Benefits of ICU admission in critically ill patients: Whether instrumental variable methods or propensity scores should be used

Table 4 Evaluation of the qualities of the potential instruments

	OR [95CI]	Partial r²	Partial F-statistic	p-value	IV effect on ICU admission Estimate (SD)	Standardized Differences
						Age	SOFA	SAPS	GCS	Karnofsky
Original cohort						0.311	-0.205	0.092	0.166	-0.352
Instrumental Variable (IV)
Country of enrolment	2.90 [2.58-3.28]	0.006	54.90	< .0001	0.15 (0.01)	0.306	0.075	0.209	-0.001	-0.143
Physician's age	2.03 [1.81-2.27]	0.001	13.42	.0003	-0.13 (0.01)	-0.003	-0.036	0.042	-0.031	-0.013
Physician's main specialization	2.22 [1.96-2.51]	0.003	25.55	< .0001	0.10 (0.01)	-0.071	0.075	0.080	-0.094	0.048

Partial r²: square of the partial correlation between the instrument and the treatment. GCS: Glasgow Coma Scale. OR: odds ratio. 95CI: 95% confidence interval. The IV effect on ICU admission (denominator of the Wald estimator) is expressed as the estimate (SD) of the linear regression that models ICU acceptance according to the IV.

ISSN: 1471-2288