Pediatric Risk of Mortality (PRISM III and PRISM IV)

CategoryMortality and severity
Versionv2.2.1
Reviewed2026-08-04
ValidationIndependent clinical validation: pending

Assessment

Data collection window

First 4 hours of PICU care (PRISM IV)First 12 hours of PICU care (PRISM III-12)First 24 hours of PICU care (PRISM III-24)

The window sets which model your data belong to, so it is asked rather than assumed. It changes nothing about the score, which is computed identically for all three. It decides only whether a mortality estimate can be shown: the first 4 hours is PRISM IV, whose equation is published in full and shown here (laboratory values from 2 hours before admission through hour 4), and it shows a probability only once all four admission-context questions have been answered — leave any of them blank and the score still appears while the probability is withheld, because a blank is not an answer of "no". The 12- and 24-hour windows are PRISM III, whose mortality equations are not published in the source article and are separately licensed, so those windows give the score and its two subscores and no probability. The four admission-context questions belong to PRISM IV alone and are ignored on those windows.

Age

Sets the age band for blood pressure, heart rate, creatinine, BUN and PTT, and is separately a term in the PRISM IV equation. PRISM III bands: neonate under 1 month, infant 1 to under 12 months, child 12 months to under 12 years, adolescent 12 years and over. Accepted 0–18 years

Cardiovascular

Systolic blood pressure (lowest)

mmHgkPa

The lowest value recorded in the window. Accepted 0–300 mmHg

Temperature (lowest)

Rectal, oral, blood or axillary. Under 33 °C scores 3; a patient with both a low and a high excursion still scores 3 once, not twice. Accepted 20–45 °C

Temperature (highest)

Above 40.0 °C scores 3, on the same single row. Accepted 20–45 °C

Heart rate (highest)

Do not assess during crying or iatrogenic agitation. Thresholds are age-banded. Accepted 0–350 bpm

Neurological

Glasgow Coma Scale (lowest)

Enter ONLY for a patient with known or suspected acute CNS disease; leave blank otherwise. Do not assess within 2 hours of sedation, paralysis or anaesthesia — use the closest period free of them. Under 8 scores 5. One of the two neurologic items. Accepted 3–15

Pupillary reflexes

Both reactiveOne fixed, one reactiveBoth fixed

A non-reactive pupil must be larger than 3 mm. Do not assess after iatrogenic dilation. At 11 points this is the heaviest single item in the score, and the other neurologic item.

Acid–base and blood gas

pH (lowest)

Arterial, capillary or venous. Shares one row with the lowest total CO₂ — whichever is worse scores once, never both. Accepted 6.5–8

pH (highest)

A separate row from the lowest pH, and both can score in the same patient: the source is explicit that points may be assigned for the low and the high ranges together. Accepted 6.5–8

Total CO₂ (lowest)

Total CO₂ or bicarbonate. Use a calculated bicarbonate from a blood gas only if total CO₂ is not measured routinely. Accepted 0–60 mmol/L

Total CO₂ (highest)

Above 34.0 scores 4, independently of the lowest value's row. Accepted 0–60 mmol/L

PCO₂ (highest)

mmHgkPa

Arterial, capillary or venous. Accepted 0–200 mmHg

PaO₂ (lowest)

mmHgkPa

Arterial measurements only. Accepted 0–700 mmHg

Chemistry

Glucose (highest)

mg/dLmmol/L

Above 200 mg/dL scores 2. A whole-blood value should be increased by 10% before scoring. Accepted 0–1500 mg/dL

Potassium (highest)

mEq/Lmmol/L

Above 6.9 scores 3. A whole-blood value should be increased by 0.4 mmol/L before scoring. Accepted 0–15 mEq/L

Creatinine (highest)

mg/dLµmol/L

Age-banded; infant and child share one cutoff. Accepted 0–25 mg/dL

Blood urea nitrogen (highest)

mg/dLmmol/L

A two-band split only: neonates above 11.9 mg/dL, everyone else above 14.9. Accepted 0–300 mg/dL

Haematology

White blood cell count (lowest)

Only leukopenia scores: under 3,000 scores 4, and a high count scores nothing. Accepted 0–200000 cells/mm³

Platelet count (lowest)

Unusually shaped: a merely low-normal 150,000 already scores 2, 50,000-99,999 scores 4, and under 50,000 scores 5. Accepted 0–2000000 cells/mm³

