Assessment and optimization of mortality prediction tools for admissions to pediatric intensive care in the United Kingdom

OBJECTIVE. To assess the Pediatric Risk of Mortality (PRISM, PRISM III-12, and PRISM III-24) systems and the Pediatric Index of Mortality (PIM and PIM2) systems for use in comparing the risk-adjusted mortality of children after admission for pediatric intensive care in the United Kingdom. METHODS. All PICUs in the United Kingdom were invited to participate. Predicted probability of PICU mortality was calculated using the published algorithms for PIM, PIM2, and PRISM and compared with observed mortality. These scores, along with PRISM III-12 and PRISM III-24, whose algorithms are not published, were optimized for the United Kingdom. RESULTS. Of 26 PICUs in the United Kingdom, 22 (85%) were recruited, and sufficient prospective data were collected from 18 (69%) units on 10 197 (98%) of 10 385 admissions between March 2001 and February 2002. All published tools were found to have poor calibration but provided good discriminatory power. After estimation of UK-specific coefficients, only PIM2, PRISM III-12, and PRISM III-24 had satisfactory calibration. All models provided good discriminatory power. Funnel plots for all of the recalibrated models indicated that the risk-adjusted mortality for all units was consistent with random variation. CONCLUSIONS. PIM2, PRISM III-12, and PRISM III-24 all were found to be suitable for use in a UK PICU setting. All tools provided similar conclusions in assessing the distribution of risk-adjusted mortality in UK PICUs. It now is important that these tools be used to monitor outcome and improve the quality of pediatric intensive care within the United Kingdom.