About the Lille Model for Corticosteroid Response in Alcohol-Related Hepatitis
Unlike every other alcohol-related hepatitis score, the Lille model is calculated after seven days of corticosteroid treatment rather than at admission — it exists to answer a question no baseline score can, which is whether the steroids are working. It combines five admission variables (age, albumin, prothrombin time, renal insufficiency and day-0 bilirubin) with the one dynamic term that carries the model: how far bilirubin has fallen by day 7. A score of 0.45 or above defines a non-responder, which described about 40% of treated patients in the derivation work and predicts roughly 25% six-month survival, against about 85% for those scoring below 0.45. Because prolonging ineffective immunosuppression in a jaundiced, infection-prone patient does harm, identifying non-response early is the point.
Formula
R = 3.19 − 0.101 × age + 0.147 × albumin(g/L) + 0.0165 × (bilirubin day 0 − bilirubin day 7, µmol/L) − 0.206 × renal insufficiency − 0.0065 × bilirubin day 0 (µmol/L) − 0.0096 × prothrombin time (s); Lille score = e^(−R) / (1 + e^(−R))- age
- In years.
- albumin
- Day-0 albumin in g/L (multiply g/dL by 10).
- Δ bilirubin
- The FALL in bilirubin from day 0 to day 7 in µmol/L, so a decrease is a positive number. A rising bilirubin makes this term negative and drives the score up.
- renal insufficiency
- 1 if creatinine exceeds 1.3 mg/dL (about 115 µmol/L), otherwise 0.
- bilirubin day 0
- Baseline bilirubin in µmol/L, entering with a negative coefficient.
- prothrombin time
- In seconds, not INR.
- The published model uses SI units — albumin in g/L and bilirubin in µmol/L. Feeding mg/dL and g/dL values into these coefficients gives a meaningless score.
- The bilirubin term is the FALL from day 0 to day 7. Getting the sign backwards inverts the model's central variable and will misclassify responders as non-responders.
- Prothrombin time is in seconds. Substituting INR is a common and consequential error, since the two are on entirely different scales.
- Renal insufficiency is dichotomised at a creatinine of 1.3 mg/dL rather than entered continuously, so a creatinine of 1.4 and one of 4.0 contribute identically.
- The final logistic transformation constrains the output to between 0 and 1, which is why the score reads as a probability rather than a points total.
Interpreting the result
A score below 0.45 marks a responder, with roughly 85% six-month survival, and continuing the corticosteroid course is generally appropriate. A score of 0.45 or above defines non-response — around 40% of treated patients in the derivation cohort — and predicts approximately 25% six-month survival. In that group the reasoning shifts: continuing steroids is not merely futile but actively harmful given the infection risk they carry in a patient who is already immunocompromised by liver failure, so the accepted course is to stop them. What follows depends on the individual — early liver transplantation has become an option for carefully selected non-responders, and where it is not appropriate the emphasis moves to supportive care, infection control, nutrition, abstinence support and honest prognostic discussion. Combining Lille with a static baseline score such as ABIC or MELD predicts survival better than either approach used alone.
| Score | Band | What it means | Action |
|---|---|---|---|
| < 0.45 | Steroid responder | Approximately 85% six-month survival | Continue the corticosteroid course; maintain infection surveillance, nutrition and abstinence support |
| ≥ 0.45 | Steroid non-responder | Approximately 25% six-month survival; about 40% of treated patients fall here | Stop corticosteroids; consider early transplant assessment in selected patients, and otherwise focus on supportive care and prognostic discussion |
What the Lille Model needs (6 inputs)
- Age (years)
- Enters with a negative coefficient in the linear predictor R.
- Albumin, day 0 (g/dL)
- Baseline albumin, converted to g/L for the published formula.
- Total bilirubin, day 0 (mg/dL)
- The baseline value, used both on its own and to calculate the fall by day 7.
- Total bilirubin, day 7 (mg/dL)
- The value after a week of corticosteroids. The difference from day 0 is the dynamic term that gives the model its power.