Prothrombin time (highest)

Shares one row with PTT. Above 22.0 s at any age; the row awards 3 once even when both analytes qualify. Accepted 0–200 s

Partial thromboplastin time (highest)

Above 85.0 s in a neonate, above 57.0 s at every other age. Accepted 0–400 s

One physiologic score; one published mortality model. Seventeen variables are scored against age-banded thresholds and summed: 0 to 74, decomposing into a neurologic subscore (pupillary reflexes 0-11 plus mental status 0-5, maximum 16) and a non-neurologic subscore (the remaining fifteen variables, maximum 58). Several rows have shapes that are easy to get wrong. Acidosis is one row satisfied by either the lowest pH or the lowest total CO₂, awarded once at the worse tier. The highest pH is a separate row, so a patient whose pH swung from 6.9 to 7.6 scores on both. Total CO₂ likewise scores once at the low end and again at the high end. Prothrombin and partial thromboplastin time share a single row. That score is computed the same way whichever window you collected. Only PRISM IV turns it into a probability, and it does not use the total: it weights the neurologic subscore at 0.197 per point and the non-neurologic at 0.163, adds age, admission source, pre-admission CPR, cancer and low-risk system of primary dysfunction, and finishes with P = 1 / (1 + e^-R). PRISM III's own mortality equations are not published in the source article — Pollack 1996 prints the score sheet in full but contains no regression coefficients, in any table and with no supplement — and they are separately licensed, so this platform presents the physiologic score only for the 12- and 24-hour windows.

PRISM estimates mortality risk for a POPULATION and is a case-mix and benchmarking instrument, not a bedside prognosis for the child in front of you. Summed across a cohort it gives an expected death count for a standardised mortality ratio; applied to one patient it says nothing actionable. NO MORTALITY FIGURE IS SHOWN FOR THE 12- AND 24-HOUR WINDOWS, and that is a deliberate absence rather than a missing feature or a zero. PRISM III's mortality equations are not published in the source article: Pollack 1996 prints the score sheet in full — every variable, every range, every point value — but contains no regression coefficients, in any of its tables and with no supplement, and the paper's author note reserves the equations for research use while stating that non-research uses may attract compensation. They are therefore separately licensed, and this platform presents the physiologic score only for those windows: the total, the neurologic subscore and the non-neurologic subscore. The authors' own network does the same, publishing a PRISM III calculator that returns the score and its subscores and no probability. The score itself is unaffected and is identical for all three windows. PRISM IV, the 4-hour window, does show a probability: its coefficients are printed in full in Pollack 2016, a paper whose stated objective included placing the algorithms in the public domain. IT SHOWS ONE ONLY WHEN ALL FOUR ADMISSION-CONTEXT QUESTIONS HAVE BEEN ANSWERED — admission source, CPR in the 24 hours before admission, cancer, and a low-risk system of primary dysfunction. Leave any one of them blank and the score and its two subscores still appear while no probability does. This is deliberate and is the same rule as above: each of those four contributes nothing at its reference level, so treating a blank as an absent term would not compute a probability without it, it would compute the reference patient — admitted from the operating room, no CPR, no cancer, no low-risk system — and hand that curve to everyone who skipped a question. An unanswered question is never answered with a default. Those four questions belong to PRISM IV alone and are ignored on the 12- and 24-hour windows. Read what that number is precisely — the estimated probability of HOSPITAL mortality, for a FIRST PICU admission, from data collected in the first 4 hours of PICU care — where PRISM III's outcome was PICU mortality on 1990s practice. The windows are not interchangeable: PRISM IV takes the first 4 hours with laboratory values from 2 hours before admission, PRISM III the first 12 or 24 hours, so this calculator asks which you collected. Blank components score zero, so a partially entered score reads lower than the patient is. Mental status should be entered only for known or suspected acute CNS disease, and not within 2 hours of sedation, paralysis or anaesthesia. Whole-blood chemistry needs correcting before entry: glucose up by 10%, sodium by 3 mmol/L, potassium by 0.4 mmol/L. [NEEDS SOURCE]: no published worked example exists for either model, so the test cases were constructed from the threshold table and verified by arithmetic rather than against a published case; the authors' own CPCCRN calculators are the natural oracle, and both calculators' input and output sets were read on 2026-08-03, but no case has yet been round-tripped through either one, so the constructed cases remain unreconciled against the authors' own implementation. The patent's printed neonate heart-rate band appears to contain an OCR error (215-255 against a >225 cutoff on the next line) and 215-225 is used here, following an independent reproduction. The glucose row prints 200 mg/dL and 11.0 mmol/L as if equivalent when 200 mg/dL is 11.1 mmol/L; the mg/dL limb is authoritative here. REGIONAL CALIBRATION, WHERE THIS PLATFORM ACTUALLY DEPLOYS. A Riyadh PICU series (Alkhalifah 2022, n = 3396, children under 14) evaluated paediatric mortality models and found them to have sufficient discrimination and poor calibration; PRISM III discriminated best in the 60-120-month age band (AUC 0.87), and the WORST calibration and the worst discrimination were both in infants under 12 months — the youngest patients, scored by a model that separates them least well. Part of the same shape is on record from Dubai for the neighbouring model: PIM3 in a single Dubai PICU (Malhotra 2019, n = 583, 46 deaths, 7.9%) reached AUC 0.78 with an overall SMR of 0.53, over-predicting deaths for the unit as a whole, while UNDER-predicting by 2.1 in the sepsis subgroup. That second study is a PIM3 finding and not a PRISM one; it is carried here because what generalises is the pattern — acceptable discrimination alongside calibration that does not travel — which is what both series show. WHAT NEITHER SERIES SUPPORTS IS A CLAIM ABOUT THE LOW END OF THE PROBABILITY SCALE, and the Dubai paper is itself the reason: cut finely it reports SMR 2.67 among children whose predicted probability was 1-5%, severe under-prediction, while cut coarsely at a predicted probability of 14.3% it reports SMR 0.33 below the cut against 0.72 above it, over-prediction across that same low range. Both cuts come from that one cohort of 583 and they point in opposite directions, so both are recorded here rather than whichever one is convenient. THE CONCLUSION A READER NEEDS: discrimination travels between populations, calibration frequently does not, and the under-prediction that survives its own paper is the one in sepsis rather than one attached to any band of the probability scale. Three limits on what those studies say about the figure on this page. They evaluated PRISM III, not PRISM IV, so nothing here asserts a regional evaluation of the model that produces this calculator's probability. Every Dubai figure above is PIM3's. And neither model has been recalibrated for this region, so a PRISM IV probability shown here is an uncalibrated population estimate, least trustworthy in the youngest infants.

