A validated, interpretable prediction model for advanced fibrosis risk stratification
in patients with indeterminate FIB-4 scores (1.3–2.67)
Approximately 25–35% of patients with metabolic dysfunction-associated steatotic liver disease (MASLD) fall in the indeterminate FIB-4 zone (1.3–2.67), where current guidelines offer limited diagnostic guidance. This project presents a simple, laboratory-based logistic regression model that provides additional risk stratification using 10 routinely available clinical variables.
| AUROC — Internal Validation | AUROC — Asian External Validation | AUROC — NHANES Generalizability |
An interactive calculator implementing the prediction model is freely available:
https://laithomari.github.io/advanced_fibrosis_calculator/
Enter routine clinical values and receive an instant risk classification (Low / Intermediate / High) with predicted probability of advanced fibrosis.
The model uses 10 features selected via recursive feature elimination from 24 candidates:
| Variable | Type | Unit |
|---|---|---|
| Age | Continuous | years |
| BMI | Continuous | kg/m² |
| AST | Continuous | U/L |
| ALT | Continuous | U/L |
| GGT | Continuous | U/L |
| Platelet count | Continuous | ×10&sup9;/L |
| HbA1c | Continuous | % |
| Total cholesterol | Continuous | mg/dL |
| AST/ALT ratio | Derived | — |
| Diabetes status | Binary | Yes/No |
| Category | Threshold | Clinical Action |
|---|---|---|
| Low Risk | Probability < 0.25 | Reassurance; monitor in primary care |
| Intermediate | 0.25 ≤ Probability < 0.62 | Consider second-line testing (e.g., VCTE) |
| High Risk | Probability ≥ 0.62 | Refer to hepatology |
Figure 1. Receiver operating characteristic curves across all three validation cohorts.
| Cohort | N | Prevalence | AUROC (95% CI) | O/E Ratio |
|---|---|---|---|---|
| Internal Validation | 213 | 41.8% | 0.819 (0.758–0.873) | 0.94 |
| Asian (External) | 203 | 44.3% | 0.740 (0.671–0.810) | 1.12 |
| NHANES | 1,503 | 10.7% | 0.753 (0.711–0.798) | 0.41 |
Figure 2. Calibration plots comparing predicted vs. observed event rates.
Figure 3. Decision curve analysis demonstrating net benefit across clinically relevant thresholds.
| Threshold | Sensitivity | Specificity | PPV | NPV |
|---|---|---|---|---|
| Rule-out (0.25) | 95.5% | 32.3% | 50.3% | 90.9% |
| Youden (0.49) | 66.3% | 79.8% | 70.2% | 76.7% |
| Rule-in (0.62) | 42.7% | 89.5% | 74.5% | 68.5% |
| Cohort | Source | N (Indeterminate) | Reference Standard |
|---|---|---|---|
| NAFLD DB2 | NIDDK Central Repository | 608 | Liver biopsy (F3–F4) |
| LiveFbr Asian | China & Malaysia | 203 | Liver biopsy (S3–S4) |
| NHANES 2017–2020 | CDC | 1,503 | LSM ≥10 kPa |
advanced_fibrosis_calculator/
index.html # Online calculator (deployed via GitHub Pages)
README.md # This file
assets/ # Figures and images
banner.svg
figure3_roc_curves.png
figure4_calibration.png
figure5_dca.png
- NAFLD DB2: Available from the NIDDK Central Repository
- NHANES 2017–2020: Publicly available from the CDC
- LiveFbr cohort: Described in Sang et al., Front Med 2021
If you use this calculator or model in your research, please cite:
A Simple Model to Predict Advanced Fibrosis in MASLD Patients with Indeterminate FIB-4. [Manuscript in preparation]
This project is licensed under the MIT License. See LICENSE for details.