The Lille model answers a specific, high-stakes question: after a week of corticosteroids for severe alcoholic hepatitis, is this treatment working? Published by Louvet and colleagues in 2007, it converts six day-0 and day-7 measurements into a single score that separates likely responders from likely non-responders — informing whether to continue steroids or pursue an alternative strategy.
How the Lille model works
The Lille model is a logistic-regression score built from a derivation cohort of patients with severe alcoholic hepatitis already receiving corticosteroids. Six variables — age, albumin, bilirubin at day 0, bilirubin at day 7, renal insufficiency, and prothrombin time — combine into a linear predictor R, which then passes through the logistic transform exp(−R) / (1 + exp(−R)) to produce a score between 0 and 1.
The published coefficients were fit using SI units: albumin in g/L, bilirubin in µmol/L, and prothrombin time in raw seconds (not INR). This calculator accepts the US-conventional g/dL and mg/dL units clinicians here typically chart in, and converts them internally (×10 for albumin, ×17.1 for bilirubin) before applying the coefficients — so entering familiar units still produces the formula's intended result.
Why the day-7 bilirubin trajectory dominates
Unlike admission-severity scores such as Maddrey's discriminant function, the Lille model is deliberately measured after a week of treatment. The single most influential term is the 'bilirubin evolution' — the drop from day 0 to day 7 — because it is a direct biological readout of whether the liver is responding to steroids, independent of how sick the patient looked at baseline. A patient with very high day-0 bilirubin who shows a large drop by day 7 can still score as a responder; a patient with more modest day-0 bilirubin whose level barely moves can score as a non-responder.
Renal insufficiency (creatinine above 1.3 mg/dL, or dialysis) and an elevated prothrombin time both push the score toward non-response, reflecting worsening synthetic liver function and early multi-organ involvement.
Inputs and what they mean
All six inputs are required. Age and albumin (day 0) are single measurements. Bilirubin is entered twice — day 0 and day 7 — since it's the change, not either value alone, that drives the score. Creatinine auto-derives the renal insufficiency flag at the 1.3 mg/dL threshold; the dialysis checkbox overrides it, since a dialysis patient's creatinine no longer reflects native kidney function. Prothrombin time must be entered in seconds — the model was never validated against INR, and there is no dependable conversion between the two, so INR is intentionally not an accepted input here.
Limits and when to look further
The Lille model was derived and validated specifically in patients with severe alcoholic hepatitis already started on corticosteroids — it is not a general liver-severity score and should not be used to decide whether to start steroids in the first place (that's Maddrey's discriminant function). The 0.45 cutoff is a probability threshold from a specific cohort; some centers use it to guide steroid discontinuation, but that remains an individualized hepatology decision that weighs the whole clinical trajectory, not a single automated number. This calculator does not replace a full hepatology assessment, and it cannot account for factors the six-variable model doesn't capture, such as infection, gastrointestinal bleeding, or acute-on-chronic liver failure staging.