The patient’s tolerance limit was the same used for instability during IMT as described above. This test, similar to a training stimulus, was only performed in each group on the first day of inclusion, then weekly and immediately before extubation. We think that this test could help for the assessment of inspiratory muscles even if, to our knowledge, this is the first time this test has been used on ICU population.

Other secondary outcomes included MV duration (number of days between inclusion and successful extubation, with no reintubation for > 48 h), ICU length of stay (days), weaning success defined by a 2-day ventilator-free after extubation, reintubation rate, and safety (occurrence of adverse events related to the study, numbers of completed IMT sessions).

Statistical analysis

The primary analysis was conducted on an intent-to-treat basis, utilising the “last observation carried forward” (LOCF) strategy to replace missing data. In cases where MIP values were missing on Day 30 or successful extubation, the change in MIP between baseline and the last available follow-up value was used. If a participant experienced failed extubation on D30 with an end-of-study MIP measurement on the same day, the MIP value before extubation was used. Three patients who were tracheostomized before Day30 were considered in premature termination of monitoring and were treated using the LOCF strategy. The primary outcome was compared using a linear regression model adjusted for study centre (stratification variable), baseline MIP centred on the median, and both the presence of respiratory pathology at inclusion (defined as chronic obstructive pulmonary disease—COPD, asthma, chronic respiratory failure other than COPD, and neuromuscular disease), and length of MV before inclusion, which were regarded as a prognostic factor for extubation failure.

The same analytical strategy (including LOCF) was used to compare Ppk between Day 1 and successful extubation or Day 30. One participant with a missing baseline Ppk value was excluded from these analyses. Analysis of the other secondary outcomes was performed on participants with available data for variables included in the regression models, without imputation of missing data (i.e. complete cases analyses).

The primary analysis was performed using a two-sided overall type I error rate of 5%, with a p value threshold of 2.5% for each of the two comparisons (MI vs. LI and MI vs HI). A hierarchical testing procedure was predefined in the statistical analysis plan where for any particular comparison, if the null hypothesis for the primary endpoint (MIP) was rejected, then a statistical comparison of the secondary endpoint (Ppk) was performed at the same two-sided alpha threshold of 2.5%. If the null hypothesis for the primary outcome was not rejected, no hypothesis testing regarding Ppk was performed for that comparison. No hypothesis testing was conducted for the remaining secondary outcomes. All statistical analyses were performed using SAS v.9.4 software (SAS Institute).