EuroSCORE II Cardiac Surgery Risk Calculator

The full published logistic model — intercept and all 27 coefficients from Table 6 of the derivation paper — with the calibration findings that say where a 2010 model over- and under-predicts today.

Last updated 8 Oct 2026

EuroSCORE II

Logistic model · 18 inputs

Age (years)The model does not use the age itself. Table 6’s note reads: the age variable is 1 if the patient is 60 or under, and increases by one per year thereafter — so 1 at 60, 2 at 61, 3 at 62. That floor is the whole difference between a correct implementation and one that quietly credits every patient under 60 with negative risk; at 40 a missing floor subtracts 0.57 from the linear predictor. The 2003 logistic EuroSCORE on the additive EuroSCORE uses a different offset again — one at 59 and under — so the two models are a year apart on the same patient.

SexFemale sex carries a positive coefficient in this model: the paper states plainly that “females have a higher mortality than males” in cardiac surgery.

Renal functionCreatinine clearance by the Cockcroft–Gault formula, which the paper names — not by CKD-EPI and not by an eGFR off the report, because Cockcroft–Gault carries weight and the others do not (see CKD-EPI 2021). Note what the published table actually says: a dialysis patient’s coefficient (0.642) is lower than a clearance of 50 or less without dialysis (0.859). That is not a transcription error on this page — it is the fitted model, and the usual reading is that established dialysis is a managed state while unsupported severe impairment is not.

Extracardiac arteriopathyAs published: claudication; carotid occlusion or more than 50 per cent stenosis; amputation for arterial disease; or intervention, past or planned, on the abdominal aorta, limb arteries or carotids.

Poor mobility“Severe impairment of mobility secondary to musculoskeletal or neurological dysfunction.” Not a frailty index and not deconditioning; functional capacity in METs is the separate question.

Previous cardiac surgery“One or more previous major cardiac operation involving opening the pericardium.” The second largest coefficient in the model, after salvage surgery.

Chronic pulmonary dysfunctionThe 2012 paper gives no definition for this item, which is a real gap: the 1999 additive model defined it as long-term use of bronchodilators or steroids for lung disease, and that is the definition in general use. Postoperative pulmonary risk in non-cardiac surgery is ARISCAT‘s question, not this one’s.

Active endocarditis“Patients still on antibiotic treatment for endocarditis at the time of surgery.” Whether the case is endocarditis is the Duke criteria‘ question.

Critical preoperative stateAny one or more, in the same admission as the operation: ventricular tachycardia or fibrillation or aborted sudden death; cardiac massage; ventilation before arrival in the anaesthetic room; inotropes; intra-aortic balloon counterpulsation or a ventricular assist device before arrival in the anaesthetic room; acute renal failure with anuria or oliguria under 10 mL/h. It is a closed list, not a judgement.

Insulin-dependent diabetesInsulin treatment, as in Revised Cardiac Risk Index — the route of treatment, not the diagnosis and not the control. A patient on metformin with a very high HbA1c scores zero.

NYHA class for dyspnoeaThe paper’s own four one-line definitions, quoted above. The model takes the class, not functional capacity in METs.

CCS class 4 angina“Inability to perform any activity without angina or angina at rest.” Class 4 only; classes 1 to 3 score nothing. A patient who has presented with possible ischaemia is GRACE‘s question, not this model’s.

Left ventricular functionFour categories, and note how little separates the bottom two: 0.808 against 0.935. The step that matters is from moderate to poor, which more than doubles the contribution. The 1999 additive model had only two categories split at 30 per cent.

Recent myocardial infarctionNinety days, as published — and a smaller coefficient than female sex, which surprises most readers.

Pulmonary artery systolic pressureThe paper contradicts itself on these cut-offs. Table 6 prints “31–55 mmHg” and “≥55”; the Results text prints “30–55 mmHg” and “56 mmHg and above”. The two readings disagree about 30, 31 and 55 mmHg, and nothing in the paper resolves it. This record follows Table 6, which is the table the coefficients come from. See also pulmonary vascular resistance.