References

Pollack MM, Patel KM, Ruttimann UE. PRISM III: an updated Pediatric Risk of Mortality score. Crit Care Med. 1996;24(5):743-752.The derivation paper, and the source of the score. It publishes the score sheet in full (Figure 1) and NO regression coefficients: its eight tables were enumerated against the full text and Table 3, the one the Results section points to for the risk-factor models, compares model fit (chi-square, df, AIC, AUC, Hosmer-Lemeshow) rather than listing coefficients. There is no supplement. Its author note reserves the mortality equations for research use and states that non-research uses may attract compensation.PMID 8706448
Pollack MM, Holubkov R, Funai T, et al. The Pediatric Risk of Mortality Score: Update 2015. Pediatr Crit Care Med. 2016;17(1):2-9.PRISM IV. Source of the subscore split and every coefficient in Table 3. Its stated objective included placing the algorithms in the public domain.PMID 26492059 DOI 10.1097/PCC.0000000000000558
Pollack MM. Method, apparatus and medium for allocating beds in a pediatric intensive care unit and for evaluating quality of care. US patent 5,809,477. 1998.Primary source for the full PRISM III threshold table and the scoring notes quoted in the help text — the parts of the score the 1996 paper also publishes. Status: Expired - Lifetime, anticipated expiration 2015-09-21. It also states mortality equations the paper does not, but those are NOT implemented: the document is a single source for them, its transcription carries known internal inconsistencies, and it offers no page or table to cite.Source
Collaborative Pediatric Critical Care Research Network. PRISM IV calculator.The authors' own implementation, and the natural oracle for reconciling this one. Input and output sets read 2026-08-03: its input list matches Table 3 of Pollack 2016 one-to-one — same variables, same categories, same reference levels — and its reference age band tops out at 18 years. No case has been round-tripped through it, so the constructed fixtures here remain unreconciled against it.Source
Collaborative Pediatric Critical Care Research Network. PRISM III calculator.Retrieved 2026-08-03. Takes the 17 physiologic variables and an age band and returns SCORE, NEUROLOGIC and NON-NEUROLOGIC — no mortality, and it collects no risk factors with which to produce one. Pollack's own network had the coefficients and shipped the score without them, which is the practice this calculator now matches.Source
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. Its conclusion for the models it evaluated was 'sufficient discrimination ability and poor calibration', with the worst calibration AND discrimination in infants under 12 months. PRISM III discriminated best in the 60-120-month band (AUC 0.87). It evaluated PRISM III, not PRISM IV.DOI 10.3389/fped.2022.926686
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.A PIM3 evaluation, not a PRISM one, and cited here for that reason explicitly. Latifa Hospital, Dubai; n = 583, 46 deaths (7.9%). Stable findings: AUC 0.78 (95% CI 0.69-0.87), overall SMR 0.53, and SMR 2.1 in sepsis. Its predicted-probability strata contradict each other and are carried as unstable: SMR 2.67 in the 1-5% band against SMR 0.33 below a predicted probability of 14.3% and 0.72 above it, so the same paper shows under-prediction and over-prediction in the same low range depending on where the bands are cut. It is the second Gulf data point for the pattern that does hold — discrimination survives the move between populations, calibration does not — which is why this page carries it alongside the Riyadh series rather than only naming its own model.DOI 10.1159/000505205

