About the Maddrey's Discriminant Function for Alcoholic Hepatitis
A value of 32 or above is what defines severe alcoholic hepatitis and opens the corticosteroid conversation — that single threshold is why this score exists. The number itself comes from two laboratory values: DF = 4.6 × (patient PT − control PT, in seconds) + total bilirubin (mg/dL), with the control PT taken from the same assay run. The 32 cutoff has held since 1989, though the largest modern trial found prednisolone's 28-day mortality benefit at that threshold fell just short of statistical significance, which matters before treating the number as a promise of benefit.
Formula
DF = 4.6 × (patient PT − control PT, seconds) + total bilirubin (mg/dL)- patient PT − control PT
- The prolongation of the patient's prothrombin time over their own laboratory's control, in seconds. Both values must come from the same assay run — the formula was derived and validated using a control PT, not a normal-range midpoint or the international normalised ratio.
- total bilirubin
- Serum total bilirubin in mg/dL, entered without floor or cap.
- There is no unit toggle needed for the PT terms, but bilirubin is often reported in µmol/L outside the US — divide by 17.1 to convert to mg/dL before entering, or use the calculator's unit toggle.
- Some published variants substitute INR for the PT-ratio term (a 'modified' or INR-based discriminant function), which produces a different number from the original formula and is not the version the 32 threshold was validated against.
- The score has no upper bound; a very high value does not indicate a data-entry error the way a MELD score above 40 would.
Interpreting the result
Below 32, the trial evidence does not support starting corticosteroids, and management is supportive: abstinence support, nutrition, and treatment of alcohol withdrawal. At or above 32, corticosteroids are conventionally considered, but the strength of that recommendation should be read against the largest trial to date, which found a mortality benefit that did not reach statistical significance. In practice this means: cross the threshold, have the steroid conversation, but do not treat a DF of 32 as a mortality-benefit guarantee, and reassess with the Lille score at day 7 to decide whether to continue.
| Score | Band | What it means | Action |
|---|---|---|---|
| < 32 | Mild to moderate | Lower short-term mortality; steroid trials have consistently excluded this group | Abstinence support, nutrition, thiamine, and withdrawal management |
| ≥ 32 | Severe | 28-day mortality around 17% in the placebo arm of the largest modern trial; 66-day mortality was substantially higher (about 55%) in the smaller 1992 trial that first confirmed steroid benefit at this threshold | Exclude infection and GI bleeding, consider corticosteroids, reassess with Lille score at day 7 |
Scroll the table sideways for every column.
What the Maddrey's DF needs (3 inputs)
- Total bilirubin (mg/dL)
- Serum total bilirubin, entered directly into the equation without a floor or ceiling.
- Patient PT (seconds)
- The patient's prothrombin time in seconds, from the same assay run as the control.
- Control PT (seconds)
- The laboratory's control PT for that same run — not a textbook or remembered value. Using the wrong control value is the single commonest source of an incorrect discriminant function, because the formula depends on the difference between the two.
Units. Total bilirubin is entered in mg/dL in the base formula; outside the US it is often reported in µmol/L — divide by 17.1, or use the unit toggle, before entering. Both prothrombin times are in seconds; there is no SI equivalent conversion needed for that term, but they must be read from the same assay run.
What it returns
- Discriminant function
- A continuous number with no fixed ceiling. Reported to two decimal places; 32 is the only clinically actionable cutoff.
- Severity band
- Mild-to-moderate (below 32) or severe (32 or above).
How it is calculated
The function came from stepwise discriminant analysis of a 1978 randomized trial of prednisolone in alcoholic hepatitis, which found that prothrombin time prolongation and bilirubin level, of all the laboratory values measured, were the two independently associated with death. The equation weights PT prolongation heavily (a factor of 4.6 per second) because coagulopathy in this context reflects the liver's remaining synthetic capacity more directly than bilirubin does, which mainly reflects excretory function. The 32 threshold was not part of the original 1978 derivation — it was established over a decade later as the enrollment criterion for a methylprednisolone trial and has been used as the field's steroid-decision cutoff ever since.
Facts & figures
| Trial | Year | n | Mortality endpoint | Result |
|---|---|---|---|---|
| Maddrey et al. (derivation) | 1978 | 55 | Death during 28–32 day trial + 5 days after | 6/31 placebo died vs 1/24 prednisolone; PT and bilirubin independently predicted death |
| Ramond et al. | 1992 | 61 | Death by day 66 | 16/29 (55%) placebo died vs 4/32 (12.5%) prednisolone (p=0.001) |
| STOPAH (Thursz et al.) | 2015 | 1,103 | Death by day 28 | 17% placebo-placebo vs 14% prednisolone-placebo; odds ratio 0.72 (95% CI 0.52–1.01, p=0.06) — did not reach significance |
Scroll the table sideways for every column.
