Non-Invasive Prediction of Non-Selective Beta-Blockers
Response in Patients with Liver Cirrhosis and High-Risk Varices

Instructions: The prediction algorithm was developed on observed percentage changes in liver and spleen stiffness (kPa) and heart rate (bpm) measured on the first day of treatment and three months after the initiation of non-selective beta-blocker (NSBB) therapy. The following predictive model is can be applied only to patients with liver cirrhosis starting NSBB as primary prophylaxis of variceal hemorrhage (i.e., patients with high-risk varices, without previous or current history of variceal hemorrhage or pharmacological/endoscopic treatment). For enhanced accuracy and user adaptability, the provided interface allows the user to manually adjust the percentages corresponding to these changes using the dedicated sliders located within the designated green section. In the event that these percentages have yet to be established, the blue section offers a calculator to compute these values by inputting the baseline values along with those registered three months after NSBB initiation. In addition, once the three values are selected via the slider, the model will automatically predict the NSBB response probability and provide a real-time assessment of the risk with results interpretation (red: high probability of being a non-responder; green: high probability of being a responder).

SPLEEN STIFFNESS % CHANGE
(3 Months From Baseline):
0.0


LIVER STIFFNESS % CHANGE
(3 Months From Baseline):
0.0


HEART RATE % CHANGE
(3 Months From Baseline):
0.0

Baseline Value
(Start of Treatment)
Value at Third Month:
Change (%)
0.0%

Interpreting the results: The model was trained on a cohort of 119 consective patients (AUC 0.96) and validated in a cohort of 34 consecutive patients (AUC 0.91), referred to the Liver Clinic at Trieste University Hospital (Italy). The combined model outperformed univariate models involving spleen stiffness (AUC 0.89, p = 0.045), liver stiffness (AUC 0.78, p < 0.001), and heart rate (AUC 0.72, p < 0.001). We selected a cut-off of 0.90, with a specificity of 100%, in order to have low false positive rates (i.e., patients who have not responded to NSBB, that were predicted as responders by the model). The cut-off also showed a statistically significant higher sensitivity (94.2%) if compared to univariate models with spleen stiffness (78.8%, p = 0.022), liver stiffness (23.1%, p < 0.001), and heart rate (22.1%, p < 0.001).

*** Disclaimer: the following model currently lack external validation and should not be used to guide patient care nor as a substitute for clinical judgment.

Developed by University of Trieste (Department of Medical, Surgical and Health Sciences) in collaboration with Yale School of Medicine (Human + Artificial Intelligence Lab)

The following machine learning risk is based on the article "Predicting Non-Invasive Response to Non-Selective Beta-Blockers with Liver and Spleen Stiffness and Heart Rate in Patients with Compensated Advanced Chronic Liver Disease and High-Risk Esophageal Varices" published in [Journal Name][Data of Publication] - DOI: [DOI].