Reproduction rights

Freely reproducible.The PRISM III score — the variables, their age-banded ranges and every point value — is published in full in Pollack 1996 (Figure 1) and reproduced verbatim in US patent 5,809,477 (Pollack), which shows status 'Expired - Lifetime' with an anticipated expiration of 2015-09-21. PRISM IV's coefficients are printed in Table 3 of Pollack 2016 (PMID 26492059), whose stated objective included 'placing the algorithms (Pediatric Risk of Mortality IV) in the public domain', and the authors' own network publishes a free public calculator of it. PRISM III's MORTALITY equations are a separate matter and are NOT shipped: they appear in no table of the 1996 article, that paper's author note reserves them for research use and states that non-research uses may attract compensation, and their only source is the patent's transcription, which carries known internal inconsistencies and offers no page to cite. They were removed on 2026-08-03; see the changelog.

Independent clinical validation: pending

Two independent clinical validators will be named here once review is complete.

Validator slot open 1

Validator slot open 2

2026-07-31v1.0.0Initial releasePRISM III and PRISM IV as one calculator: one physiologic score, and the mortality equation matching the collection window.
2026-08-03v2.0.0Output withdrawnREMOVED: the PRISM III predicted-mortality percentage. The 12-hour and 24-hour windows no longer show any mortality figure — they now show the PRISM score, the neurologic subscore and the non-neurologic subscore, and nothing else. The 4-hour PRISM IV window is unchanged and still shows its probability. Why: the PRISM III mortality equations are not published in the source article. Pollack 1996 prints the score sheet in full but contains no regression coefficients in any of its eight tables and has no supplement, and the paper's author note reserves the equations for research use while stating that non-research uses may attract compensation — they are separately licensed. There is consequently no page this platform could cite for the numbers it was showing. Two further reasons the figure was worse than it looked: it came from the score-only model, which carries no risk-factor adjustment at all, so against the published risk-adjusted model it OVER-predicts for post-operative admissions and for an acute diagnosis of diabetes, and UNDER-predicts after pre-ICU cardiac massage, with cancer, and after a previous ICU admission; and one score maps to a spread of roughly 10% to 65% once those factors are applied, where a single number was shown to every patient. THE SCORE ITSELF NEVER CHANGED. No threshold, no age band, no point value and no subscore moved, in any window; only the mortality overlay was withdrawn. Where a mortality estimate is needed, PRISM IV is the model with published coefficients — collect the first 4 hours and select that window.
2026-08-03v2.1.0Output withdrawnWITHHELD: the PRISM IV probability, whenever an admission-context question is left blank. On the 4-hour window, admission source, CPR within the 24 hours before admission, cancer, and low-risk system of primary dysfunction must now each carry an answer. If any one of them is blank the calculator shows the PRISM score, the neurologic subscore and the non-neurologic subscore and no probability at all — the same honest absence the 12- and 24-hour windows already carry. Why: every one of those four contributes zero at its reference level, so a blank falling through to the equation did not compute a probability without that term, it computed the REFERENCE PATIENT — admitted from the operating room or post-anaesthesia care, no CPR, no cancer, no low-risk system — and handed that curve to every clinician who skipped a question. That is the reference-patient form of the defect removed on 2026-08-03, where one number stood in for every patient. An unanswered question is now answered with nothing rather than with a default. The four inputs are deliberately still NOT declared required: they belong to PRISM IV alone and mean nothing on the 12- and 24-hour windows, so an unconditional requirement would reject a legitimate score-only entry; the requirement is conditioned on the window instead. NO NUMBER MOVED. No threshold, age band, point value, subscore or coefficient changed, and a fully answered 