OBJECTIVE: To compare the probabilities of death predicted by Pediatric Index of Mortality 2 (PIM2) and Pediatric Index of Mortality 1 (PIM1) models against actual mortalities in 3 pediatric intensive care units in Hong Kong. METHODS: All consecutive patients admitted to pediatric intensive care units in 3 regional non-teaching hospitals between Jan 2006 and December 2006 were included. The data required for PIM1 and PIM2 were collected. The probabilities of death predicted by PIM1 and PIM2 were validated against the actual probability of mortality. RESULTS: Of the 540 patients were included in this study, only seven deaths were observed (1.3%). PIM 1 and PIM 2 overestimated the mortality rate by giving a greater mortality rate. (PIM1: 13.4, Standardized mortality ratio = 0.52, 95% CI = 0.14 to 0.91; PIM2: 14.2, Standardized mortality ratio =0.49, 95% CI = 0.13 to 0.86) The discrimination of PIM1 and PIM2 were satisfactory as reflected by area under receiver-operator characteristic curve of 0.889 (95% CI: 0.703 to 1.000) and 0.904 (95% CI: 0.738 to 1.000) respectively. Calibration was not possible due to insufficient death cases. CONCLUSIONS: The current study showed that PIM2 and PIM1 had similar accuracy in mortality prediction in Hong Kong. The current study also demonstrated the difficulty to calibrate the mortality model when actual mortality rate is low.
OBJECTIVE: To compare the probabilities of death predicted by Pediatric Index of Mortality 2 (PIM2) and Pediatric Index of Mortality 1 (PIM1) models against actual mortalities in 3 pediatric intensive care units in Hong Kong. METHODS: All consecutive patients admitted to pediatric intensive care units in 3 regional non-teaching hospitals between Jan 2006 and December 2006 were included. The data required for PIM1 and PIM2 were collected. The probabilities of death predicted by PIM1 and PIM2 were validated against the actual probability of mortality. RESULTS: Of the 540 patients were included in this study, only seven deaths were observed (1.3%). PIM 1 and PIM 2 overestimated the mortality rate by giving a greater mortality rate. (PIM1: 13.4, Standardized mortality ratio = 0.52, 95% CI = 0.14 to 0.91; PIM2: 14.2, Standardized mortality ratio =0.49, 95% CI = 0.13 to 0.86) The discrimination of PIM1 and PIM2 were satisfactory as reflected by area under receiver-operator characteristic curve of 0.889 (95% CI: 0.703 to 1.000) and 0.904 (95% CI: 0.738 to 1.000) respectively. Calibration was not possible due to insufficient death cases. CONCLUSIONS: The current study showed that PIM2 and PIM1 had similar accuracy in mortality prediction in Hong Kong. The current study also demonstrated the difficulty to calibrate the mortality model when actual mortality rate is low.