- Prothrombin time (seconds)
- Baseline prothrombin time in seconds, not INR — the model was published in seconds.
- Creatinine (mg/dL)
- Used as a binary term: renal insufficiency is defined as a creatinine above 1.3 mg/dL, which contributes a fixed amount.
Units. The Lille model is an SI-unit formula: albumin in g/L, bilirubin in µmol/L, prothrombin time in seconds. Convert with albumin g/dL × 10 = g/L and bilirubin mg/dL × 17.1 = µmol/L. Renal insufficiency is defined at creatinine > 1.3 mg/dL (≈ 115 µmol/L). This calculator accepts conventional units and converts internally.
What it returns
- Lille score
- A probability between 0 and 1. Higher values indicate a poorer response to corticosteroids.
- Response category
- Responder (below 0.45) or non-responder (0.45 or above).
How it is calculated
The model was built on 320 patients with severe alcohol-related hepatitis — defined by a discriminant function of 32 or above — prospectively treated with corticosteroids, with a further 118 used for validation. Baseline data and the change in bilirubin at day 7 were tested by logistic regression against death at six months. Six reproducible variables survived: age, renal insufficiency, albumin, prothrombin time, day-0 bilirubin, and the evolution of bilirubin at day 7. The structure is worth noticing — five of the six are baseline measurements, and only one describes what has happened during treatment, yet it is that single dynamic term that turns a prognostic model into a treatment-response model. The linear predictor is then passed through a logistic function so the result is expressed as a probability of poor outcome.
Facts & figures
| Lille score | Category | Six-month survival | Proportion of patients |
|---|---|---|---|
| < 0.45 | Responder | ≈ 85% | ≈ 60% |
| ≥ 0.45 | Non-responder | ≈ 25% | ≈ 40% |
Derived in 320 corticosteroid-treated patients with a discriminant function ≥ 32, validated in a further 118.
| Static scores (Maddrey, MELD, ABIC, GAHS) | Lille model | |
|---|---|---|
| Timing | Admission | Day 7 of corticosteroids |
| Question answered | How severe is this, and who should be treated? | Is the treatment working? |
| Requires a repeat bilirubin | No | Yes — day 0 and day 7 |
| Applies to untreated patients | Yes | No |
Joint-effect models pairing a static score with Lille outperformed either alone for two- and six-month survival (derivation n = 538, validation n = 604).
Evidence
Derivation — Louvet
2007 · n = 320320 patients with severe alcohol-related hepatitis (discriminant function ≥ 32) prospectively treated with corticosteroids, forming the development cohort, with 118 further patients used for validation. Baseline variables and the change in bilirubin at day 7 were entered into a logistic regression predicting death at six months.
A model combining six reproducible variables — age, renal insufficiency, albumin, prothrombin time, bilirubin and the evolution of bilirubin at day 7 — was highly predictive of death at six months. A score above 0.45 defined non-response, describing about 40% of treated patients, with roughly 25% six-month survival compared with about 85% below the threshold.
Joint static-dynamic modelling — Louvet
2015 · n = 604Patients with severe alcohol-related hepatitis treated with corticosteroids from French and UK databases (derivation n = 538), validated in corticosteroid trial cohorts from the United States, France, Korea and Belgium (n = 604). Three joint-effect models were compared: Maddrey+Lille, MELD+Lille and ABIC+Lille.
Every joint model predicted two- and six-month survival significantly better than static or dynamic scores used alone, in both derivation and validation cohorts — establishing that baseline severity and day-7 response carry complementary rather than overlapping information.
STOPAH trial context — Thursz
2015A multicentre randomised trial of prednisolone and pentoxifylline in severe alcohol-related hepatitis, which framed the modern evidence on how much benefit corticosteroids actually deliver and therefore on the stakes of continuing them in a non-responder.
Prednisolone's mortality benefit at 28 days did not reach statistical significance, and corticosteroid treatment was associated with more serious infections — sharpening the argument for stopping steroids early in patients the Lille model identifies as not responding.
