Objective
To describe the predictive and calibration capabilities of PIM and PRISM in infants and children following cardiac surgery.
Design
Between December 1997 and November 1998, 250 consecutive infants and children were studied. No child died in theatre. There were; 53 patients <1 month, 75 from 1 month-1 year and 122 >1 year. Median age 11.43 months (range 0.02-229). Survivors were defined by ICU discharge.
Results
Crude mortality was 6% (15/250) all deaths occurred in children <1 year old. Median age of death (range) was 0.33 months (0.02-11.83). Median time (range) to death was 53 h (2-264).
Calibration using the Hosmer-Lemeshow goodness of fit test, showed a ?2 16. 15, df8, significance 0.04 for PRISM and ?2 17.05, df8, significance 0.03 for PIM. Using a cut off at P = 0.5, sensitivity and specificity for PRISM was 98.3% and 33.3%, and 99.2% and 26.7% for PIM.
Conclusion
Neither PRISM, nor the new scoring system PIM are well calibrated for predicting individual mortality. However, despite the small numbers, the area under the ROC plot for PIM compares favourably with the original work by Shann et al. [1] (0.87 vs 0.83). Therefore we would concur with their conclusion that PIM is accurate enough to describe the risk of mortality in groups of children, and has the added advantage of needing less data collection than PRISM.
Table.
Observed vs (predicted) deaths and area under the ROC curve for PRISM and PIM
| Mortality% | <1 | 1-4.99 | 5-14.99 | 15-29.99 | >30 | Total | ROC |
|---|---|---|---|---|---|---|---|
| PRISM | 0 (0.42) | 0 (2.8) | 5 (3.22) | 4 (3.96) | 6 (5.78) | 15 (16.1) | 0.93 |
| No. | 73 | 113 | 33 | 19 | 12 | 250 | (0.88-0.97) |
| Mean risk | 0.58 | 2.52 | 9.78 | 20.88 | 48.22 | ||
| PIM | 0 (0.24) | 3 (3.79) | 4 (1.99) | 4 (4.64) | 4 (3.19) | 15 (13.8) | 0.87 |
| No. | 30 | 177 | 22 | 22 | 8 | 250 | (0.77-0.99) |
| Mean risk | 0.79 | 2.14 | 8.68 | 21.08 | 39.8 |
References
- Shann. et al. Intensive Care Med. 1997;23:201–207. doi: 10.1007/s001340050317. [DOI] [PubMed] [Google Scholar]