Pediatric mortality probability estimated from pre-ICU severity of illness.

Abstract

The Pediatric Risk of Mortality (PRISM) score is a measure of illness severity based on abnormalities observed in the bedside examination and laboratory assessment. PRISM scores obtained after pediatric intensive care unit (PICU) admission predict mortality probability, but no previous efforts to evaluate mortality risk before PICU admission have been reported. Our study was performed on patients admitted to PICUs at four pediatric tertiary care centers to derive a quantitative estimate of hospital mortality probability as a function of PRISM scores obtained at referring hospitals before PICU transfer. Performance of the model was tested by evaluating accuracy of mortality predictions obtained from pre-ICU PRISM scores in a separate validation set of patients. Patients were randomized to the derivation or validation sets. Data were recorded prospectively from observations made at hospitals referring to the study PICUs. Patients included 780 infants and children with medical and surgical emergencies and trauma. Electively admitted patients were excluded from analysis. The relationship between mortality probability (P) and the pre-ICU PRISM score is expressed by the equation: P = er/(1 + er). In this equation, r is an empirical function of the pre-ICU PRISM score: r = .197 x PRISM - 4.705. The mortality probability rises from near 0 at low scores, approaching 1 (certainty) above a PRISM score of 40. Mortality probability exceeds 10% at a score of 13 and exceeds 50% at a score of 24. Performance of predictions in the validation set of patients was evaluated for five categories of mortality probability. The observed number of deaths corresponded to predicted mortality across the range of illness severity. When compared for each tertiary institution, observed mortality rates were similar to predictions for three of four institutions. For data obtained at institution D, the observed mortality of 17% significantly exceeded the 7% predicted rate. In infants younger than 1 year, as well as children 1 year and older, observed mortality rates were similar to predicted. The pre-ICU PRISM score as a measure of illness severity provides an estimate of hospital mortality probability. Further investigation is required to determine the use of pre-ICU mortality estimates in making clinical decisions.