Lille Model for Alcoholic Hepatitis

Learn how the Lille model estimates early corticosteroid non-response in severe alcoholic hepatitis using age, albumin, bilirubin at day 0 and 7, creatinine, and prothrombin time—with the 0.45 cutoff and clinical caveats.

Lille Model (Severe Alcoholic Hepatitis)

The Lille model estimates the probability of not responding to a week of corticosteroids using age, day-0 albumin and bilirubin, day-7 bilirubin, day-0 creatinine (renal insufficiency flag), and day-0 prothrombin time. Use values from the same steroid course (typically day 7 after starting prednisolone/prednisone). For severity at presentation, clinicians often use modified Maddrey discriminant function (mDF) and MELD (including MELD-Na) alongside histology and clinical context.

Age (years)

Serum albumin (day 0)

Albumin unit

Total bilirubin day 0

Total bilirubin day 7

Bilirubin unit

Evolution term uses (day 0 − day 7) bilirubin in µmol/L after internal conversion. A falling bilirubin increases R and lowers the Lille score.

Serum creatinine (day 0)

Renal insufficiency in the model: creatinine >1.3 mg/dL (≈115 µmol/L).

Creatinine unit

Prothrombin time (day 0, seconds)

Use the same PT assay as in the derivation cohort (seconds), not INR, unless your laboratory defines an equivalent.

Disclaimer: The Lille model was derived in patients with severe alcoholic hepatitis treated with corticosteroids. It supports—but does not replace—clinical judgment, infection screening, and multidisciplinary decisions (including transplant referral pathways). Cutoffs describe group-level outcomes from published cohorts.

Lille model — equation and units

Overview

The Lille model is a logistic regression–based index developed to identify patients with severe alcoholic hepatitis who are unlikely to benefit from continued corticosteroids after an initial treatment period. It combines baseline characteristics with the change in total bilirubin at day 7 of steroid therapy. Primary reference: Louvet A et al., Hepatology 2007;45(6):1348–1354. DOI 10.1002/hep.21607.

Logistic form

The published probability index is computed from a linear predictor R (some sources write the equivalent using exp(−R) in the numerator). This implementation uses the stable form:

Lille score = 1 / (1 + exp(R))

Algebraically the same as exp(−R) / (1 + exp(−R)). Values range from 0 to 1; higher scores indicate lower likelihood of treatment response under the original steroid strategy.

Linear predictor R (Louvet 2007)

All bilirubin terms use total bilirubin in µmol/L after conversion.

R = 3.19 − 0.101 × age (years) + 0.147 × albumin at day 0 (g/L) + 0.0165 × (bilirubin day 0 − bilirubin day 7) [µmol/L] − 0.206 × renal insufficiency (1 if present, else 0) − 0.0065 × bilirubin day 0 (µmol/L) − 0.0096 × prothrombin time at day 0 (seconds)

Bilirubin evolution is (day 0 − day 7). When bilirubin falls during therapy, this term is positive, R increases, and the Lille score tends to decrease.
Renal insufficiency in the original model is a binary factor: serum creatinine > 1.3 mg/dL (often quoted as >115 µmol/L). Enter creatinine in your usual units; the calculator converts µmol/L to mg/dL using the conventional 88.4 factor for the threshold.
Prothrombin time is the day-0 value in seconds, as in the derivation cohort—not INR.

Interpretation (published cutoff)

Lille score	Common use
< 0.45	Associated with better 6-month outcomes when steroids were continued in the validation cohort; typically interpreted as responder trajectory.
≥ 0.45	Associated with poor outcomes despite steroids; widely used to define non-response and prompt re-evaluation of therapy (including guidelines on early transplant assessment where appropriate).

Cutoffs describe group-level risks from specific study populations. Always integrate infection screening, bleeding risk, psychosocial factors, and local pathways.

Unit conversions in this calculator

Albumin: g/dL × 10 → g/L.
Bilirubin: mg/dL × 17.104 → µmol/L (conventional clinical conversion).
Creatinine: µmol/L ÷ 88.4 → mg/dL for the renal-insufficiency flag.

What the Lille model is

The Lille model is a validated prognostic tool used in severe alcoholic hepatitis (SAH) for patients who have started corticosteroids (commonly prednisolone or an equivalent regimen). Unlike scores that summarize severity only at presentation, the Lille model is dynamic: it explicitly incorporates how selected laboratory values evolve after approximately one week of therapy, with particular emphasis on the change in total serum bilirubin between baseline and day 7.

The model outputs a continuous probability-style score between 0 and 1. In routine practice, the score is interpreted against a widely adopted 0.45 threshold to distinguish patients whose early trajectory suggests likely response to continued steroids from those whose trajectory suggests likely non-response, in whom prolonged corticosteroid exposure may offer little incremental benefit while retaining meaningful adverse-effect burden.

Clinical context: why a “day 7” rule helps

Severe alcoholic hepatitis carries substantial short-term mortality. Corticosteroids remain a cornerstone pharmacologic strategy for selected patients, yet a meaningful fraction do not demonstrate biochemical improvement early enough to justify a full course. Continuing steroids in that setting may increase risks such as infection and metabolic complications without improving survival, and may delay consideration of alternative pathways when those pathways are appropriate for an individual patient and care setting.

The Lille model was designed to support an operational question clinicians already ask at the bedside: after a standardized early treatment interval, does this patient’s early laboratory evolution suggest meaningful responsiveness? By quantifying that question with reproducible inputs, the model reduces ambiguity when teams discuss whether to continue, taper, or stop steroids and when to escalate discussions such as clinical trials, transplant evaluation in selected programs, or intensified supportive care.

