About the EVendo Score for Oesophageal Varices Needing Treatment
Most patients with cirrhosis who undergo screening endoscopy turn out to have no varices or only small ones, and the EVendo score exists to identify who can safely skip the procedure. Six routinely available items feed it — INR, AST, platelet count, blood urea nitrogen, haemoglobin and the presence of ascites — combined as a ratio rather than a points total. At the published cut-off of 3.90 or below, applying the score to the derivation population would have spared 30.5% of endoscopies while missing only 2.8% of varices needing treatment, rising to 40.0% of endoscopies spared and 1.1% of VNT missed in Child-Turcotte-Pugh class A cirrhosis. It was developed with a random forest algorithm rather than conventional regression, which is unusual among liver scores.
Formula
A = (8.5 × INR) + (AST / 35); B = (platelets / 150) + (BUN / 20) + (haemoglobin / 15); EVendo = A / B + (1 if ascites present)- INR
- International normalised ratio, multiplied by 8.5 — the largest weight in the formula.
- AST
- In U/L, divided by 35.
- platelets
- In ×10³/µL (equivalent to ×10⁹/L), divided by 150.
- BUN
- Blood urea nitrogen in mg/dL, divided by 20.
- haemoglobin
- In g/dL, divided by 15.
- ascites
- Adds 1 to the final ratio, not to the numerator — it is applied after the division.
- The ascites point is added to the quotient, after A has been divided by B. Adding it inside the numerator gives a different and incorrect result.
- This is a ratio, not a sum, so the terms interact: the same platelet count contributes differently depending on the BUN and haemoglobin alongside it.
- BUN is not the same as urea. Where a laboratory reports urea in mmol/L, convert to BUN in mg/dL by dividing by 0.357 — this is the most likely unit error outside the United States.
- Because ascites adds a full point, its presence alone pushes most patients above the 3.90 threshold, which is clinically appropriate but means the score adds little in patients with obvious ascites.
Interpreting the result
A score of 3.90 or below identifies a low-probability group in whom screening endoscopy may reasonably be deferred, with clinical follow-up and re-scoring at intervals rather than discharge from surveillance. Above 3.90, varices needing treatment cannot be excluded and endoscopy should proceed. The figures behind the threshold matter for how confidently it is applied: in the full derivation population it would have avoided 30.5% of endoscopies at the cost of missing 2.8% of varices needing treatment, whereas in Child-Turcotte-Pugh class A patients it would have avoided 40.0% while missing only 1.1%. That difference is the practical message — the score is safest and most useful in compensated disease, and a clinician should be more cautious about deferring endoscopy in a patient with more advanced cirrhosis. Discrimination for varices needing treatment specifically (AUROC 0.74–0.75) was consistently lower than for varices of any size (0.82–0.84).
| Score | Band | What it means | Action |
|---|---|---|---|
| ≤ 3.90 | Low probability of varices needing treatment | Would have spared 30.5% of endoscopies overall (40.0% in CTP class A) while missing 2.8% of VNT (1.1% in CTP class A) | Screening endoscopy may reasonably be deferred, with clinical follow-up and re-scoring later — not discharge from surveillance |
| > 3.90 | Screening endoscopy indicated | Varices needing treatment cannot be excluded | Proceed with screening upper endoscopy |
What the EVendo Score needs (6 inputs)
- INR
- Weighted 8.5 in the numerator, making it the dominant single term — small changes in INR move the score substantially.
- AST (U/L)
- Divided by 35 in the numerator, so it contributes on a scale of multiples of the upper reference range.
- Platelet count (×10³/µL)
- Divided by 150 in the denominator, so falling platelets raise the score — the portal-hypertension signal.
- Blood urea nitrogen (mg/dL)
- Divided by 20 in the denominator.
- Haemoglobin (g/dL)
- Divided by 15 in the denominator, so anaemia raises the score.
- Ascites
- Adds a flat 1 point to the final result when present.
Units. The formula expects US conventional units: BUN in mg/dL (not urea in mmol/L — divide mmol/L by 0.357), haemoglobin in g/dL (divide g/L by 10), platelets in ×10³/µL (numerically the same as ×10⁹/L), and AST in U/L. The BUN conversion is the commonest error outside the United States.
What it returns
- EVendo score
- A continuous ratio, reported to two decimal places.
- Screening recommendation
- Low probability of varices needing treatment (≤ 3.90) or screening endoscopy indicated (> 3.90).
How it is calculated
Dong and colleagues collected data on 238 patients with cirrhosis undergoing screening endoscopy at three Los Angeles hospitals between January 2016 and December 2017, recording sex, age, ethnicity, platelets, haemoglobin, sodium, AST, ALT, bilirubin, INR, albumin, urea nitrogen, creatinine, cirrhosis aetiology, and the presence of ascites and hepatic encephalopathy. Rather than fitting a logistic regression, they applied a random forest algorithm to identify which factors were associated with varices and with varices needing treatment. Six emerged, and the resulting formula was then tested prospectively in 109 further patients at the same centres during 2018. The machine-learning derivation explains the score's unusual ratio structure — the form was chosen to fit the data rather than for interpretability, which is why it does not resemble a conventional points scale.
