Of all the 2256 adult cardiac surgical patients operated upon during a 12-month period from 1st February 1992 in three units, only 162 (7.2%) spent more than 48 h in the intensive care unit (ICU) (median 6 days, range 3-90). There were 47 deaths in ICU, 7 more before hospital discharge, and a further 10 before the study end-point of one year after surgery. All 98 1-year survivors were at home with 86 of them reporting their quality of life, on formal evaluation, to be within the reference range which we have established for a less complicated cohort of cardiac surgical patients. Prospectively collected physiological measurements were used in a mathematical model to test how well we could predict which patients will die and which of the survivors have a poor quality of life. The algorithm performs well for cardiac surgery patients with a specificity of 98%. If treatment had been withdrawn when death or poor quality of life became predictable, the maximum number of ICU bed days that could be freed was of the order of 2%. The plight of these patients is distressing, but most survive and do well and they are infrequent compared with the large majority who survive to leave hospital after a short ICU stay.