We performed a random-effects meta-analysis17 for each outcome of interest. For categorical outcomes (mortality, infec- tious morbidity) the effects size was estimated by the risk ratio (RR), while for continuous outcome (length of stay) the mean difference (MD) was used. In the calculation of the pooled RR, a correction factor of 0.5 was added to all cell frequencies of studies where no patient had the outcome in either GLN or control groups. We made sure of the absence of any possible bias due to sparse data by applying also the fixed method of Peto (results not shown) which confirmed the results of the present analysis.18 Mean and standard deviation of length of stay was calculated according to method of Hozo et al19 for the studies where only median and range (or interquartile range) were reported. The weights assigned to each study were computed according to the inverse of the variance. Heterogeneity was quantified using I2 and t2 indexes and testing the null hypothesis that all studies share a common effect size. We used I2 1⁄4 30% as a threshold to establish the presence of moderate heterogeneity. Moreover, we investigated the presence of publication bias using funnel plots.20
Finally, some stratified analyses were performed according to the following indicators: GLN dosage (>0.3g/kg/day or