Pediatric Mortality Score Outperforms Neonatal Gold Standards in Surgical NICU Study

A mortality prediction score designed for children’s intensive care units may be just as good as—and in some cases better than—the specialized scoring systems traditionally used for newborns, according to new research from one of the world’s most renowned pediatric hospitals. The study, conducted at Great Ormond Street Hospital in London, suggests that the Paediatric Index of Mortality 3, or PIM-3, could reshape how clinicians assess risk in the growing number of surgical neonatal units that operate independently of maternity services.

The findings, published in Frontiers in Pediatrics, come at a time when specialized children’s hospitals increasingly manage critically ill newborns in dedicated neonatal intensive care units embedded within pediatric tertiary referral centers. These units differ fundamentally from conventional neonatal units attached to delivery suites: their patients are often transferred from other hospitals, may be admitted days or weeks after birth, and frequently arrive with complex surgical conditions that traditional neonatal scoring systems were never designed to capture.

For decades, neonatologists have relied on tools such as the Score for Neonatal Acute Physiology II (SNAP II) and its perinatal extension (SNAPPE II) to estimate mortality risk in newborns. These scores incorporate variables collected during the first twelve hours of admission, including mean blood pressure, lowest body temperature, oxygenation ratios, blood pH, seizure activity, and urine output. SNAPPE II additionally factors in Apgar scores, birth weight, and whether the infant is small for gestational age—perinatal characteristics long considered cornerstones of neonatal risk assessment.

The PIM-3 score, by contrast, was developed for pediatric intensive care populations and draws on a different set of parameters measured within the first hour of admission: pupil reactivity, mechanical ventilation status, systolic blood pressure, base excess, fractional inspired oxygen concentration, and arterial partial pressure of oxygen. It also accounts for whether the admission was planned or unplanned and whether the patient carries a high-risk primary diagnosis, such as necrotizing enterocolitis or severe intracranial hemorrhage. Rather than assigning categorical points, PIM-3 uses an algorithm that multiplies each variable by a calibrated constant to generate a mortality probability expressed as a percentage.

Notably absent from PIM-3 are gestational age, birth weight, and Apgar scores—the very variables that many argue are indispensable for assessing newborn vulnerability. This omission has been the principal objection to applying the pediatric score in neonatal settings. The Great Ormond Street team, led by Rachel Atkinson and Maria Chalia, set out to test whether that objection holds up in a population where acute surgical pathology may matter more than perinatal history.

The researchers conducted a retrospective analysis of all admissions to the Great Ormond Street Neonatal Intensive Care Unit during calendar year 2022. Of 233 confirmed admissions, 197 patients met the inclusion criteria after exclusions for stays shorter than twelve hours, incomplete data, or missing PIM-3 calculations. The cohort comprised 122 preterm and 75 term neonates, with a median age at admission of seven days. Seventeen patients died, yielding a mortality rate of 8.6 percent in the study population—consistent with previously reported UK figures.

Using receiver operating characteristic analysis, the team calculated the area under the curve (AUC) for each score as a measure of discriminatory ability. PIM-3 achieved the highest AUC at 0.788, followed by SNAPPE II at 0.738 and SNAP II at 0.706. All three exceeded the conventional threshold of 0.7 for acceptable discrimination, though none reached 0.8, underscoring the inherent difficulty of predicting outcomes in this heterogeneous surgical population. At their optimal cut-off values—9 percent for PIM-3, 30 for SNAPPE II, and 11 for SNAP II—the PIM-3 score delivered the best balance of sensitivity and specificity, achieving 77 percent and 76 percent respectively, compared with notably lower specificities for the two neonatal scores.

Subgroup analyses revealed striking patterns. In preterm neonates, PIM-3 was the only score to achieve statistically significant mortality discrimination, with an AUC of 0.794. Among ventilated infants, PIM-3 again stood alone with a significant AUC of 0.858, while SNAP II fell to 0.601. The score also performed remarkably well in small-for-gestational-age neonates, reaching an AUC of 0.932, though the authors caution that this subgroup contained only 46 patients. Conversely, in term neonates, SNAPPE II outperformed PIM-3 with an AUC of 0.907, likely reflecting the greater relevance of perinatal variables in that group. All three scores performed poorly in planned admissions, where patients are typically physiologically stable despite requiring complex interventions.

The authors propose that the strong performance of PIM-3—despite its omission of gestational age and birth weight—may reflect the nature of mortality in tertiary surgical neonatal care. Conditions such as necrotizing enterocolitis, congenital diaphragmatic hernia, and severe intracranial pathology drive risk largely independently of perinatal characteristics, and PIM-3 explicitly incorporates several of these high-risk diagnoses while the neonatal scores do not. Furthermore, many patients in this cohort were admitted well beyond the immediate postnatal period, with preterm infants arriving at a median of fifteen days of life, by which point their physiology has diverged substantially from the neonatal transition that SNAP-based scores were designed to model.

The study also found that PIM-3 correlated more strongly with length of stay than the other two indices, though the relationship remained modest and the difference was not statistically significant. The authors emphasize that their findings require validation in larger, prospective, multicenter cohorts before any change in clinical practice, noting that the retrospective single-center design, small number of deaths, and reliance on retrospective documentation all limit the strength of their conclusions. Nevertheless, as surgical neonatal units proliferate within specialized children’s hospitals in the United Kingdom and beyond, the prospect of a single, rapidly calculable score that performs comparably to neonatal-specific tools offers an appealing path toward more consistent benchmarking and risk stratification in this increasingly distinct patient population.

Subject of Research: Comparison of PIM-3, SNAPPE II, and SNAP II mortality prediction scores in a tertiary surgical neonatal intensive care unit

Article Title: Can PIM-3 score predict risk for neonatal mortality vs. SNAPPE II and SNAP II scores in a tertiary surgical neonatal intensive care unit?

Article References: Atkinson, R., Ray, S., Hannam, S., & Chalia, M. (2026). Can PIM-3 score predict risk for neonatal mortality vs. SNAPPE II and SNAP II scores in a tertiary surgical neonatal intensive care unit?. Frontiers in Pediatrics, 14, Article 1933189. https://doi.org/10.3389/fped.2026.1933189

Image Credits: AI Generated

DOI: 10.3389/fped.2026.1933189

Keywords: neonatology, PIM-3, SNAPPE II, SNAP II, mortality prediction, neonatal intensive care, pediatric critical care, Great Ormond Street Hospital, score, predict, risk, neonatal