Sample size and statistical analyses
The study was designed to detect a plausible reduction
of reported infections (with glutamine) within the first
14 days, a 15% reduction from 50% to 35%. About 340
patients would be required to detect this difference
(80% power, two sided 5% significance level). To
allow for loss of patients to follow-up during the six
months for other outcomes, the trial recruitment target
was inflated to 500 patients.
All outcomes were analysed on the basis of intention
to treat (that is, all participants analysed within the
group into which they were randomised irrespective
of subsequent compliance with nutritional support).
We also undertook a prespecified analysis of the primary
outcomes in participants who received trial parenteral
nutrition for at least five days.10
The principal comparisons were between (a) all
those randomised to glutamine formulation versus all
those not allocated glutamine and (b) all those randomised
to extra selenium versus all those not allocated
selenium. The primary outcome measures (infections
and mortality) were analysed using logistic regression.
Lengths of stay in the critical care or high dependency
unit and in acute hospital were analysed with Cox proportional
hazards models. Duration of treatment with
antibiotics and modified SOFA score were analysed
with the Mann-Whitney test.
All regression models included covariates indicating
whether the patient had been allocated to receive glutamine
or selenium and the minimisation variables
(excluding trial centre). A two sided significance level
of 5% was considered evidence of statistical
significance, and corresponding confidence intervals
were calculated. An interaction between glutamine
and selenium was tested for by adding the interaction
term to the model (two sided 5% significance level).
Subgroup analyses, defined a priori, assessed severity
of disease (Acute Physiology and Chronic Health
Evaluation II (APACHE II) scores split by median
score and the minimisation variables) by adding treatment
subgroup interaction terms to the models. We
used a two sided 1% significance level to test the treatment
by subgroup, interactions reflecting the more
speculative nature of these analyses.
RESULTS
Atotal of 1134 patients were screened for participation
from June 2004 to November 2008, of whom 506 did
not meet trial inclusion criteria (fig 1). For 53 patients,
trial parenteral nutrition was not immediately available,
leaving 575 patients from whom consent was
sought. The 502 who were randomised were located
in 10 centres in Scotland.
Baseline characteristics
Baseline characteristics were well matched across
groups (table 1). Participants had a mean (SD) age of
63.8 (14.9) years, 39% were women, 89% required
level 3 care, 25% were medical patients (defined as
non-surgical cause for admission), and the median
APACHE II score was 20 (interquartile range 16–25).
At randomisation, 89% were already receiving antibiotics,
and 27% were undernourished. The nature of
previous nutrition support was also well balanced