Facts & figures
| Outcome | Training (n = 238) | Validation (n = 109) | CTP class A (n = 235) |
|---|---|---|---|
| Oesophageal varices (any) | AUROC 0.84 | AUROC 0.82 | AUROC 0.81 |
| Varices needing treatment | AUROC 0.74 | AUROC 0.75 | AUROC 0.75 |
Performance for varices needing treatment — the clinically decisive outcome — is consistently lower than for varices of any size, which is the score's main statistical limitation.
| Population | Endoscopies spared | VNT missed |
|---|---|---|
| All patients | 30.5% | 2.8% |
| Child-Turcotte-Pugh class A | 40.0% | 1.1% |
The score performs best where the pre-test probability is lowest — compensated cirrhosis — which is also where deferring endoscopy is most attractive.
Evidence
Derivation — Dong (random forest)
2019 · n = 238238 patients with cirrhosis undergoing screening oesophagogastroduodenoscopy at three separate hospitals in Los Angeles from January 2016 through December 2017. Demographic, laboratory and clinical variables — including platelets, haemoglobin, sodium, AST, ALT, bilirubin, INR, albumin, urea nitrogen, creatinine, aetiology, ascites and hepatic encephalopathy — were analysed with a random forest algorithm to identify predictors of varices and of varices needing treatment.
The resulting EVendo score, based on INR, AST, platelets, urea nitrogen, haemoglobin and ascites, identified oesophageal varices with an AUROC of 0.84 in training and varices needing treatment with an AUROC of 0.74. A cut-off below 3.90 would have spared 30.5% of patients an endoscopy while missing only 2.8% of varices needing treatment.
Prospective validation — Dong
2019 · n = 109109 patients undergoing screening endoscopy prospectively at the same three medical centres from January through December 2018, used to test the accuracy of the derived formula.
AUROC 0.82 for varices of any size and 0.75 for varices needing treatment, closely reproducing the training performance. In the 235 patients with Child-Turcotte-Pugh class A cirrhosis, AUROCs were 0.81 and 0.75, and the same cut-off would have spared 40.0% of endoscopies while missing 1.1% of varices needing treatment.
How it compares
EVendo Score vs Baveno VI criteria (liver stiffness < 20 kPa and platelets > 150)
Baveno VI is the guideline-endorsed route to avoiding endoscopy but requires transient elastography; EVendo reaches a similar decision from blood tests alone, which makes it usable where elastography is not.
The Baveno VI criteria spare screening endoscopy in compensated advanced chronic liver disease when liver stiffness is below 20 kPa and platelets exceed 150 × 10⁹/L, and they carry consensus backing that EVendo does not. Their limitation is access: elastography is unavailable in many services and unreliable in obesity and ascites. EVendo needs only a full blood count, a liver panel, urea and a clinical examination, so it extends the same decision to settings without elastography — at the cost of a lower evidence base and no guideline endorsement.
EVendo Score vs Child-Pugh score
Child-Pugh grades overall cirrhosis severity and does not predict varices; EVendo predicts varices needing treatment specifically, and its own performance is stratified by Child-Pugh class.
The two are complementary rather than competing, and the relationship runs one way: knowing a patient is Child-Turcotte-Pugh class A is what tells you EVendo is at its most reliable, where it spared 40.0% of endoscopies while missing 1.1% of varices needing treatment. Child-Pugh itself was never designed to answer whether a varix needs banding, and using it that way would be a misapplication.
EVendo Score vs Glasgow-Blatchford score
Opposite clinical moments — EVendo decides whether an asymptomatic patient with cirrhosis needs elective screening endoscopy, while the Glasgow-Blatchford score triages a patient who is already bleeding.
The pairing is worth distinguishing because both scores use haemoglobin and urea and both bear on whether endoscopy is needed, which invites confusion. Glasgow-Blatchford is a pre-endoscopy risk score in acute upper gastrointestinal bleeding, identifying who can be managed as an outpatient. EVendo operates before any bleeding has occurred, in surveillance. A patient with acute variceal bleeding needs urgent therapeutic endoscopy regardless of any EVendo score.
Pearls & pitfalls
- The ascites point is added after the division, not to the numerator. Getting this wrong changes the result substantially.
- BUN is not urea. Outside the United States, divide urea in mmol/L by 0.357 to obtain BUN in mg/dL — otherwise the denominator is badly wrong.
- INR carries a weight of 8.5 and dominates the numerator, so it is the value most worth double-checking, particularly in a patient on anticoagulation where the INR does not reflect liver function at all.
- The score predicts varices needing treatment less well (AUROC ~0.75) than varices of any size (~0.83). Read the headline discrimination figures carefully, because the lower number is the clinically relevant one.
- A low score defers endoscopy; it does not exclude varices. Re-score at follow-up rather than treating a single result as a permanent clearance.
