What is a Lille Score?
The Lille score is a prognostic model calculated on day 7 of corticosteroid treatment in severe alcoholic hepatitis that predicts 6-month mortality and identifies patients who are not responding to therapy, with a score ≥0.45 indicating treatment failure and warranting cessation of corticosteroids. 1
Purpose and Clinical Use
The Lille score serves as an on-treatment prognostic tool specifically designed to assess therapeutic response to corticosteroids in patients with severe alcoholic hepatitis (Maddrey score ≥32). Unlike pre-treatment scores (mDF, MELD, GAHS, ABIC) that predict baseline mortality risk, the Lille score incorporates early changes in liver function during the first week of treatment to identify non-responders 2.
Calculation Formula
The Lille score uses the following variables 2:
R = 3.19 – 0.101 × age (years) + 0.147 × albumin day 0 (g/L) + 0.0165 × (bilirubin day 0 – bilirubin day 7 in μmol/L) – 0.206 × renal insufficiency – 0.0065 × bilirubin day 0 (μmol/L) – 0.0096 × PT (seconds)
Lille score = exp(–R) / [1 + exp(–R)]
- Renal insufficiency is coded as 0 if creatinine <1.3 mg/dL or 1 if ≥1.3 mg/dL
- Evolution in bilirubin = bilirubin level on day 0 minus day 7
Interpretation and Mortality Risk
Critical thresholds 1:
- Lille score <0.45 (Responders): 6-month survival of 70-80%
- Lille score ≥0.45 (Non-responders): 6-month survival of only 20-30%
- Lille score ≥0.56 (Null responders): Corticosteroids are no more effective than placebo; must stop steroids immediately 1
- Lille score 0.45-0.56: Consider stopping corticosteroids on a case-by-case basis 1
The original validation study showed the Lille model identifies approximately 75% of observed deaths with typically 75% mortality within 6 months if score ≥0.45 2, 3.
Timing of Assessment
Standard practice: Calculate on day 7 of corticosteroid therapy 1
Emerging evidence: Recent studies suggest the Lille score can be calculated as early as day 4 with similar prognostic accuracy 1, 4, 5, 6. One study showed day 4 Lille had comparable area under ROC curve (0.77 vs 0.75, P=0.558) with 90.3% concordance with day 7 assessment 5, 6. However, the 2022 French guidelines note this requires further validation before routine implementation 1.
Even earlier assessment at day 2 shows promise, with one 2025 study demonstrating 85% concordance with day 7 and similar predictive accuracy (AUC 0.818 vs 0.809) 7.
Clinical Decision-Making Algorithm
When Lille score ≥0.56:
- Stop corticosteroids immediately - these patients derive no benefit and face unnecessary infection risk 1
When Lille score 0.45-0.56:
- Individualized decision required
- Consider stopping if clinical deterioration or infection develops
- Weigh infection risk against potential marginal benefit
When Lille score <0.45:
- Continue prednisolone 40mg daily (or methylprednisolone 32mg daily) for full 28-day course 1
- These patients benefit from corticosteroid therapy
Combining with Other Scores
The 2022 French guidelines strongly recommend combining Lille and MELD scores for optimal mortality prediction 1. This joint-effect model outperforms either score alone and is particularly valuable for selecting candidates for early liver transplantation 1, 4. For example, a patient with MELD 21 and Lille 0.45 has 1.9-fold higher 2-month mortality risk than one with MELD 21 and Lille 0.16 (23.7% vs 12.5%) 4.
Common Pitfalls
Infection screening is mandatory before calculating the Lille score 1. Perform comprehensive bacteriological workup (blood cultures, urine culture, ascites fluid analysis) as 20-30% of patients have infection at admission and 25% develop infection during the first month of corticosteroid treatment 1. Infection developed under corticosteroids significantly increases short-term mortality risk 1.
Do not use the Lille score for initial treatment decisions - it is an on-treatment assessment tool, not a pre-treatment prognostic score 2. Use mDF ≥32 or MELD >20 to determine initial corticosteroid eligibility 4.
The Lille score has excellent sensitivity (95%) but poor specificity (19%) for predicting 30-day survival, meaning it reliably identifies responders but less accurately identifies non-responders 8. This supports its primary use as a stopping rule rather than a continuation rule.