Maddrey vs MELD: which score for alcohol-associated hepatitis?

How the Maddrey and MELD scores differ

The Maddrey and MELD scores differ in what they were built for and in what they measure. The Maddrey discriminant function was made for alcohol-associated hepatitis: Maddrey and colleagues derived it in 1978 from a trial of 55 patients, and Carithers and colleagues gave the modified form in 1989, 4.6 × the prolongation of the prothrombin time in seconds plus the serum bilirubin in mg/dL. The Model for End-Stage Liver Disease (MELD) was first created to estimate survival after a transjugular intrahepatic portosystemic shunt (TIPS); Kamath and colleagues validated it in 2001 across broader groups with end-stage liver disease, and it now ranks patients for liver transplantation. MELD combines bilirubin, creatinine and the INR on a logarithmic scale, and MELD-Na adds serum sodium. Both scores use bilirubin and a clotting test. Only MELD includes kidney function, and only MELD uses the INR, which standardizes the prothrombin time across laboratories, where the Maddrey score needs the local control time.

What AASLD, EASL and ACG guidance say

The AASLD, EASL and ACG guidance all accept both scores, but they lean in different directions. The 2018 EASL Clinical Practice Guidelines call the Maddrey discriminant function the first score that reliably identified patients at highest risk of early death, and the most widely used in practice and in trials; in its modified version a cut-off of 32 identifies severe disease and is usually the threshold for starting specific therapy. EASL notes that retrospective studies suggest patients with a MELD above 20 are at high risk of 90-day death, and its treatment algorithm uses an mDF of 32 or more or a Glasgow alcoholic hepatitis score of 9 or more. The 2019 AASLD practice guidance states that the MDF of 32 or more should be used to assess the need for corticosteroids or other medical therapies, and that a MELD above 20 should also prompt that consideration. The 2024 ACG guideline calls the mDF the historical index and defines severe disease as MELD above 20.

All three discuss corticosteroids for severe disease when there is no contraindication, and EASL and AASLD use the Lille score after seven days of treatment to decide whether it is working. The ACG adds that patients with MELD scores from 25 to 39 derive the most benefit. These pages give scores only, never drug amounts.

Which score predicts death better

MELD has predicted short-term death better than the Maddrey score in the largest comparisons, though the gap is modest. Dunn and colleagues at the Mayo Clinic studied 73 patients with alcoholic hepatitis from 1995 to 2001 and found MELD was the only independent predictor of death; its c-statistic for 30-day mortality was 0.83 against 0.74 for the discriminant function, and 0.86 against 0.83 at 90 days, differences that were not statistically significant. A MELD of 21 predicted 90-day death with 75% sensitivity and 75% specificity. The global study by Morales-Arráez and colleagues, 2,581 patients from 85 centers in 11 countries published in 2022, found that the area under the ROC curve for the mDF was significantly lower than for every other score, and its authors concluded that the mDF should no longer be used for prognosis. In the post hoc analysis of 1,068 patients from the STOPAH trial cited by the ACG, the mDF scored 0.67 against 0.704 to 0.726 for MELD, Glasgow and ABIC.

Neither static score captures response to treatment. Louvet and colleagues built the Lille model in 2007 from 320 patients treated with corticosteroids; for death at six months its AUROC was 0.89, against 0.66 for the Maddrey score, because it adds the change in bilirubin over the first seven days.

Worked examples: when Maddrey and MELD agree and disagree

Two hypothetical patients show where the Maddrey and MELD scores give the same answer and where they part. The first has a prothrombin time of 22 seconds against a laboratory control of 12, a bilirubin of 12 mg/dL, an INR of 1.9, a creatinine of 1.0 mg/dL and a sodium of 132 mmol/L. The prolongation is 10 seconds, so the modified discriminant function is 4.6 × 10 + 12 = 58.0, well above 32. MELD, calculated with the US transplant network's rules, is 23 and MELD-Na 26, both above 20, so every guideline calls the disease severe. The second has a prothrombin time of 15 seconds against a control of 12, a bilirubin of 8 mg/dL, an INR of 1.3, a creatinine of 2.0 mg/dL and a sodium of 130 mmol/L. The mDF is 4.6 × 3 + 8 = 21.8, below 32, yet MELD is 24 and MELD-Na 28, mostly because of the creatinine that the Maddrey formula leaves out.

Under the EASL 2018 mDF cut-off the second patient is not severe, while the ACG 2024 cut-off of MELD above 20 and the AASLD statement on MELD would flag the case. Enter both patients in the Maddrey discriminant function calculator and the MELD calculator to check the numbers.