The direction of effect for prednisolone has been consistent across four decades of trials, but its statistical strength has weakened as trial size has grown — the largest and most recent trial found a 28-day benefit that fell just short of the conventional significance threshold, and no benefit at 90 days or one year.
Evidence
Derivation — Maddrey et al.
1978 · n = 55A 28- to 32-day randomized, double-blind trial of prednisolone 40 mg/day versus placebo in 55 patients with alcoholic hepatitis. Stepwise discriminant analysis of the laboratory values measured identified prothrombin time prolongation and serum bilirubin as independently associated with death.
4 of 31 placebo-treated patients died during the trial and 2 more within 5 days of completion, versus 1 of 24 prednisolone-treated patients (Fisher exact test, p=0.10 for the raw comparison; corticosteroid therapy significantly reduced mortality, p<0.05, once treatment was included as a variable in the discriminant model).
The 32 threshold — Carithers et al.
1989 · n = 66Randomized, double-blind, multicenter trial of methylprednisolone in 66 patients with alcoholic hepatitis, enrolled using either spontaneous hepatic encephalopathy or a discriminant function above 32 as the severity criterion — the trial that established 32 as the field's steroid-decision threshold.
Methylprednisolone decreased short-term mortality in patients meeting either severity criterion (encephalopathy or DF > 32); this is the study from which the ≥ 32 cutoff used ever since is drawn.
Confirmatory trial — Ramond et al.
1992 · n = 61Randomized, double-blind trial of prednisolone 40 mg/day versus placebo for 28 days in 61 patients with biopsy-proven alcoholic hepatitis, entered with either spontaneous encephalopathy (n=19) or a discriminant function above 32; 57 of 61 had cirrhosis on biopsy.
By day 66, 16 of 29 placebo recipients had died (survival 45±8%) versus 4 of 32 prednisolone recipients (survival 88±5%); log-rank p=0.001. The survival advantage persisted after adjustment for prognostic factors and stratification by centre.
STOPAH — the largest trial to date
2015 · n = 1,103Multicentre, double-blind, 2×2 factorial randomized trial of prednisolone and pentoxifylline in 1,103 UK patients with a clinical diagnosis of severe alcoholic hepatitis; 1,053 patients contributed to the primary 28-day mortality analysis.
28-day mortality was 17% (45/269) in the double-placebo group versus 14% (38/266) with prednisolone alone; odds ratio 0.72 (95% CI 0.52–1.01, p=0.06). No significant difference at 90 days or 1 year. Serious infections occurred in 13% of prednisolone-treated patients versus 7% of those who did not receive it (p=0.002). Pentoxifylline showed no survival benefit.
How it compares
Maddrey's DF vs Lille score
Not competitors — Maddrey's DF decides whether to start steroids, and the Lille score, calculated after a week of treatment, decides whether to continue them.
Maddrey's DF uses baseline labs to flag severe disease. The Lille score incorporates the change in bilirubin over the first week of steroid treatment and was built specifically to identify early non-responders, for whom continuing steroids adds infection risk without a survival benefit. The two scores are sequential steps in the same treatment decision, not alternatives to each other.
Maddrey's DF vs Glasgow Alcoholic Hepatitis Score
GAHS was derived to correct Maddrey's DF omission of renal function and white cell count, and the two are often used together rather than as substitutes.
Maddrey's DF is built from PT and bilirubin alone. GAHS adds age, white cell count and renal function (as blood urea) alongside PT ratio and bilirubin, on the reasoning that renal impairment and a marked inflammatory response are prognostically important and were missing from the older score. In practice, many units check both: Maddrey's DF for the trial-standard steroid threshold, and GAHS as a second opinion that captures markers the DF does not.
Pearls & pitfalls
- Patient PT and control PT must come from the same assay run. A remembered 'normal' control value instead of the actual run's control is the most common source of a wrong discriminant function.
- The formula uses PT in seconds, not INR. Some online calculators substitute INR into a modified formula — that produces a different number that was never validated against the 32 cutoff used here.
- The score has no upper bound. Do not treat a very high number as an entry error the way you would for a capped score like MELD.
- A DF of 32 is not a mortality-benefit guarantee for steroids — the largest trial to date (STOPAH) found a 28-day benefit that did not reach statistical significance.
- Reassess with the Lille score at day 7 of steroid treatment. Lille was built specifically to identify non-responders early, which the discriminant function itself cannot do.
- Bilirubin units matter. Enter mg/dL; convert from µmol/L by dividing by 17.1, or use the unit toggle.