UrgencySalvage carries the largest coefficient in the whole model.

Weight of the interventionIsolated CABG is the reference level. The second level is all but weightless — 0.0062118, which is to say that in this cohort a single non-coronary operation was no riskier than a bypass. The jump is from one procedure to two.

Surgery on the thoracic aortaCounted in addition to the weight of the intervention, not instead of it. The paper gives no separate definition.

6.15%Example

72-year-old woman for coronary bypass plus aortic valve replacement; creatinine clearance 70 mL/min; insulin-treated diabetes; chronic lung disease; NYHA III; ejection fraction 40 per cent; elective

The model, and what is known about its calibration

predicted mortality = e^y / (1 + e^y) where y = −5.324537 + Σ β_ix_i
age term: β = 0.0285181, x = 1 at age 60 or under, +1 per year thereafter

the intercept: −5.324537. A patient with every risk factor absent and an age of 60 or under still carries the age term once, so the floor of the model is y = −5.324537 + 0.0285181 = −5.2960189, which is 0.50 per cent. The model cannot return zero and cannot return one: it is bounded strictly inside (0, 1) by construction, and equals exactly 0.5 when y is 0
discrimination: the derivation paper reported an area under the ROC curve of 0.8095 (95% CI 0.7820–0.8360) in its validation set, against 0.7896 for the 2003 logistic EuroSCORE on the same data. A meta-analysis of 22 studies and 145 592 procedures pooled it at 0.792 (0.773–0.811). A systematic review of four Latin American studies and 8 372 patients found a mean area of 0.77, with individual studies at 0.76, 0.76, 0.80 and 0.81
calibration, which is the part that moves: the same 22-study meta-analysis pooled the observed-to-expected ratio at 1.019 (0.899–1.139) — close to ideal overall — but found the model over-predicted in isolated coronary bypass (O/E 0.829, 0.677–0.982) and under-predicted in high-risk patients (O/E 1.253). The Latin American review found far worse: observed 7.08 per cent against 3.89 per cent expected, a stated ratio of 2.04, with per-study ratios of 1.00, 1.43, 2.19 and 2.46. That review’s own totals — 593 observed deaths against 326 expected — give 1.82 rather than 2.04, and it does not explain the gap; both numbers are printed here because neither can be preferred from what is published
what it leaves out: frailty, which is not in the model at all; body mass; anaemia; the specific valve; coronary anatomy; and every outcome except in-hospital death — no stroke, no renal failure, no prolonged ventilation, no reoperation for bleeding. The STS Predicted Risk of Mortality models several of those and cannot be reproduced anywhere, because its coefficients are not published
which operation this is for: major cardiac surgery in adults. This model is for CARDIAC surgery. The RCRI and ARISCAT are for non-cardiac surgery and were derived in cohorts from which cardiac operations were excluded; the two sets of numbers are not interchangeable in either direction.
and it is not the last word: the EuroSCORE project’s own site says a third model, EuroSCORE 3, is in development and expected to be larger and more accurate. A reader holding a EuroSCORE should know which one it is, exactly as MELD, MELD-Na and MELD 3.0 are three numbers

Worked example

72-year-old woman for coronary bypass plus aortic valve replacement; creatinine clearance 70 mL/min; insulin-treated diabetes; chronic lung disease; NYHA III; ejection fraction 40 per cent; elective
Age term: 72 − 59 = 13, so x = 13 and the contribution is 0.0285181 × 13 = 0.3707353. Not 72 × 0.0285181, and not 13 × the age
Female 0.2196434 + clearance 51–85 0.303553 + chronic pulmonary dysfunction 0.1886564 + insulin-dependent diabetes 0.3542749 + NYHA III 0.2958358 + moderate LV 0.3150652 + two procedures 0.5521478 = 2.2291765
y = −5.324537 + 0.3707353 + 2.2291765 = −2.7246252
e^−2.7246252 / (1 + e^−2.7246252) = 6.15 per cent
The same patient on the additive EuroSCORE scores 8 additive points and 11.19 per cent by the 2003 logistic EuroSCORE. Three models, one patient, 6.15 against 8 against 11.19 — and the 2010 cohort that produced the lowest of the three is the only one whose observed mortality was measured this century
Change the operation to isolated bypass and the weight term falls from 0.5521478 to 0
Change nothing else but make it a salvage operation and y rises by 1.362947 to −1.3616782, giving 20.41 per cent — one item, a threefold change, which is what a logistic model does and an additive one cannot