4-hour entry returns exactly the probability it returned before. Also in this version: interpretationStatus moves from 'pending' to 'not-applicable', because PRISM IV outputs a continuous probability rather than a band and its calibration tables bin by predicted probability rather than by score, while PRISM III score-only has no published severity band at all — so nothing is awaiting a later pass and 'pending' was asserting strata that do not exist. And the provenance note about the authors' CPCCRN calculators is corrected: both calculators' input and output sets were read on 2026-08-03; what remains outstanding is that no case has been round-tripped through either, so the constructed fixtures stay unreconciled against the authors' own implementation.
2026-08-03v2.1.1ClarificationRelabels the two entries above. NOTHING ABOUT THE SCORE CHANGED — no threshold, age band, point value, subscore, coefficient or output — and the two summaries are word for word what they were. Both entries were tagged 'Formula correction', which was the closest label the platform had and was wrong in a way worth correcting: correcting a formula leaves the reader with the same output computed better, while v2.0.0 removed the PRISM III mortality percentage outright and v2.1.0 stops showing the PRISM IV probability when an admission-context question is blank. Both are withdrawals of a number a clinician previously read, which is a materially larger event than a corrected equation, and both now carry 'Output withdrawn'. The label is a new one, added to the platform's shared changelog vocabulary for exactly this case, so no other score's history is affected.
2026-08-04v2.2.0New referenceAdds regional calibration evidence for the region this platform deploys in. NOTHING ABOUT THE SCORE OR THE MODEL CHANGED — no threshold, age band, point value, subscore, coefficient or output moved, and a 4-hour entry returns the same probability it returned yesterday. A Riyadh PICU series (Alkhalifah 2022, n = 3396, under 14 years) evaluated paediatric mortality models and found sufficient discrimination with poor calibration; PRISM III discriminated best in the 60-120-month band (AUC 0.87), and both the worst calibration and the worst discrimination were in infants under 12 months. Alongside it, and labelled as the PIM3 finding it is, the Dubai evaluation of PIM3 (Malhotra 2019, n = 583, 46 deaths, 7.9%): AUC 0.78, overall SMR 0.53 — over-predicting for the unit as a whole — while under-predicting by 2.1 in sepsis and by 2.67 in the 1-5% predicted-probability band. The notes now draw the conclusion instead of leaving the reader to assemble it: discrimination travels between populations, calibration frequently does not, and the under-prediction sits in the low-probability band and in sepsis, which is where a reassuring number is most likely to be believed. Two limits are stated with it — both studies evaluated PRISM III rather than PRISM IV, so no regional evaluation of the model that produces this page's probability is claimed, and neither model has been recalibrated for this region.
2026-08-04v2.2.1ClarificationCorrects the conclusion v2.2.0 drew from the Dubai PIM3 series, which that paper's own other stratification contradicts. NOTHING ABOUT THE SCORE OR THE MODEL CHANGED — no threshold, age band, point value, subscore, coefficient or output, and a 4-hour entry returns the same probability. v2.2.0 concluded that the under-prediction sits in the low-probability band and in sepsis, resting the first half of that on SMR 2.67 in the 1-5% predicted-probability band. Malhotra 2019 also reports SMR 0.33 below a predicted probability of 14.3% against 0.72 above it — over-prediction, not under-prediction, across that same low range. Both figures are from one cohort of 583 and point in opposite directions depending on where the bands are cut, so the notes now carry BOTH, record the predicted-probability strata as unstable, and make no claim about the low end of the scale. The sepsis finding is untouched and stays prominent: it is the half nothing in the paper contradicts. The Riyadh figures, the PIM3-not-PRISM attribution and the statement that neither series evaluated PRISM IV are unchanged.

Other mortality and severity scores

Paediatric Index of Mortality 3 (PIM3)v1.2.2Open →

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.