Limitations of score-based daily outcome predictions in the individual intensive care patient. An example of the RIAHDH algorithm.
OBJECTIVE: The nature of score-based predictions is probabilistic, and their accuracy depends on the reliability and validity of the applied system. As an example, the present study investigates the accuracy of the RIP-algorithm (RIP = Riyadh Intensive Care Program) based on daily APACHE II scores, and compares it with published results of that algorithm from other investigators. DESIGN: Prospective observational study and review of the literature. PATIENTS AND METHODS: 1,986 consecutive admissions of 1,808 patients to a surgical intensive care unit were documented. Daily changes of score values were used to derive a risk of death estimation. Sensitivity and the rate of false predictions were calculated for score-based predictions. Health status one year after discharge was assessed in survivors predicted to die. RESULTS: Daily application of the algorithm identified 109 situations leading to death predictions in 56 patients. Five of these patients were discharged alive from the hospital (positive predictive value 91%). One year later 3 of these patients were still alive. The algorithm identified 51 of the non-survivors (sensitivity 19%), 110 died in the ICU without prediction. Altogether 270 patients died during their hospital stay. Among the 6 independent validation studies, similar results were found, but differences occurred due to the problematic assessment of consciousness. CONCLUSIONS: Sequential assessment of scores in intensive care could identify high risk patients, but with some degree of uncertainty. Therefore, the scores should only be used by those familiar with their limitations and risks.