Every coefficient in the published model

Table 6 of the derivation paper, which prints a standard error, a z value and a confidence interval beside each of these. Read the column rather than the labels: a recent infarction weighs less than being female, and one non-coronary operation weighs almost nothing against an isolated bypass.
Variable	Level	β
Constant	—	−5.324537
Age	x = 1 at 60 or under, +1 per year after	0.0285181
Sex	Female	0.2196434
Renal impairment	Creatinine clearance 51–85 mL/min	0.303553
	Clearance 50 or less, not on dialysis	0.8592256
	On dialysis, whatever the creatinine	0.6421508
Extracardiac arteriopathy	Present	0.5360268
Poor mobility	Present	0.2407181
Previous cardiac surgery	Present	1.118599
Chronic pulmonary dysfunction	Present	0.1886564
Active endocarditis	Present	0.6194522
Critical preoperative state	Present	1.086517
Insulin-dependent diabetes	Present	0.3542749
NYHA	II / III / IV	0.1070545 / 0.2958358 / 0.5597929
CCS class 4 angina	Present	0.2226147
LV function	Moderate 31–50% / poor 21–30% / very poor ≤20%	0.3150652 / 0.8084096 / 0.9346919
Recent myocardial infarction	Within 90 days	0.1528943
PA systolic pressure	31–55 mmHg / 55 or above	0.1788899 / 0.3491475
Urgency	Urgent / emergency / salvage	0.3174673 / 0.7039121 / 1.362947
Weight of intervention	1 non-CABG / 2 procedures / 3 or more	0.0062118 / 0.5521478 / 0.9724533
Surgery on the thoracic aorta	Present	0.6527205

A 2010 model, read in 2026

EuroSCORE II is a logistic regression with an intercept of −5.324537 and 27 coefficients, published in full in Table 6 of its derivation paper. That matters more than it sounds: many cardiac risk models are distributed only as a web form, and a model whose coefficients are printed can be audited, recalculated and argued with. The Society of Thoracic Surgeons’ Predicted Risk of Mortality cannot be, because its coefficients are not published. EuroSCORE II can, which is why this page exists.

The arithmetic has three traps. The first is the age term, which is not the age: the published variable is 1 at 60 and under and rises by one per year thereafter, and the 2003 logistic EuroSCORE uses the same trick with a different offset. The second is the renal item, which is a Cockcroft–Gault creatinine clearance and not an eGFR, so the number off the laboratory report is the wrong number. The third is that dialysis carries a smaller coefficient (0.6421508) than an unsupported clearance of 50 or less (0.8592256). Readers assume one of those is a typo. Neither is.

Calibration is where the honest version of this page lives. Pooled across 22 studies and 145 592 procedures the observed-to-expected ratio is 1.019 — close to ideal — but that average hides two opposite errors: the model over-predicts in isolated coronary bypass (ratio 0.829) and under-predicts in high-risk patients (1.253). Outside Europe it can fail much harder. A systematic review of four Latin American studies found observed mortality of 7.08 per cent against 3.89 per cent expected, with one of the four studies calibrated perfectly and another out by a factor of 2.46. Which direction the model errs in depends on the population, so there is no safe general correction to apply.

