PIM3 score (logit) = 3.8233 × pupils + 0.9763 × ventilated − 0.5378 × elective + 0.0671 × |base excess| − 0.0431 × SBP + 0.1716 × (SBP² ÷ 1000) + 0.4214 × (FiO₂ × 100 ÷ PaO₂) − 1.2246 × bypass-cardiac recovery − 0.8762 × non-bypass-cardiac recovery − 1.5164 × non-cardiac recovery + 1.6225 × very-high-risk diagnosis + 1.0725 × high-risk diagnosis − 2.1766 × low-risk diagnosis − 1.7928, where each indicator is 1 when present and 0 otherwise. Predicted mortality (probability) = 1 ÷ (1 + e^−logit). No severity bands are published, so none are shown. THE THREE DIAGNOSIS TIERS ARE ONE VARIABLE, NOT THREE. Very-high-risk outranks high-risk, which outranks low-risk, and only the highest applies. Counting two tiers together is the commonest porting defect: on the paper’s own worked example it yields 72.34% instead of 47.22%. MISSING VALUES SUBSTITUTE THE MODEL’S OWN FIGURES rather than being dropped or read as normal: SBP → 120, base excess → 0, oxygenation term → 0.23 (a PIM3 correction; PIM2 used 0). This is the ordinary path, not an edge case, PaO₂ having been missing in 55.8% of the derivation cohort. Systolic BP also carries three coded entries that are not measurements: cardiac arrest at admission → 0, shocked with an unmeasurable BP → 30, unknown → blank, which the model reads as 120. The paired SBP terms are U-shaped with a minimum near 125.6 mmHg, which is how the arrest code acquires its weight, about +2.70 logit against the default.

PIM3 estimates hospital-mortality probability from data at first ICU contact, for unit-level case-mix and SMR benchmarking. The derivation paper states the limit itself: “These models are not intended for prognostic use on individual patients” (Straney 2013). MEASUREMENT WINDOW AND CODING. Use the FIRST value of each variable from first face-to-face ICU-team contact to 1 hour after ICU arrival, not the worst. Pupils count only when both are larger than 3 mm and fixed to bright light; a drug, toxin or direct eye-injury explanation does not count. Ventilated in the first hour covers invasive ventilation, mask or nasal CPAP, BiPAP and negative-pressure ventilation. A tracheostomy breathing spontaneously without support is no, which is an ANZPIC Registry data-entry convention (January 2019 booklet) rather than a rule in the paper, and the paper is silent on the case. Elective means the admission could have been deferred by more than 6 hours without harm. Each of the three diagnosis-tier lists is complete as published and applies to the main reason for admission; if you are unsure, record none. Where a condition from more than one tier applies, the model applies the highest tier only. POST-LIVER-TRANSPLANT ADMISSIONS ARE CODED OPPOSITELY BY THE TWO CUSTODIAN REGISTRIES. ANZPICR excludes planned post-transplant recovery from liver failure; PICANet includes it. This score follows ANZPICR, the stricter reading. Both agree that a readmission for graft failure qualifies. AGE. Read the model as applying to children younger than 16. The paper CONTRADICTS itself here, its abstract saying younger than 18 and its inclusion criteria younger than 16, and the contradiction is in the source rather than resolved here. CALIBRATION TRAVELS FAR WORSE THAN DISCRIMINATION. Italy AUC 0.88 with SMR 0.98; Argentina 0.83 with SMR 1.3; South Africa 0.81 with SMR 1.28, its highest SMR of 6.67 falling in the LOWEST-risk decile. The Gulf has its own evidence. Dubai (n = 583): AUC 0.78 with an overall SMR of 0.53, yet SMR 2.1 in SEPSIS, an under-prediction inside an over-predicting unit, and that sepsis signal is the finding that survives its own paper. Riyadh (n = 3,396): sufficient discrimination, poor calibration, worst in infants under 12 months. Newborns are systematically over-predicted, sitting below the SBP nadir, and haemato-oncology admissions are under-predicted, observed mortality 18.73% against 7.13% predicted. Recalibrate and monitor locally before comparative use. DO NOT MIX COEFFICIENT SETS. ANZICS publishes regional recalibrations, PIM3-anz13 and PIM3-anz15, whose coefficients are entirely different. Registry exports use sentinel values, 999 meaning unknown, so any future import path must map them before scoring. PICANet publishes citable plausibility ranges for systolic BP, PaO₂ and base excess; no registry publishes ranges for platelets, bilirubin, creatinine or MAP, because none collects them.