- It is most reliable in Child-Turcotte-Pugh class A cirrhosis. Be more cautious about deferring endoscopy in class B or C, where fewer validation data support the threshold.
- Because ascites alone adds a full point, the score contributes little in a patient with clinically obvious ascites, who will almost always cross the threshold anyway.
Critical actions
- Convert urea to BUN in mg/dL before calculating if your laboratory does not report BUN directly.
- Apply the ascites point to the quotient, after dividing A by B.
- Arrange endoscopy without delay for any score above 3.90, and do not use a borderline score to justify deferral.
- Re-score patients at surveillance intervals rather than treating one low result as lasting reassurance — portal hypertension progresses.
- Check whether the INR is raised by anticoagulation rather than by liver disease, since that inflates the score's dominant term and will falsely trigger endoscopy.
- Consider the Baveno VI criteria alongside EVendo where transient elastography is available, as the two can be used together to strengthen a decision to defer.
Why this score exists
The authors' opening premise was economic and ethical rather than diagnostic: endoscopic screening for varices is, in their framing, a potentially deferrable procedure that increases patient risk and healthcare cost, and most of the patients who undergo it are found to have no varices or only small ones. That framing dictated the design target. They aimed not at detecting varices — which would have been the easier statistical problem, and indeed the score does it better — but at detecting varices *needing treatment*, because a small varix found at endoscopy changes nothing. The choice of a random forest over regression follows the same pragmatism: the goal was a formula that worked on readily available data, not one whose coefficients told a physiological story, which is why the published expression is a ratio with no obvious clinical interpretation.
About the creator
First author, 2019 derivation and validation study
Applied a random forest algorithm to routinely available data to derive the score, then tested it prospectively.
Senior author
Led the work aimed at safely deferring screening endoscopy in patients unlikely to have varices needing treatment.
Limitations
- Discrimination for varices needing treatment — the outcome that actually matters — is only moderate, with AUROCs of 0.74 to 0.75, notably lower than for varices of any size.
- The cut-off still misses varices needing treatment in 2.8% of patients overall, so deferral is a calculated risk rather than a safe exclusion.
- It was derived and validated at three hospitals in a single city, with no independent multi-centre validation in the original publication.
- Validation data are strongest in Child-Turcotte-Pugh class A cirrhosis; the threshold is less well supported in class B and C.
- The random forest derivation produced a formula with no physiological interpretation, which makes it harder to reason about when a component is confounded.
- INR is heavily weighted and is unreliable in anticoagulated patients, in whom the score will be systematically inflated.
- It carries no guideline endorsement comparable to the Baveno criteria, so deferring endoscopy on EVendo alone is a departure from consensus recommendations.
- BUN is affected by renal function, diet, gastrointestinal bleeding and hydration, none of which relate to portal hypertension.
If you are the patient
The EVendo score helps doctors decide whether someone with cirrhosis needs a camera test (endoscopy) to look for enlarged veins in the food pipe, called varices. Most people with cirrhosis who have this test turn out to have no varices, or only small ones that need no treatment, so avoiding an unnecessary procedure is worthwhile. The score uses five blood results — a clotting test (INR), a liver enzyme (AST), your platelet count, a kidney test (urea), and your haemoglobin — plus whether you have fluid in your abdomen. A score of 3.90 or below suggests the test can safely be put off for now, with continued monitoring and re-checking later. Above that, the camera test is recommended. It is not a perfect test: in the original study, deferring the procedure at this cut-off missed a small number of varices that did need treatment, which is why your team will keep reviewing you rather than dropping the check altogether.
Frequently asked questions
What is the EVendo score cut-off?#
3.90. A score of 3.90 or below identifies patients unlikely to have varices needing treatment, in whom screening endoscopy may reasonably be deferred with continued follow-up. Above 3.90, endoscopy should proceed.
How many endoscopies does the EVendo score avoid?#
At the 3.90 cut-off it would have spared 30.5% of endoscopies overall while missing 2.8% of varices needing treatment, and 40.0% in Child-Turcotte-Pugh class A cirrhosis while missing only 1.1%.
How accurate is the EVendo score?#
AUROC was 0.84 in training and 0.82 in prospective validation for varices of any size, but 0.74 and 0.75 for varices needing treatment — the lower figure is the clinically relevant one.
Is EVendo better than the Baveno VI criteria?#
Not better, but more widely applicable. Baveno VI carries consensus endorsement and requires transient elastography plus a platelet count; EVendo needs only blood tests and an examination, so it works where elastography is unavailable, at the cost of a smaller evidence base.
Does EVendo use urea or BUN?#
Blood urea nitrogen (BUN) in mg/dL. If your laboratory reports urea in mmol/L, divide by 0.357 to convert. Entering urea directly will give an incorrect score.
Can a low EVendo score rule out varices?#
No. It identifies a low-probability group in whom endoscopy can reasonably be deferred, not excluded — 2.8% of varices needing treatment were still missed at the cut-off, so continued surveillance and re-scoring are essential.