Results

1) Data characterization

Study group comprised a total of 1809 patients admitted at three volunteers Portuguese PICU’s (Oporto, Coimbra, Lisbon), 977 male, with male:female ratio of 1.2:1 and a mean age of 4.6 years (range: 0.8 years). The average duration of internment was 7.7 days, during which 57.4% of patients were submitted to mechanical ventilation. Medical (59.9%) and surgical (37.3%) were the most common reasons for admission, being a minority of them (2.9%) justified by monitoring and prevention intentions. A majority of total admissions (68.6%) was unplanned. On the total 1809 admissions included in the analysis, 8.6% died in the admitted PICU. An additional 1.4% died before discharge from the hospital (Table 2).

Table 2 – Study group characteristics (Patients’ demographic, type of admission and general diagnostic categories).

Table 3 – Expected mortality and Standardized Mortality Ratio according to original score prediction.

PIM, PIM2, PRISM and PRISM III quantities for original scores predicted mortality percentage means were respectively 6, 5.3, 9.9 and 7.4, compared to 8.6 (%) of observed mortality. Concerning the Standardized Mortality Ratios, 1.43, 1.62, 0.87 and 1.16 were the obtained values, respectively (Table 3).

2) Discrimination and calibration assessment

Figure 1 – Receiver operating characteristics (ROC) curves.

Table 4 – Discrimination and calibration according to original scores prediction.

Hosmer-Lemeshow statistics showed good calibration for PRISM (p=0,305), PRISM III (p=0,375) and PIM (0,132), but poor calibration for PIM2 (p=0,027). Discrimination was generally good for all models, with areas under the receiver operating characteristic curves (AUC) ranged from 0,84 (PIM) to 0,91 (PRISM III) (Table 4).

2) Optimized version of PIM2 (C-PIM2)

In what concerns optimized version of PIM2’s (C-PIM2), discrimination and calibration tests information showed good discriminatory power (AUC= 0,87) and good calibration (p=0,477).