Table 1.

Reanalysis of the number of units of red blood transfused in the Trial in Gastrointestinal Transfusion (TRIGGER) trial

Analysis method	Is the analysis method appropriate for use with a small or medium number of clusters?	Difference in meansa (95 % CI)	P value
Unadjusted for clustering	No	−0.7 (−1.1 to −0.4)	<0.001
GEE (no correction)	No	−0.7 (−1.2 to −0.2)	0.01
Mixed-effects model (no correction)	No	−0.7 (−1.2 to −0.2)	0.01
GEE (with correction)	Yes	−0.7 (−1.8 to 0.4)	0.15
Mixed-effects model (with correction)	Yes	−0.7 (−1.4 to 0.0)	0.04
Cluster-level analysis	Yes	−0.7 (−1.6 to 0.3)	0.12

^aThe difference in means is presented as the restrictive transfusion policy minus the liberal transfusion policy

CI confidence interval, GEE generalised estimating equation