Prospective evaluation of the Paediatric Risk of Mortality (PRISM) score

Abstract

The performance of the admission day Paediatric Risk of Mortality (PRISM) score for outcome prediction was assessed prospectively in 270 consecutive admissions, aged 3 days to 18.6 years, to a paediatric intensive care unit. Using a cut off of r = 0.00 (expected mortality = 50%), the overall sensitivity (correct prediction of death) was 48% while specificity (correct prediction of survival) was 99%, comparable with the original validation data of the score in the USA. Outcome prediction was most accurate when the stay in the paediatric intensive care unit was between one and four days. Sensitivity was appreciably lower for operative patients (17%) compared with non-operative patients (71%) because of a failure to predict deaths after cardiac surgery. The sensitivity (41%) and specificity (99%) using five variables (systolic blood pressure, Glasgow coma scale, carbon dioxide tension, and serum bicarbonate and serum calcium concentrations) was similar to that using all 14 variables. Six variable ranges related differently with non-survival compared with the score. It is concluded that the performance of the PRISM score is institution independent and good for short stay patients. It underpredicts deaths after cardiac surgery. Only five variables may be needed for satisfactory outcome prediction. Some of the variables need reweighting for paediatric intensive care units in the UK.