Table 3.

Advantages and disadvantages of different approaches to estimating attributable mortality and length of intensive care unit stay

Method	Advantages	Disadvantages
Crude comparison	Simple	Ignores influence of confounding factors, possibly
		yielding biased estimates
Matched cohort method	Integrates biologic rationale for matching	May fail to adjust for important confounding factors,
	patients; can be used in multiple databases;	possibly yielding biased estimates; compared with
	compared with regression method, avoids bias	regression model, event rate over time is not
	if event rate is not constant over time	considered
Model-based matched cohort	Analysis customized to the database; compared	Compared with matched cohort method, chance
method	with crude or matched cohort method, more	associations may generate biased estimates due to
	likely to adjust for important confounding factors;	'overmatching'; compared with regression model,
	compared with regression method, avoids bias	event rate over time is not considered
	if event rate is not constant over time
Regression method	Analysis customized to the database; uses all	Complex; compared with matched cohort method,
	patient data; considers patterns of events and	chance associations may generate biased estimates
	predictors over time and generates most	due to 'overmatching'; biased estimates may also
	precise estimates if event rate is constant	result if event rate is not constant over time
	over time

The advantages and disadvantages of several approaches to estimating the attributable mortality and length of intensive care unit stay associated with clinically important bleeding are presented (a crude comparison of bleeding and non-bleeding patients, and the three methods used in these analyses).