Critical actions
- Exclude active infection and gastrointestinal bleeding before starting corticosteroids — both are common in this population and both are worsened by steroids.
- At DF ≥ 32, discuss corticosteroids, but set expectations using STOPAH's result rather than the older, smaller trials alone.
- Reassess at day 7 with the Lille score; a non-response there is the trigger to stop steroids rather than continuing on the strength of the baseline discriminant function.
- Address nutrition and thiamine, and manage alcohol withdrawal actively regardless of which side of 32 the score falls.
- Recheck the score if the clinical picture changes — it is not a one-off measurement, and a patient can cross the threshold during admission.
Why this score exists
Maddrey's original 1978 paper was not built to produce a general severity score — it was the byproduct of asking which laboratory values predicted death in a small steroid trial, and prothrombin time and bilirubin were what came out of that analysis. The 32 threshold that everyone now associates with the score was not in that paper at all; it was fixed over a decade later, in 1989, as the enrollment criterion for a different trial, and it has persisted mainly because every subsequent alcoholic hepatitis trial adopted it for comparability rather than because 32 has any special biological meaning.
About the creator
First author, 1978 derivation study
Published the discriminant function in the controlled trial of prednisolone from which the threshold defining severe disease is taken.
Limitations
- Derived from a very small trial (n=55) by modern standards, and the 32 threshold was fixed by a separate, later trial rather than by the derivation study itself.
- Uses PT in seconds against the local laboratory's control, which is less standardised across centres and instruments than INR — the same patient can generate different discriminant function values in different laboratories.
- Does not include renal function or a marker of the systemic inflammatory response, both of which later scores (GAHS, MELD, Lille) were built partly to add.
- The largest and most recent randomized trial found the 28-day mortality benefit of prednisolone at this threshold did not reach statistical significance, and found no benefit at 90 days or one year — a meaningfully different conclusion from the trials that established the score's clinical use.
- Says nothing about diagnosis; it assumes alcoholic hepatitis has already been established rather than helping to establish it.
If you are the patient
Maddrey's discriminant function is a number your medical team calculates from two blood tests — your bilirubin and how much longer your blood takes to clot compared with a normal sample — to judge how severe your alcohol-related liver inflammation is. A result of 32 or higher is 'severe' by this score, and it is the point at which steroid treatment is usually discussed, because that is the group in whom steroid trials have shown a benefit. It is important to know that the largest and most recent study found this benefit was smaller than earlier, smaller studies suggested, and steroids also raise the risk of serious infection — so this is a decision made with you, weighing a modest possible benefit against a real risk, not an automatic prescription once the number crosses 32. If steroids are started, your team will usually recheck things after about a week (the Lille score) to see whether they are working before continuing.
Frequently asked questions
What is Maddrey's discriminant function?#
A score for alcoholic hepatitis calculated as 4.6 × (patient PT − control PT, in seconds) + total bilirubin (mg/dL). A value of 32 or above defines severe disease and is the conventional threshold for considering corticosteroids.
What Maddrey's DF score needs steroids?#
32 or above is the conventional threshold, established by the 1989 Carithers trial and used as the entry criterion for every major alcoholic hepatitis steroid trial since, including STOPAH in 2015.
Do steroids actually reduce mortality in severe alcoholic hepatitis?#
The evidence is mixed by trial size. Smaller trials in 1978 and 1992 found a significant mortality benefit. The largest trial to date, STOPAH (n=1,103, 2015), found a 28-day mortality odds ratio of 0.72 for prednisolone that did not reach statistical significance (p=0.06), and no benefit at 90 days or one year, alongside a higher rate of serious infections.
What labs do you need for Maddrey's discriminant function?#
Total bilirubin and the prothrombin time, both the patient's value and the laboratory's control value from the same assay run. No other inputs are used.
Is Maddrey's DF the same as the Lille score?#
No. Maddrey's DF is calculated at baseline to decide whether to start steroids. The Lille score is calculated after roughly a week of steroid treatment, using the change in bilirubin, to decide whether to continue them.
Can I use INR instead of PT seconds in the formula?#
Not in the original, validated formula. INR-based 'modified' discriminant functions exist but produce different numbers, and the 32 threshold has only been validated against the PT-in-seconds version described here.
References
Original / primary reference
- Maddrey WC, Boitnott JK, Bedine MS, Weber FL Jr, Mezey E, White RI Jr. Corticosteroid therapy of alcoholic hepatitis. Gastroenterology. 1978;75(2):193-199.
- Carithers RL Jr, Herlong HF, Diehl AM, et al. Methylprednisolone therapy in patients with severe alcoholic hepatitis: a randomized multicenter trial. Ann Intern Med. 1989;110(9):685-690.