We describe an audit system used in our Medical/Surgical Intensive Care Unit (ICU) during 1989-90. The system emphasizes the integration of data acquisition (database function) with the analysis and use of data (decision function). Resource input (human and technological) included patient demographics, diagnoses, complications, procedures, severity of illness (Apache II), therapeutic interventions (TISS), and nursing workload (GRASP and TISS). The output was assessed by survival, length of stay and ability to return home. The annual operating cost for 277 admissions (249 patients) to this ICU was $7,333. The implementation costs were $58,261 including program development and computer purchases. Non-survivors of ICU and hospital had higher Apache II scores on admission (P less than 0.0001) and longer ICU length of stay (P less than 0.05) than survivors. The nursing workload (both TISS and GRASP) on the day of admission and the last day in ICU were greater in non-survivors (P less than 0.0001) than survivors. Limitations of this audit system included the delay (6-9 mos) from ICU admission until data entry, the large number of diagnostic groups in the ICD.9.CM classification, and lack of a documented cause/effect relationship between interventions and complications. This audit system was more useful for utilization management than for quality assurance purposes.