Twelve randomised and quasi-randomised controlled trials were included (n=898).
The quality of the included studies was generally poor. Methodological problems included: inadequate concealment of allocation (9 studies); lack of intention-to-treat analysis (8 studies); exclusion of patients after randomisation (3 studies); lack of blinding of outcome assessors (10 studies); and short-term follow-up (only 5 studies followed patients for 6 months or more). All studies defined inclusion or exclusion criteria well.
Mortality (8 trials).
There were no significant differences in terms of overall mortality between the supplement and no supplement groups (RR 0.92, 95% CI: 0.56, 1.50). Marginally significant heterogeneity between the trials was observed (P<0.1). There was no evidence of an effect for mortality when split by the three different feeding routes. There was considerable heterogeneity between the three nasogastric feeding trials shown (P=0.04). When the trials were subgrouped according to nutritional status, mortality tended to be reduced in the supplemented groups in those trials that targeted malnourished patients(RR 0.61, 95% CI: 0.28, 1.34), but not in the other trials (RR 1.20, 95% CI: 0.63, 2.28). However, this was not statistically significant.
Unfavourable outcome (mortality or complications) (3 trials).
A significant benefit in favour of the oral supplement group was shown (RR 0.52, 95% CI: 0.32, 0.84). No significant statistical heterogeneity was observed. However, the results of further analyses based on the number of patients randomised (in which it was assumed that all excluded patients in the supplemented group had complications at follow-up) showed that for both complications and unfavourable outcomes the results were no longer significant (RR 0.70, 95% CI: 0.37, 1.34 and RR 0.67, 95% CI: 0.42, 1.05, respectively.
Protein only supplementation (3 trials).
No significant differences in relation to mortality were shown between the supplemented and non-supplemented groups (RR 1.38, 95% CI: 0.82, 2.34). No significant statistical heterogeneity was found.