We have investigated biases in physicians' decisions regarding the for
m of life support to withdraw from critically ill patients in whom the
decision to withdraw has already been made. Using a specially designe
d instrument that solicited both self-reported preferences and also re
sponses to experimentally varied clinical vignettes, we surveyed 862 A
merican internists, of whom 481 (56%) responded. Physicians do have pr
eferences about the form of life support withdrawn. From most likely t
o least likely the order is: blood products, haemodialysis, intravenou
s vasopressors, total parenteral nutrition, antibiotics, mechanical ve
ntilation, tube feedings, and intravenous fluids. Four biases in decis
ion making were also identified. Physicians prefer to withdraw forms o
f therapy supporting organs that failed for natural rather than iatrog
enic reasons, to withdraw recently instituted rather than longstanding
interventions, to withdraw forms of therapy resulting in immediate de
ath rather than delayed death, and to withdraw forms of therapy result
ing in delayed death when confronted with diagnostic uncertainty. Beca
use these biases may have clinical, social, and ethical consequences c
ounter to patient goals, and because they may affect the underlying de
cision whether to withdraw life support at all, they may represent imp
ediments to rational and compassionate decision making in critical car
e.