How this instrument works
The Lille model comes from a 2007 study by Louvet, Naveau, Mathurin, and colleagues in Hepatology, built specifically for patients already started on corticosteroids for severe alcoholic hepatitis. The question it answers isn't whether the liver is failing — that's usually already established — but whether this particular patient, on this particular treatment, is responding well enough by day 7 to justify continuing it.
The model's central insight is about trajectory rather than a snapshot. Two patients can arrive with very different starting bilirubin levels; what predicts outcome is less where each one started than how far bilirubin has fallen after a week of treatment. A patient responding to steroids shows a clear early drop in bilirubin; one who is not responding shows little to no drop, and the six weighted terms — age, albumin, the day-0-to-day-7 bilirubin change, renal insufficiency, day-0 bilirubin, and prothrombin time — are combined through a logistic regression specifically to capture that early signal.
The output is a probability-like value between 0 and 1, and the derivation study's validated cutoff sits at 0.45. Below that line, six-month survival in the original cohort was around 85%; at or above it, survival dropped sharply, and the finding that matters clinically is what to do with a poor-response result: continuing steroids past day 7 in a patient who has not responded does not reliably rescue outcomes, but it does keep the patient exposed to steroid-related infection risk during a period they are unlikely to benefit from. A score at or above the cutoff is the standard trigger to reconsider that exposure.
The model is narrow by design. It was derived and validated in patients with severe alcoholic hepatitis already committed to a course of corticosteroids, not in liver failure from other causes and not as a tool for deciding who should start steroids in the first place — that earlier decision typically rests on Maddrey's discriminant function or a similar severity score, calculated before day 7 even arrives.
- Enter Age in years and Albumin at admission in grams per litre.
- Enter Bilirubin at day 0 and Bilirubin at day 7, both in µmol/L.
- Set Renal insufficiency (creatinine above 115 µmol/L) to Yes or No.
- Enter Prothrombin time in seconds, then read the Lille score (0–1); the working block shows the linear combination behind it.
Worked example — bilirubin falling from 300 to 200
A 50-year-old with albumin 30 g/L, bilirubin falling from 300 to 200 µmol/L over the week, no renal insufficiency, and a prothrombin time of 18 seconds. Age term: −0.101 × 50 = −5.05. Albumin term: 0.147 × 30 = 4.41. Bilirubin-change term: 0.0165 × (300 − 200) = 1.65. Renal term: 0. Day-0 bilirubin term: −0.0065 × 300 = −1.95. PT term: −0.0096 × 18 = −0.1728. Added to the 3.19 constant: 3.19 − 5.05 + 4.41 + 1.65 − 0 − 1.95 − 0.1728 = 2.077. Lille = e⁻²·⁰⁷⁷ ⁄ (1 + e⁻²·⁰⁷⁷) ≈ 0.111 — well under the 0.45 cutoff, a good response.
Change only the trajectory and the picture flips. A 65-year-old with lower albumin (22), bilirubin barely moving from 400 to 390, renal insufficiency, and a longer prothrombin time (24 seconds) works out to a linear combination of about −3.012, giving a Lille score near 0.953 — well above 0.45, a poor response that typically prompts stopping steroids. A 40-year-old whose bilirubin falls hard, from 250 to 100, with albumin 35 and a prothrombin time of 15 seconds, instead gives a linear combination near 5.001 and a Lille score near 0.0067, an excellent response. All three patients are plugged into the identical six-term equation — the bilirubin trajectory is what moves the number.
Questions
Does the absolute bilirubin level matter, or only how much it fell?
Both appear in the equation, but they play different roles. The day-0-to-day-7 change carries the largest bilirubin-related weight and is the term most directly tied to treatment response; day-0 bilirubin alone carries a smaller separate weight reflecting how sick the patient was to begin with. Two patients with an identical day-7 bilirubin can still score very differently depending on where they started and how much ground they covered in a week — the trajectory is the point of the model, not a snapshot.
What happens clinically if the Lille score is 0.45 or higher?
A score at or above 0.45 is the validated marker of poor response and is the standard trigger to reconsider continuing corticosteroids. The rationale is that patients who have not shown a meaningful bilirubin drop by day 7 are unlikely to benefit from further steroid exposure, while they remain at elevated risk of steroid-related infection during that same window — so continuing treatment in a clear non-responder tends to add risk without a matching chance of benefit.
Why does renal insufficiency push the score toward a poor response?
Renal insufficiency (creatinine above 115 µmol/L) subtracts 0.206 from the linear combination, and because a lower linear combination produces a higher Lille score, renal impairment independently pushes the result toward poor response. In the derivation cohort it marked more severe, multi-organ involvement associated with worse outcomes regardless of how bilirubin was trending, so the model weights it separately from the liver-specific terms.
How does the Lille model relate to Maddrey's discriminant function?
They answer different questions at different points in care. Maddrey's discriminant function is typically calculated at admission to decide whether a patient's alcoholic hepatitis is severe enough to justify starting corticosteroids in the first place. The Lille model is calculated a week later, only in patients already on steroids, to decide whether that treatment is working well enough to continue. One is a starting gate, the other a day-7 checkpoint.
Can the Lille model be used for liver disease that isn't alcohol-related?
No — it was derived and validated specifically in patients with severe alcoholic hepatitis who had already started corticosteroids, and its coefficients reflect that population. Applying it to viral hepatitis, drug-induced liver injury, or other causes of acute liver failure isn't supported by the original data and isn't how the tool was intended to be used.
Is a score of exactly 0.45 a hard line?
Treat it as a validated threshold rather than a precise biological boundary. The cutoff was chosen because it best separated survivors from non-survivors in the derivation and validation cohorts, but a score of 0.40 and a score of 0.50 both sit close enough to that line that the surrounding clinical picture — overall trend, infection status, organ function — should weigh into the decision alongside the number itself.
References
- Louvet et al. 2007, Hepatology — the Lille model derivation study (PubMed)
- EASL Clinical Practice Guidelines — alcohol-related liver disease, 2018
Read this first: This instrument computes a screening figure from population formulas — it is not a diagnosis, and it cannot see the whole picture a clinician can. Use it to inform a conversation, not to replace one.