What follows is the thing to hold on to: the cohort was recruited over twelve weeks in 2010, and surgical mortality, case mix and transcatheter alternatives have all moved since. A model’s predicted probability is not this patient’s probability: it is the rate observed in the cohort the model was fitted to, among patients who shared these inputs and differed in everything the model does not record. Frailty, body mass, anaemia, the specific valve and the surgeon are all in that second category. Current guidance uses a risk score alongside functional capacity in METs, NT-proBNP and a multidisciplinary discussion rather than instead of them. Every figure here comes from a named cohort, and cohorts differ in case mix, era, outcome definition and treatment; where your own institution’s protocol differs, it takes precedence. This page reports what a stratum predicted in a named study. It recommends no action.

Frequently asked questions

Why does the model subtract 59 from the age?

It does not, quite. The published age variable is 1 for any patient aged 60 or under and increases by one per year thereafter, which is max(1, age − 59). The floor is the part implementations get wrong: without it, a 40-year-old gets a negative age contribution of −0.54 and the model credits youth with protection it never measured, because the derivation cohort had almost nobody that young.

Which creatinine clearance should I use?

Cockcroft–Gault, which the paper names explicitly. An eGFR from CKD-EPI is a different quantity — it is normalised to body surface area and does not use weight — and substituting it changes which of the three renal levels a patient falls into, most often in the direction of looking better. The levels are above 85, 51 to 85, and 50 or less, with dialysis as a separate level whatever the creatinine.

Is EuroSCORE II still accurate?

Its discrimination has held up: pooled area under the curve 0.792 across 22 studies and 145 592 procedures, against 0.8095 in the original validation. Its calibration has not held up uniformly. It over-predicts in isolated coronary bypass and under-predicts in high-risk patients, and in some non-European cohorts it under-predicts by a factor approaching 2.5. A score that ranks patients well can still attach the wrong absolute number to each of them, and those are separate properties.

Can I use this for a transcatheter valve procedure?

No. The cohort was surgical — 22 381 patients having cardiac surgery in 2010, when transcatheter aortic valve implantation was confined to patients judged inoperable. A surgical model’s predicted surgical mortality says nothing about a different procedure, and this page does not compare the two.

Someone quoted me a EuroSCORE of 8. What is that?

Ambiguous: 8 additive points, 8 per cent from the 2003 logistic model, or 8 per cent from this one. For the worked patient above those readings are 8 points, 11.19 per cent and 6.15 per cent. Ask which model — the additive EuroSCORE shows the first two side by side.

Related calculators

References

Nashef SAM, Roques F, Sharples LD, et al. EuroSCORE II. Eur J Cardiothorac Surg. 2012;41(4):734–745. Read in full from the EuroSCORE project’s own copy; Table 6 carries the intercept and all 27 coefficients with their standard errors.
An independent reproduction of EuroSCORE II’s full coefficient table to four decimal places, read as the corroborating source; all 27 coefficients and the intercept agree with the derivation paper to the rounding.
Guida P, Mastro F, Scrascia G, Whitlock R, Paparella D. Performance of the European System for Cardiac Operative Risk Evaluation II: a meta-analysis of 22 studies involving 145 592 cardiac surgery procedures. J Thorac Cardiovasc Surg. 2014;148(6):3049–3057.
Rendimiento de EuroSCORE II en Latinoamérica: una revisión sistemática. Rev Med Chile. Four studies, 8 372 patients from Argentina and Brazil; the review’s own totals and its stated pooled ratio disagree, and both are quoted here.
The EuroSCORE project’s own site, read for the model’s history, for the statement that the 2003 logistic model “significantly overpredicted mortality”, for the curation by Royal Papworth Hospital and for the absence of any terms of use, copyright notice or licence statement.
Halvorsen S, Mehilli J, Cassese S, et al. 2022 ESC Guidelines on cardiovascular assessment and management of patients undergoing non-cardiac surgery. Eur Heart J. 2022;43(39):3826–3924.

Not medical advice. For healthcare professionals and education. Reference intervals vary by laboratory and assay — always use your own laboratory's. Never base a dose or a treatment decision on this page alone. Full disclaimer at calcengines.com/disclaimer/