How it compares
Lille Model vs Maddrey's Discriminant Function
Maddrey's DF decides who starts corticosteroids at admission; the Lille model decides who should stop them at day 7 — they are sequential steps in one pathway, not competing scores.
A discriminant function of 32 or above is the conventional entry criterion for corticosteroid treatment, and it was the definition of 'severe' used to assemble the Lille derivation cohort. Lille then evaluates what happened over the following week using the fall in bilirubin. Neither substitutes for the other: without Maddrey's DF there is no indication to treat, and without Lille there is no defined point at which treatment is judged to have failed.
Lille Model vs ABIC score
ABIC grades baseline risk into three tiers at admission while Lille measures response at day 7, and using them together predicts survival better than either in isolation.
ABIC is a static admission score built from age, bilirubin, INR and creatinine, producing low, intermediate and high-risk groups. Lille is dynamic and requires a week of treatment before it can be computed. Louvet and colleagues formally tested the pairing — ABIC+Lille alongside Maddrey+Lille and MELD+Lille — across 538 derivation and 604 validation patients, and every joint model beat static or dynamic scoring alone for two- and six-month survival.
Lille Model vs Glasgow Alcoholic Hepatitis Score
GAHS predicts at admission who is likely to benefit from corticosteroids; Lille confirms at day 7 whether that prediction held for the individual patient in front of you.
GAHS separates patients above and below a threshold of 9, with steroid survival benefit confined to the higher group — a predictive claim made before treatment starts. Lille makes no prediction at all; it measures. The two are complementary in an obvious way: GAHS informs the decision to treat, Lille audits it a week later, and a GAHS-positive patient who turns out to be a Lille non-responder is exactly the patient in whom stopping steroids early matters most.
Pearls & pitfalls
- The bilirubin term is the fall from day 0 to day 7, so a decrease enters as a positive number. Reversing the sign inverts the model's most important variable.
- Prothrombin time is in seconds, not INR. This is the most frequent unit error in Lille calculation and it is not a small one.
- The formula expects SI units — albumin in g/L, bilirubin in µmol/L — which is the opposite convention from ABIC and Maddrey's DF, both of which use mg/dL.
- The day-0 bilirubin must be captured when steroids start. There is no way to compute a Lille score retrospectively without it, and it is easily lost if treatment begins before the score is anticipated.
- Renal insufficiency is binary at a creatinine of 1.3 mg/dL, so it does not scale with the severity of kidney injury.
- A non-responder score is a reason to stop steroids, not merely to note poor prognosis — continued immunosuppression in that setting adds infection risk without survival benefit.
- Lille says nothing about untreated patients and is not a severity score; a patient never started on steroids has no meaningful Lille result.
Critical actions
- Record the day-0 bilirubin, albumin, prothrombin time and creatinine at the moment corticosteroids are started, so the day-7 score can actually be calculated.
- Calculate the score on day 7 and act on it — stop corticosteroids in non-responders rather than completing a fixed course out of habit.
- Screen actively for infection before and during steroid treatment, and again when a non-responder result prompts a change of plan.
- Assess early transplant candidacy in selected non-responders, ideally through a service with an established pathway, rather than treating non-response as the end of options.
- Pair the Lille result with a baseline score such as ABIC or MELD, since the combination predicts survival better than either alone.
- Address abstinence, nutrition and thiamine in every patient regardless of response category — none of these appear in the model, and all affect outcome.
Why this score exists
The authors' framing was that early identification of non-responders was 'crucial', and the word choice reflects a specific clinical frustration: corticosteroids were the only therapy with any evidence base in severe alcohol-related hepatitis, which created pressure to persist with them even when a patient was visibly not improving. Their contribution was to make 'not improving' measurable at a point early enough to act on, and to define it as a probability rather than a clinical impression. The explicit secondary purpose was to enable trials of alternative therapies — a salvage-therapy trial is impossible without an agreed, reproducible definition of who has failed first-line treatment. That the term 'non-responder' now has a numeric definition at all is the model's real legacy.
