A) To examine the accuracy of the Acute Physiology and Chronic Health Evaluation (APACHE II) and the Glasgow Coma Scores as predictors of the outcome of patients following resuscitation from cardiac arrest; b) to study the impact of the components of APACHE II on the prediction. A nationwide study in Finland with prospectively collected data on all patients admitted to intensive care after cardiac arrest during a 14-month period. Two thirds of the cardiac arrest patients included in the study were randomly selected to derive predictive models, and the remaining one third constituted the validation sample. A total of 25 medical and surgical ICUs in Finland (13 in tertiary referral centers). Six-hundred nineteen consecutive cardiac arrest patients. Fifteen patients less than 16 yrs were excluded. Variables included in the APACHE II or Glasgow Coma Scores were collected at the time of ICU admission and then three times after admission, at 24-hr intervals. ICU- and hospital-mortality rates and a 6-month mortality rate after ICU admission were studied. Of 604 study patients, 370 (61.3%) patients died in the hospital. The most accurate prediction of hospital outcome was based on data collected after the first day of ICU care, not on the admission values. Twenty-one (21.9%) of 96 patients with a low APACHE II score (less than or equal to 9) died compared with 66 (84.6%) of 78 patients with a high APACHE II score (greater than or equal to 25) (p less than .001). Of 160 patients with a normal Glasgow Coma Score (14 to 15), 45 (28.1%) died, whereas there were 114 (81.4%) nonsurvivors among 140 patients with a low Glasgow Coma Score of 3 (p less than .001). The performance of predictive models, including age, the Chronic Health Evaluation, and either the Acute Physiology Score (Acute Physiology Score model) or the Glasgow Coma Score (Glasgow Coma Score model) were compared with the prediction according to the APACHE II in the validation sample. When using 80% probability of death as a decision rule, the Acute Physiology Score model determined 35 of 153 patients to have high risk of death, 29 of whom died (the positive predictive value being 82.9%). The Glasgow Coma Score model predicted 34 patients to die, 26 of whom died (positive predictive value 76.5%), and the APACHE II score predicted seven deaths, five of whom actually died (positive predictive value 71.4%). The APACHE II scoring system cannot be recommended as a prognostic tool to support clinical judgement in cardiac arrest patients, but by modifying it, a more accurate prediction of poor outcome could be achieved. The Glasgow Coma Score explained to a great extent the predictive power of the APACHE II.