The device used, the ventilation mode while training, training pressure, duration, frequency, and progression of training were recorded for the experimental group and for the control group if it received sham training. The method of inspiratory muscle training (isocapnic/normocapnic hyperpnoea, inspiratory resistive training, threshold pressure loading, or adjustment of ventilator pressure trigger sensitivity) was also recorded. Outcome measures: Primary outcome measures were measures of inspiratory muscle strength at a controlled lung volume (eg,

maximal inspiratory pressure at residual volume), inspiratory Temsirolimus muscle endurance, the duration of unassisted breathing periods, weaning success (ie, proportion of patients successfully weaned, defined as spontaneous breathing without mechanical support for at least 48 hours), weaning duration (ie, from the identification of readiness to wean, as determined by the authors and/or commencement of inspiratory muscle training, to the discontinuation of mechanical ventilation) and reintubation (ie, proportion of extubated patients who were reintubated within the follow-up period of the study). Secondary outcomes were tracheostomy (ie, proportion GS-7340 cell line of extubated patients tracheostomised after the commencement

of training), survival, adverse effects, and length of stay in hospital or the intensive care unit. The relevant data including study characteristics and outcome data were extracted from the eligible studies by two reviewers (LM and JR) using a standard form and the third author (ME) arbitrated in cases of disagreement. The reviewers extracted information about the method (design, participants,

and intervention) and outcome data for the experimental and control groups. Authors were contacted where there was difficulty in interpreting or extracting data. The data analysis was performed using Revman 5.1 (Revman 2011). A fixed-effect model was used unless there was substantial heterogeneity (I2 > 50%), when a random-effects model was used. Continuous outcomes were reported as weighted mean differences with next 95% CIs, while dichotomous outcomes were reported as risk ratios with 95% CIs. The search retrieved 816 studies. After screening titles and abstracts, 797 were excluded and 19 full text articles were identified. After evaluation of the full text, nine studies were excluded on the basis of participants not meeting the inclusion criteria. A further three were excluded on the basis of the intervention not meeting the inclusion criteria. Therefore seven papers (Cader et al 2010, Caruso et al 2005, Martin et al 2006a, Martin et al 2006b, Martin et al 2007, Martin et al 2009, Martin 2011) met the inclusion criteria for the review. One trial was reported across five publications (Martin et al 2006a, Martin et al 2006b, Martin et al 2007, Martin et al 2009, Martin et al 2011), so the seven included papers provided data on three trials.