Eligibility criteria
Types of study
Randomized controlled trials (RCTs) will be included.
Types of participants
Patients aged ≥18 years old with mechanical ventilation in ICU will be included.
Types of intervention
We are interested in any protocols of early rehabilitation to improve the outcomes of mechanically ventilated patients. Also, both simple intervention and multiple interventions will be included.
Comparators
We will make comparisons between usual care and early rehabilitation or between different protocols of early rehabilitation.
Outcome measures
Primary outcome
The incidence of ICU-AW of mechanically ventilated patients, which will be evaluated by MRC-score23.
Secondary outcomes
We will include the duration of mechanical ventilation, length of stay in ICU and hospital.
Search strategy
PubMed, Cochrane library, Web of Science, EMbase, grey literature and referent literature will be searched from inception to August 2019. The search strategies will be (“early mobilization” OR “early rehabilitation” OR “active mobilization” OR “passive mobilization” OR “whole-body vibration” OR “early passive tilting” OR “early cognitive and physical rehabilitation” OR “passive leg cycling” OR “early In-Bed cycling” OR “in-bed cycling exercise” OR “interactive video games” OR “electrical muscle stimulation” OR “neuromuscular electrical stimulation” OR “NMES”) AND (“ICU” OR “intensive care unit*” OR “burn unit*” OR “coronary care unit*” OR “recovery room” OR “respiratory care unit*”). And an experienced expert will be invited to review the selected search strategy. In addition, we will manually search for the unpublished literature and will also try our best to contact the authors via email to require unreliable literature.
Study selection
All studies will be selected in the EndNote X8 which is a literature management software. Two authors will independently review the titles and abstracts of studies to screen studies, then review the full-text to include studies for quantitative analysis. A third reviewer will be invited for the discrepancies between the two authors.
Data extraction
Two authors will independently extract data according to a predesigned form, which will include study characteristics (first author, year, country, ICU type), patient characteristics (sample size, male/female, age, interventions, and controls), and outcomes (ICU-AW, duration of mechanical ventilation, and length of stay in ICU and hospital).
Quality assessment
Two reviewers will independently assess the risk of bias using the Cochrane Handbook for Systematic Reviews (V5.1.0), which includes random sequence generation, allocation concealment, participants/personnel blinding, outcome assessor blinding, incomplete outcome data, selective reporting, and other bias24. The third reviewer will be invited to solve the discrepancies of the two reviewers.
Data synthesis and statistical methods
We will use R software (V3.6.0, University of Auckland, New Zealand) to conduct statistical analysis and graphical presentations, which will included pairwise meta-analysis, ranking possibility, inconsistency analysis, subgroup analysis and sensitivity analyses; Stata software (V13.0, StataCorp, College Station, Texas, USA) will be used to assess publication bias by funnel plot, Begg or Egger’s test if included studies were ten or more.
Network meta-analysis
Dichotomous outcomes will use odd ratio (OR) and 95% confidence interval (95%CI) as effect value; mean differences or standardised mean differences (SMDs) with 95% CI for continuous outcomes. Heterogeneity text will be assessed by using the χ2 test and I2 statistic. If p value ≥ 0.1 and I2 ≤ 50%, we will consider there is no significant statistical heterogeneity exists and use the Mantel-Haenszel fixed effect model; if in contrast, we will use the random effects model. Sensitivity analysis will be conducted to explore the sources of heterogeneity. Publication bias or small sample effect will be examined through funnel plot, Begger or Egger’s test if more than ten studies are available.
Ranking possibility
We will perform histogram of ranking probability to rank the intervention. The bigger of the value in the rank, the more possibility of it rank there.
Assessment of inconsistency
The consistency between direct and indirect evidence will be assessed by node split method in the loop, which separates direct evidence from indirect evidence for a specific comparison in the loop (inconsistency factor). P value will be used to estimate the inconsistency, p > 0.05 indicated it is consistency between direct and indirect comparisons, and it is inconsistency when p ≤ 0.05.
Subgroup analysis and sensitivity analysis
Subgroup analyses and network meta-regression will be performed according to the ICU type, country and study quality when there is high heterogeneity. And we will perform sensitivity analyses by excluding each study to check the stability of the results.
Publication bias
Potential publication bias will be assessed by funnel plot, Egger or Begger’s text.
Quality of evidence
The evidence quality will be assessed following the Grading of Recommendations, Assessment, Development and Evaluation (GRADE) method25.