Table 2.

Length of intensive care unit (ICU) stay attributable to clinically important gastrointestinal bleeding

ICU length of stay	Median difference
(days)	(95% confidence interval)
Crude comparison
Non-survivors	14.0 (8.6–19.4)
Overall	17.2 (13.2–21.3)
Matched cohort method
Non-survivors	0.6 (-5.1–6.3)
Overall	3.8 (-0.01–7.6)
Model-based matched cohort method
Matching on mortality status	5.1 (0.3–9.9)
Overall	6.7 (2.7–10.7)
Regression method
Unadjusted	7.9 (1.4–14.4)
Adjusted*	6.2 (1.0–11.4)

The model-based matched cohort probably yields the best estimate of the length of ICU stay. See Table 3 for advantages and disadvantages of these methods. ^* Adjusted for age, APACHE II score, admitting diagnosis, duration of ventilation, Multiple Organ Dysfunction Score, and bleeding status.