ELasto-ML (ELM) Score

Clinically Significant Portal Hypertension Risk Calculator

  1. Select the appropriate elastography modality (VCTE, P-SWE, or 2D-SWE) and measurement unit (kPa or m/s).
  2. Input patient's Liver Stiffness Measurement (LSM) and Spleen Stiffness Measurement (SSM) values.
  3. Enter platelet count value.
  4. Select patient's liver disease etiology.
  5. Input additional parameters: Child-Pugh score, Age, and Gender.
  6. Click "Calculate" to generate the CSPH probability assessment.
  7. Review the results and explore the Risk Interpretation Dashboard for detailed analysis.

This tool should be used for:

  • Risk stratification for patients with suspected portal hypertension.
  • Patients with compensated advanced chronic liver disease (cACLD) or compensated cirrhosis.

Do not use for patients with decompensated cirrhosis, acute liver failure, or without confirmed advanced chronic liver disease.

Risk Category Probability Score Interpretation
High Risk (Red) ≥ 0.60 Rule-in CSPH (HVPG ≥ 10 mmHg)
Intermediate Risk (Gray) 0.46 - 0.59 Cannot safely rule-in or rule-out CSPH
Low Risk (Green) ≤ 0.45 Rule-out CSPH (HVPG < 10 mmHg)
Warning: The model will display alerts when 2+ key variables fall outside the distribution observed in the training data (above the 95th percentile or below the 5th percentile), indicating potential reduced model accuracy.

This predictive tool is designed to assess the likelihood of clinically significant portal hypertension (CSPH) in patients with compensated advanced chronic liver disease (cACLD).

Key Features:

  • Compatible with three elastography modalities: VCTE, P-SWE, and 2D-SWE.
  • Primary predictors: LSM, SSM, and platelet count.
  • Validated across multiple liver disease etiologies.
  • Significantly reduces "gray zone" compared to Baveno VII criteria.
  • Features an interpretable AI model with transparent decision visualization.

Locked — you are currently using the model without spleen stiffness. Select "Spleen Stiffness Measurement Available" to view this section.

Study Results: In developing this model, we aimed to create a tool applicable across all elastography methodologies, maintaining a Negative Predictive Value (NPV) ≥ 90% while simultaneously reducing the "gray zone" — those patients who, under current Baveno VII criteria, are not classified in either the rule-out or rule-in categories.

The model's cut-offs were selected in the training set to achieve 100% sensitivity to rule-out CSPH and 100% specificity to rule-in CSPH. When tested on the external validation cohort, these cut-offs demonstrated:

  • Sensitivity: 0.94 (95% C.I.: 0.91-0.97)
  • NPV: 0.90 (95% C.I.: 0.84-0.94)
  • Specificity: 0.94 (95% C.I.: 0.80-0.98)
  • PPV: 0.96 (95% C.I.: 0.92-0.98)
  • Gray Zone: only 12.3% of patients

Comprehensive Comparison Across Models: Our analysis demonstrated significant improvement in reducing gray zone patients across all elastography modalities:

Elastography Type Baveno VII Baveno VII-SSM Dual Cut-Off Baveno VII-SSM Single Cut-Off ELM Score
VCTE-50Hz Module 47.0% 33.2% 10.6% 7.3%
VCTE-100Hz Module 47.3% 37.5% 18.2% 12.0%
2D-SWE 45.0% 44.2% 17.8% 11.6%
P-SWE 55.5% 49.1% 20.3% 12.3%

Values represent percentage of patients falling in the "gray zone" (where CSPH cannot be reliably ruled in or ruled out)

Performance by Etiology: The ELM Score showed varied performance across different liver disease etiologies:

  • Viral Etiology (HBV/HCV): 8.0% gray zone
  • Metabolic Associated Steatotic Liver Disease: 7.1% gray zone
  • Alcohol-associated Liver Disease: 10.7% gray zone
  • Autoimmune Liver Disease: 5.0% gray zone
  • Other/Mixed Etiology: 14.2% gray zone

Key Insight: The ELM Score consistently demonstrates superior performance across all elastography modalities, reducing the proportion of patients in the diagnostic "gray zone" by up to 85% compared to traditional Baveno VII criteria.

Locked — you are currently using the full model with spleen stiffness. Select "Spleen Stiffness Measurement Unavailable" to view this section.

Clinical Performance of ELM Score with Missing SSM Values: Random Forest models possess the unique capability to handle missing data through surrogate splits, allowing predictions even when certain features are unavailable. To assess the clinical applicability of ELM when spleen stiffness measurement is unavailable or technically unfeasible, we exploited this property by evaluating model performance with SSM treated as missing across all three cohorts.

The same cut-offs derived from the full model were applied (rule-out ≤ 0.45; rule-in ≥ 0.60). As shown in the table below, the model without SSM maintained reasonable discrimination across all cohorts.

RULE-OUT: ELM SCORE ≤ 0.45
Metrics Training Cohort Internal Validation Cohort External Validation Cohort
Sensitivity 0.95 (0.93–0.96) 0.95 (0.88–0.98) 0.95 (0.91–0.97)
NPV 0.88 (0.83–0.91) 0.91 (0.80–0.97) 0.88 (0.79–0.93)
RULE-IN: ELM SCORE ≥ 0.60
Metrics Training Cohort Internal Validation Cohort External Validation Cohort
Specificity 0.81 (0.76–0.84) 0.86 (0.75–0.92) 0.85 (0.79–0.91)
PPV 0.90 (0.87–0.92) 0.89 (0.80–0.94) 0.88 (0.86–0.94)
Gray Zone 110/943 (11.6%) 25/150 (16.6%) 55/342 (16.0%)

Performance of the ELM Score with spleen stiffness measurement treated as missing across training, internal validation, and external validation cohorts. Abbreviations: CSPH: clinically significant portal hypertension; NPV: negative predictive value; PPV: positive predictive value.

While the model without SSM demonstrated acceptable AUC values and maintained high sensitivity for rule-out, both NPV and PPV fell below the 0.90 threshold recommended by the Baveno VII consensus for adequate diagnostic models. Additionally, the gray-zone increased substantially compared to the full model (16.0% vs. 12.3% in external validation). These findings confirm the incremental diagnostic value of incorporating spleen stiffness measurement into the ELM Score.

Research Use Only: This tool is provided exclusively for research and educational purposes and is not intended for clinical decision-making.

Not a Diagnostic Device: The probability estimates generated by this tool do not replace invasive HVPG measurement, which remains the gold standard for assessing portal hypertension.

No Regulatory Approval: This tool is not approved or cleared by any regulatory authority (including FDA or EU regulators) and must not be used to guide medical care or treatment decisions.

Clinical Responsibility: Patients should never use this tool for self-diagnosis or self-management. Clinical evaluation and management must always be performed by qualified healthcare professionals.

ELM Score --


Developed by the LIVER-MAINDS Lab at University of Trieste

The following machine learning model is based on the article "Title" published in [Journal Name][Data of Publication] - DOI: [DOI]