Statistical analyses
Using a random effect model we report the differences in
categorical outcomes between the treatment and placebo or
control groups as relative risks with 95% confidence intervals.We
further stratified the effects of frusemide on mortality and the
need for dialysis after frusemide treatment into studies using
frusemide to prevent or to treat acute renal failure, and we tested
this interaction by relative risk ratio.9 Using a random effect
model we report the differences in length of hospital stay and the
number of dialysis sessions required as weighted mean
differences. We used the 2 statistic to assess heterogeneity
between trials and the I2 statistic to assess the extent of inconsistency.
10 One study reported tinnitus and deafness in several
patients after frusemide treatment but did not specify the
number.w5 We therefore estimated that at least three patients
would have tinnitus or deafness in the frusemide group. One
study reported the duration of continuous renal replacement
therapy until recovery.w4 We pooled the results of this study with
others that reported the total number of dialysis sessions
required until recovery of renal function.We carried out sensitivity
analyses by excluding one study that compared a single dose
of frusemide with prolonged continuous infusionw6 or by including
only studies that had adequate allocation concealment.w1-w3 w7
Publication bias was assessed by funnel plot using mortality as an
end point.We considered a P value less than 0.05 as significant.