Objective: To evaluate the use of daily Acute Physiology and Chronic Health Evaluation (APACHE) II scoring in the prediction of individual mortality rates for intensive care unit (ICU) patients.
Design: A prospective study of consecutive patients admitted to four university-affiliated ICUs.
Setting: Medical-surgical ICUs of four tertiary care academic hospitals.
Patients: Daily data from 3,350 consecutive ICU admissions, excluding postoperative cardiac patients, were collected from January to December 1991.
Measurements and main results: Daily APACHE II scores were calculated for all patients and correlated with both ICU and hospital mortality. The ability of an absolute level or a predetermined algorithm, based on these scores, to predict mortality was examined. Day 1 APACHE II scores ranged from 0 to 55 (mean 18). We were unable to replicate the suggestion by Chang et al. that 100% hospital mortality was predicted by the following APACHE II scores: a) > 35 at admission; b) 30 to 35 at admission, with a decrease of < or = 3 from day 1 to day 2; or c) > 27 on any day, with an increase of > 2 from the previous day. We were unable to adjust these criteria to avoid a false prediction of death with any remaining useful sensitivity. Mortality rates of 158 (69%) deaths per 229 patients, 68 (62%) deaths per 110 patients, and 110 (48%) deaths per 230 patients were obtained, respectively, for these criteria.
Conclusions: Admission or daily APACHE II scores do not predict individual patient mortality. The adjustments needed in the algorithm that was used to avoid a false prediction of death render sensitivity so low that it would be impractical to limit therapy on this basis alone.