Table 3.

Multivariable analysis assessing associations between respondent characteristics and transfusion thresholds in the context of a randomized, controlled trial^a

Clinical setting	Predictors remaining in final model	Estimate (β)	P value
WFNS grade 4
(lowest acceptable Hb)	Specialty (neurosurgery)	0.37	<0.0001
	Years in practice	0.01	0.009
WFNS grade 4
(highest acceptable Hb)	Specialty (neurosurgery)b Years in practicec,d	0.03	0.01
DCI
(lowest acceptable Hb)	Years in practice	0.02	0.01
DCI
(highest acceptable Hb)	Transfusion protocol	-0.35	0.003
	Years in practice	0.02	0.007
	Transfusion protocolb Specialty (neurosurgery)b	0.66	0.002

^aMultivariable analysis was performed using generalized linear models with stepwise backward elimination of the least significant variable where P > 0.05. Initial models included country (United States vs. Canada), specialty (neurosurgery vs. critical care), multimodal monitoring (yes vs. no), use of a transfusion protocol (yes vs. no) and years in practice (continuous variable). All interactions were assessed, and those for which P < 0.05 in univariate analysis were incorporated into initial multivariable models. WFNS, World Federation of Neurological Surgeons scale; Hb, hemoglobin concentration; DCI, delayed cerebral ischemia. ^bNeurosurgical specialty significantly modified practices among clinicians who use a protocol (see Results section for details); ^cWhite's heteroscedasticity-specific standard error; ^dYears in practice significantly modified practices among neurosurgeons (see Results section for details).