References

Straney L, Clements A, Parslow RC, et al; ANZICS Paediatric Study Group and PICANet. Paediatric index of mortality 3: an updated model for predicting mortality in pediatric intensive care. Pediatr Crit Care Med. 2013;14(7):673–681.Derivation paper: the 13 coefficients and intercept (Table 3, p677), the three diagnosis-tier lists with their qualifying rules and the precedence rule, and the variable coding and missing-value conventions (Appendix 1, p680). The ANZICS 'PIM2 & PIM3 for the ANZPIC Registry — Information Booklet (Version Jan 2019)' is a supporting document for registry data entry, and the only source for the rule that a tracheostomy with unassisted spontaneous breathing is not ventilation, a rule the paper does not address. It is grey literature with no DOI and is no longer retrievable at its published URL (HTTP 404), so it is credited here rather than carried as its own reference.PMID 23863821 DOI 10.1097/PCC.0b013e31829760cf
Wolfler A, Osello R, Gualino J, et al; Italian Network of Pediatric Intensive Care Units. The importance of mortality risk assessment: validation of the Pediatric Index of Mortality 3 score. Pediatr Crit Care Med. 2016;17(3):251–256.Italian multicentre validation: AUC 0.88, SMR 0.98 (Hosmer-Lemeshow p = 0.21). Source for the neonatal over-prediction observed for both PIM2 and PIM3, and for the measured cost of migrating a PIM2 cohort to PIM3 (roughly one admission in eleven changes risk tier).DOI 10.1097/PCC.0000000000000657
Lee OJ, Jung M, Kim M, Yang HK, Cho J. Validation of the Pediatric Index of Mortality 3 in a Single Pediatric Intensive Care Unit in Korea. J Korean Med Sci. 2017;32(2):365–370.Independent reproduction of the full PIM3 equation, probability transform, and the risk-diagnosis lists. Source for the haemato-oncology under-prediction (c-index 0.66 against 0.74–0.83 in other subgroups; observed mortality 18.73% against 7.13% predicted).DOI 10.3346/jkms.2017.32.2.365
Arias López MdP, Fernández AL, Ratto ME, et al. Pediatric Index of Mortality 3: an evaluation of function among ICUs in Argentina. Pediatr Crit Care Med. 2018;19(12):e653–e661.Argentine multicentre evaluation: AUC 0.83, SMR 1.3, Hosmer-Lemeshow p < 0.001. Source for the observation that HIV infection and post-liver-transplant admission — both dropped from the model as non-predictive in the derivation population — remain associated with higher mortality in a resource-varied setting.DOI 10.1097/PCC.0000000000001741
Solomon LJ, Morrow BM, Argent AC. Paediatric Index of Mortality scores: an evaluation of function in the Paediatric Intensive Care Units of a South African province. Pediatr Crit Care Med. 2021;22(9):813–821.South African multicentre evaluation: AUC 0.81, SMR 1.28, Hosmer-Lemeshow p < 0.001, with the highest SMR (6.67) in the LOWEST risk decile. The closest published comparator for deployment in a resource-varied setting, being the only multicentre evaluation of PIM3 in one.DOI 10.1097/PCC.0000000000002693
Baloglu O, Nagy LR, Sonawane A, et al. Simplified Pediatric Index of Mortality 3 score by explainable machine learning algorithm. Crit Care Explor. 2021;3(10):e0561.Source for the scale of real-world missingness in the PIM3 blood-gas inputs: base excess missing in 97.2% and the oxygenation ratio in 97.3% of a single-centre US series — the reason the imputation path is the ordinary path rather than an edge case.DOI 10.1097/CCE.0000000000000561
Malhotra D, Nour N, El Halik M, Zidan M. Performance of Pediatric Index of Mortality 3 score in a tertiary pediatric ICU in Dubai. Dubai Med J. 2019;3(1):19–25.Latifa Hospital, Dubai; single centre, n = 583 with 46 deaths (7.9%). Three findings are stable and are what this page rests on: AUC 0.78 (95% CI 0.69–0.87), an overall SMR of 0.53 — the model OVER-predicted deaths across the unit — and SMR 2.1 in the sepsis subgroup, an UNDER-prediction that nothing else in the paper contradicts. Its predicted-probability strata do contradict each other and are carried as unstable rather than as a finding: SMR 2.67 in the 1–5% band (severe under-prediction) against SMR 0.33 below a predicted probability of 14.3% and 0.72 above it (over-prediction across that same low range). Both cuts are this study's own, so neither direction can be asserted for the low end of the scale.DOI 10.1159/000505205
Alkhalifah AS, AlSoqati A, Zahraa J. Performance of pediatric risk of mortality III and pediatric index of mortality scores in a tertiary pediatric intensive care unit in Saudi Arabia. Front Pediatr. 2022;10:926686.King Fahad Medical City, Riyadh; n = 3396, children under 14. The models it evaluated had, in its own words, 'sufficient discrimination ability and poor calibration', and both the worst calibration and the worst discrimination were in infants under 12 months. The one per-model figure carried here is PRISM III's (best in the 60–120-month band, AUC 0.87); no PIM3-specific statistic from this study is asserted, because none was captured in the review that supplied it.DOI 10.3389/fped.2022.926686

Reproduction rights

Freely reproducible.The formula, its 13 coefficients, the intercept, the logistic transform and the tier-precedence rule are mathematical facts / a method and are freely implementable (pim3.md IP status). Which conditions sit in which risk tier is likewise a fact and is used; the condition names are ordinary clinical terms. The paper's and the ANZICS booklet's descriptive prose — the qualifying rules, the pupil descriptor and the SBP special-value instructions — is paraphrased in this project's own words rather than transcribed (pim3.md IP FLAG).

Other mortality and severity scores

Pediatric Risk of MortalityPhysiologic severity in the first hours of PICU care, with PRISM IV mortality

Important. For use by qualified health professionals as an informational and educational aid. It supports clinical judgment and does not replace it. Verify every result independently before making a clinical decision. This is not a medical device.