Who should the model apply to?

The Lille model is intended for patients being treated in a pathway consistent with the cohorts used for derivation and validation: adults with severe alcoholic hepatitis treated with corticosteroids, with laboratory monitoring timed to the same treatment course. It is not a screening tool for alcoholic hepatitis, nor a substitute for confirming the clinical syndrome, excluding alternative diagnoses when appropriate, and addressing alcohol use disorder care, nutrition, infection surveillance, and bleeding risk in parallel.

Practical use assumes that “day 0” and “day 7” bilirubin values correspond to the intended steroid course and that major confounders (for example, new cholestasis from an alternative cause, hemolysis, recent transfusion effects on bilirubin interpretation, or interrupted steroid dosing) are considered before treating the output as definitive.

Required variables and measurement conventions

The published logistic formulation combines six conceptually distinct elements:

Age (years).
Serum albumin at baseline (day 0), classically expressed in grams per liter (g/L) in the primary reports; many laboratories display g/dL, which converts to g/L by multiplying by 10.
Total bilirubin at day 0 and day 7, with the model’s bilirubin terms expressed in micromoles per liter (µmol/L) in the original coefficient set. If bilirubin is recorded in mg/dL, convert using the standard clinical conversion (commonly mg/dL × 17.104 ≈ µmol/L).
Serum creatinine at day 0 used to define a binary renal insufficiency indicator in the original specification (typically creatinine above a defined mg/dL threshold; µmol/L values can be converted to mg/dL using the conventional creatinine conversion factor).
Prothrombin time at day 0 in seconds, reflecting coagulopathy severity in the derivation framework. This is not the same as INR; laboratories and assays differ, so consistency with local reporting matters for interpretability.

The hallmark “dynamic” piece is the bilirubin evolution term, constructed as (bilirubin day 0 − bilirubin day 7) after conversion to µmol/L. When bilirubin falls over the first week, this difference is positive and moves the composite index in the direction associated with better early-response patterns in the original cohorts. Flat or rising bilirubin tends to move the score toward the non-responder side of the spectrum.

How the score is calculated (conceptually)

The model is a logistic regression mapping the inputs to a linear predictor, followed by a logistic transform that yields a bounded 0–1 output. Algebraically, implementations may display equivalent forms such as 1 / (1 + exp(R)) or exp(−R) / (1 + exp(−R)), where R is the weighted sum of the standardized clinical inputs with the published coefficients for age, albumin, bilirubin evolution, baseline bilirubin, prothrombin time, and the renal insufficiency indicator.

Because the output is continuous, small differences around the threshold should be interpreted as uncertainty, not as a precise biological switch. The value is most useful as a structured anchor for multidisciplinary discussion rather than as an isolated numeric label.

Interpreting the 0.45 cutoff in practice

Guideline summaries and hepatology teaching commonly describe the following operational framework:

Lille score below 0.45 is typically interpreted as an early pattern more consistent with steroid responsiveness, supporting continuation of corticosteroids in appropriate candidates alongside close monitoring for complications.
Lille score of 0.45 or higher is typically interpreted as an early pattern more consistent with non-response, prompting reconsideration of continued steroids and earlier exploration of alternatives that match patient goals, eligibility, and center capabilities.

In the original validation reporting, patients above versus below this cutoff differed markedly in observed 6‑month survival under the treatment strategies studied, and the cutoff was described as capturing a large share of observed deaths while classifying a minority of patients as “non-responders.” These are population-level performance characteristics; individual patients may deviate for reasons the model does not capture.

How the Lille model complements—not replaces—other tools

Clinicians frequently pair the Lille model with scores that quantify severity at presentation or overall liver disease burden, such as modified Maddrey discriminant function, MELD (and sodium-adjusted variants where used), or Child–Pugh in cirrhosis contexts. Each tool answers a different question: some prioritize initial risk and treatment eligibility, while the Lille model prioritizes early treatment response after steroids begin.

Integrating multiple lines of evidence reduces the risk of over-weighting a single laboratory shift. For example, infection, acute kidney injury from competing mechanisms, or intercurrent bleeding can perturb bilirubin, creatinine, and coagulation in ways that resemble “non-response” or alternatively mask true trajectory unless clinically contextualized.

Limitations and responsible use

The model reflects the patients, treatment protocols, assay methods, and outcome definitions of its development era. Performance may differ when case mix, steroid dosing, concomitant medications, or laboratory standards diverge from those cohorts. The score should not be used to withhold general medical care, to replace shared decision-making, or to override a compelling clinical story when the inputs are incomplete or unstable.

Special caution applies when day‑7 labs are not truly comparable to baseline (timing off-schedule, different assay calibrations, intravenous fluids altering albumin, or recent albumin infusion). In those situations, repeating measurements, reconciling units, and documenting confounders is preferable to treating a single computed value as ground truth.

Using this calculator on CalcMD

This implementation computes the published logistic form using your entered values, performs unit conversions where optional alternate units are offered, applies the renal insufficiency flag from creatinine after conversion to a common mg/dL scale for threshold comparison, and displays the resulting Lille score with the 0.45 interpretive bands used in clinical teaching. Always verify that the day‑0 and day‑7 bilirubin samples align with the steroid course you intend to evaluate, and document clinical reasoning alongside the numeric output.