Validation of an Electronic Pediatric Index of Mortality 2 Score in a Mixed Quaternary PICU

Emily L Joyce1,2,3, Christine M Crana1, Jonathan Yabes4,5

  • 1Division of Nephrology, Department of Pediatrics, UPMC Children's Hospital of Pittsburgh, Pittsburgh, PA.

Insights

An electronic Pediatric Index of Mortality 2 score accurately predicts outcomes in pediatric intensive care units. This validated electronic health record tool offers reliable patient risk assessment.

Area of Science:

  • Pediatric critical care medicine
  • Health informatics
  • Clinical epidemiology

Background:

  • The Pediatric Index of Mortality 2 (PIM2) score is a validated tool for assessing pediatric intensive care unit (PICU) patient mortality risk.
  • Electronic health records (EHRs) offer a rich data source for clinical research, but require validated scoring systems for retrospective analysis.

Purpose of the Study:

  • To assess the validity and accuracy of an electronic version of the Pediatric Index of Mortality 2 score (ePIM2) using retrospective EHR data.
  • To determine if ePIM2 can reliably predict mortality in a mixed-ICU population.

Main Methods:

  • A retrospective observational study was conducted in pediatric and cardiac ICUs.
  • The PIM2 score was adapted for retrospective EHR data analysis.
  • The ePIM2 was validated against manually ascertained PIM2 scores in 100 patients and assessed for discrimination, calibration, and accuracy in 12,582 patient encounters.

Main Results:

  • Nine out of 11 PIM2 components showed good agreement (ICC or Cohen κ ≥ 0.6) between manual and electronic derivations.
  • The ePIM2 demonstrated good discrimination (AUC = 0.89), appropriate calibration (no significant difference between observed and expected deaths), and excellent predictive ability (Brier score = 0.0135).

Conclusions:

  • The electronic version of the Pediatric Index of Mortality 2 score is a valid and accurate tool for retrospective analysis of EHR data.
  • ePIM2 can be reliably used for risk stratification and outcome prediction in large mixed-ICU cohorts.
Abstract