Maddrey's discriminant function is a simple bedside formula, built from two routinely-drawn labs, that flags the alcoholic hepatitis patients most likely to benefit from corticosteroid therapy. This guide covers how the formula is derived, why it weights prothrombin time so heavily, what DF≥32 does and doesn't mean, and where a second tool — the Lille model — fits into the treatment decision.
How the discriminant function works
Introduced by Maddrey and colleagues in a 1978 trial of corticosteroid therapy, the discriminant function combines prothrombin time (PT) and total bilirubin into a single severity number: DF = 4.6 × (patient PT − control PT) + bilirubin. Rather than using the patient's raw PT, the formula measures how much the PT is prolonged relative to the lab's own control value — this matters because normal PT reference ranges vary between labs, reagents, and instruments, so an absolute PT cutoff would not travel well between institutions.
The 4.6 multiplier weights PT prolongation heavily relative to bilirubin, reflecting how strongly synthetic liver failure (impaired clotting-factor production) tracked with mortality in the original cohort.
What DF≥32 means — and what it doesn't
A discriminant function of 32 or higher is the historical cutoff used to identify alcoholic hepatitis severe enough that corticosteroid therapy is considered. Historical cohorts have reported roughly 30–50% 30-day mortality without treatment at this threshold, though exact figures vary by era and population and should not be read as a prognosis for any individual patient.
Crucially, DF≥32 is a screening flag, not a treatment order. Steroids are only appropriate after ruling out contraindications — active gastrointestinal bleeding, uncontrolled infection or sepsis, renal failure, hepatorenal syndrome, and pancreatitis are the most common reasons a DF≥32 patient will not receive steroids.
The Lille model and other complementary scores
Maddrey's DF answers the question 'is this patient severe enough to consider steroids?' — it says nothing about whether steroids are actually working. The Lille model is calculated after roughly seven days of corticosteroid therapy and scores the patient's early biochemical response; a poor Lille score is used to stop steroids in non-responders rather than continue a treatment unlikely to help. The MELD score and the Glasgow Alcoholic Hepatitis Score are separate, more recently validated severity indices that some centers use alongside or instead of Maddrey's DF.
Limitations
The formula is nearly 50 years old and was derived and validated before modern liver-disease scoring systems existed, so many centers now cross-check it against MELD. The control PT is lab-specific and not standardized the way the INR was later designed to be, which is one reason some clinicians prefer INR-based scores for cross-institution comparison. This calculator reports the discriminant function and the historical threshold only — it does not screen for steroid contraindications, calculate a Lille score, or substitute for a hepatologist's evaluation.