Physiology-based prognostic models have been advocated for clinical decision making in individual patients as well as for analysis and implementation of health policies. We evaluated the strengths and limitations of a physiology-based predictor, PRISM III (Pediatric Risk of Mortality Score), in the context of these potential uses.

Methods: Data from consecutive admissions to a diverse group of 32 pediatric ICUs included physiologic profiles, mortality (ICU and hospital), and demographic information. PRISM III scores were computed using data from the first 24 hoUrs of care. For patients with multiple ICU admissions, only the initial ICU admission was considered. Statistical measures of clinical“certainty” were expressed as confidence intervals.

Results: 10,608 patients (571 deaths) were evaluated. The cutoff for an observed 100% mortality was a PRISM III score > 42 (21 deaths). From individual patient's perspective, the 95%, 90%, 80% and 70% lower confidence intervals for mortality (i.e. the lowest true mortality risk in the population of patients with a PRISM III-24 score > 42 that could result in the observation of 0/21 survivors with a “certainty” level = the confidence limit) was 86.7%, 89.6%, 92.6%, and 94.5%. Thus, in the present sample, the observed mortality for a PRISM III-24 score of >42 was 100%, while the true mortality risk might be as low as 86.7% with a 95%“certainty.” From a societal perspective, we evaluated the error rate of a policy of automatically discharging patients from the pediatric ICU who have a PRISM III-24 score of >42. There were 0 errors/10,037 patients observed. However the true error rate (potential survivors who might die because of ICU discharge) could be as high as 0.03% with a“certainty” of 95%.

Conclusions: Even with contemporary, very large data bases, objective, physiology-based scores such as PRISM III cannot be used as sole criteria for implementation of decisions for individual patients. However, the very low true error rate for societal evaluations indicates potential use in health policy evaluations.