About the creator
First author, 2007 derivation study
Derived the Lille model to define corticosteroid non-response at day 7, and later led the work combining it with static scores such as ABIC and MELD.
Senior author
Led the Lille group's programme on severe alcohol-related hepatitis, including early liver transplantation for selected non-responders.
Limitations
- It requires seven days of treatment before it yields anything, so it cannot inform the initial decision to give corticosteroids.
- It depends on a prospectively recorded day-0 bilirubin, which makes it impossible to apply retrospectively if baseline values were not captured.
- Renal insufficiency is dichotomised at a single creatinine threshold, discarding information about the severity of kidney injury that heavily influences outcome.
- The 0.45 threshold was derived in a specific corticosteroid-treated European cohort, and patients close to the cut-off should not be managed as though the boundary were sharp.
- It was developed before early liver transplantation for alcohol-related hepatitis became an option, so it identifies non-responders without addressing what should follow.
- It contains no measure of infection, which is both a major cause of death in this population and the principal harm of the treatment being assessed.
- A shorter assessment interval has been proposed by some groups, and the optimal day on which to judge response is not fully settled.
If you are the patient
The Lille score is used about a week after starting steroid treatment for severe alcohol-related hepatitis, to check whether the treatment is helping. It is worked out from your age, some blood tests taken when treatment started (a protein called albumin, a clotting test, and kidney function), and — most importantly — how much your bilirubin, the blood test that reflects jaundice, has fallen over those seven days. The result is a number between 0 and 1. Below 0.45 suggests the steroids are working and they are usually continued. At 0.45 or above, the steroids are unlikely to be helping, and because they also increase the risk of serious infection, doctors will normally stop them rather than continue. That is not the end of treatment: your team will focus on other support, and for some carefully selected people a liver transplant can be considered. Stopping alcohol completely remains the single most important factor in long-term survival.
Frequently asked questions
When is the Lille score calculated?#
On day 7 of corticosteroid treatment for severe alcohol-related hepatitis. It needs both the day-0 and the day-7 bilirubin, so the baseline value must be recorded when steroids are started.
What Lille score means steroids are not working?#
0.45 or above defines a non-responder. That group had roughly 25% six-month survival in the derivation study, compared with about 85% for those scoring below 0.45, and represented around 40% of treated patients.
What should happen if a patient is a Lille non-responder?#
Corticosteroids should generally be stopped, because continuing them adds infection risk without survival benefit. Early liver transplantation can be considered in carefully selected patients; otherwise management focuses on supportive care, infection control, nutrition, abstinence support and honest prognostic discussion.
Does the Lille model use INR or prothrombin time?#
Prothrombin time in seconds. Substituting INR is a common error and produces a badly wrong score, because the two measures are on completely different scales.
Which units does the Lille formula need?#
SI units — albumin in g/L and bilirubin in µmol/L — which is the opposite convention from ABIC and Maddrey's discriminant function. Multiply albumin g/dL by 10, and bilirubin mg/dL by 17.1.
Can the Lille score be used in patients not on steroids?#
No. It measures response to corticosteroid treatment specifically, and has no meaning as a standalone severity or prognostic score in an untreated patient.
Should Lille be used alongside a baseline score?#
Yes. Joint models pairing Lille with Maddrey's DF, MELD or ABIC predicted two- and six-month survival significantly better than either static or dynamic scoring alone, across 538 derivation and 604 validation patients.
References
Original / primary reference
Validation and combined modelling
- Louvet A, Labreuche J, Artru F, et al. Combining data from liver disease scoring systems better predicts outcomes of patients with alcoholic hepatitis. Gastroenterology. 2015;149(2):398-406.
- Thursz MR, Richardson P, Allison M, et al. Prednisolone or pentoxifylline for alcoholic hepatitis. N Engl J Med. 2015;372(17):1619-1628.