Performance of the pediatric index of mortality (PIM-3) in a Moroccan PICU: challenges in resource-limited settings

Prognostic scores such as the Pediatric Index of Mortality (PIM-3) are widely used to estimate mortality risk in PICUs, yet their performance in low- and middle-income countries (LMICs) remains uncertain. We aimed to evaluate the predictive performance of PIM-3 in a Moroccan PICU; associations with ICU length of stay (LOS) and duration of mechanical ventilation (MV) were secondary exploratory outcomes. We conducted a prospective cohort study in the Mother-Child PICU at Hassan II University Hospital (Fez, Morocco) from June 1 to September 30, 2024, including all children aged 1 month to 16 years. PIM-3 variables were collected at admission. The primary focus was model performance: discrimination (AUROC, AUPRC), calibration (Hosmer-Lemeshow), and standardized mortality ratio (SMR). Univariable and multivariable regressions were used for the secondary exploratory analyses (mortality correlates, LOS, and MV duration). Among 122 patients, observed mortality was 28.7%, exceeding PIM-3 predictions (SMR = 7.7; p < 0.001). PIM-3 showed good discrimination (AUROC = 0.86; AUPRC = 0.65) but poor calibration. In secondary exploratory analyses, higher PIM-3 scores, lower GCS, multiple organ failures, and need for MV were associated with mortality. MV increased ICU stay duration and organ failures prolonged ventilation duration.

In this Moroccan PICU, PIM-3 showed good discrimination but poor calibration and thus underestimated mortality. These findings support considering local recalibration in future work, and exploration of machine-learning-based adaptation for LMIC settings.  Clinical trial number: Not applicable.

• PIM-3 is widely used to estimate PICU mortality risk. It was developed and calibrated largely in high-resource settings. • External validations often show good discrimination but variable calibration, with under- or over-estimation in LMICs due to differences in case-mix, care processes, and resources.

• In a prospective Moroccan PICU cohort (n=122), PIM-3 showed good discrimination (AUROC 0.86; AUPRC 0.65) but markedly underestimated mortality (SMR 7.7). • Higher PIM-3 risk, lower Glasgow Coma Scale, multiple organ failures, and mechanical ventilation were associated with death, underscoring the need for local recalibration and exploration of machine learning-based adaptations for LMICs.