The research question was supported by inclusion criteria for participants, outcomes and intervention and broad criteria for study design. All languages were searched, which reduced the risk of language bias. It appeared that unpublished studies were not sought; therefore, publication bias could not be ruled out. The validity of primary studies was not assessed and so the reliability of their results was unknown; however, study samples were small and observational (without control groups) and this suggested that the included studies were not of high quality. Studies were selected by two reviewers, which reduced risks of error and bias; no similar steps were reported for data extraction. The studies were heterogeneous in terms of median hospital stay and treatment regimens; therefore, narrative synthesis appeared appropriate.
The conclusions were based on observational non-controlled study designs with small sample sizes and there was a possibility of bias within the review, so they may not be reliable. Cost was not specifically investigated and so this conclusion should be interpreted with caution.