Assessment of portal hypertension severity using machine learning models in patients with compensated cirrhosis

Simple blood tests to diagnose compensated advanced chronic liver disease and stratify the risk of clinically significant portal hypertension

BACKGROUND AND AIMS

Compensated advanced chronic liver disease (cACLD) identifies patients at risk for clinically significant portal hypertension (CSPH), and thus, for liver-related complications. The limited availability of liver stiffness measurements (LSM) impedes the identification of patients at risk for cACLD/CSPH outside of specialized clinics. We aimed to develop a blood-based algorithm to identify cACLD by fibrosis-4 (FIB-4) and CSPH by von Willebrand factor/platelet count ratio (VITRO).

APPROACH AND RESULTS

Patients with (suspected) compensated chronic liver disease undergoing FIB-4+LSM were included in the LSM/FIB-4 cohorts from Vienna and Salzburg. The HVPG/VITRO cohorts included patients undergoing HVPG-measurement + VITRO from Vienna and Bern.LSM/FIB-4-derivation-cohort: We included 6143 patients, of whom 211 (3.4%) developed hepatic decompensation. In all, 1724 (28.1%) had LSM ≥ 10 kPa, which corresponded to FIB-4 ≥ 1.75. Importantly, both LSM (AUROC:0.897 [95% CI:0.865-0.929]) and FIB-4 (AUROC:0.914 [95% CI:0.885-0.944]) were similarly accurate in predicting hepatic decompensation within 3 years. FIB-4 ≥ 1.75 identified patients at risk for first hepatic decompensation (5 y-cumulative incidence:7.6%), while in those <1.75, the risk was negligible (0.3%).HVPG/VITRO-derivation cohort: 247 patients of whom 202 had cACLD/FIB-4 ≥ 1.75 were included. VITRO exhibited an excellent diagnostic performance for CSPH (AUROC:0.889 [95% CI:0.844-0.934]), similar to LSM (AUROC:0.856 [95% CI:0.801-0.910], p = 0.351) and the ANTICIPATE model (AUROC:0.910 [95% CI:0.869-0.952], p = 0.498). VITRO < 1.0/ ≥ 2.5 ruled-out (sensitivity:100.0%)/ruled-in (specificity:92.4%) CSPH. The diagnostic performance was comparable to the Baveno-VII criteria.LSM/FIB-4-derivation cohort findings were externally validated in n = 1560 patients, while HVPG/VITRO-derivation-cohort findings were internally (n = 133) and externally (n = 55) validated.

CONCLUSIONS

Simple, broadly available laboratory tests (FIB-4/VITRO) facilitate cACLD detection and CSPH risk stratification in patients with (suspected) liver disease. This blood-based approach is applicable outside of specialized clinics and may promote early intervention.

How to assess hemostasis in patients with severe liver disease

Patients with advanced liver diseases frequently acquire profound alterations in their hemostatic system. Simultaneous changes in procoagulant and anticoagulant systems result in a reset in the hemostatic balance with a relatively neutral net effect, although there are notable hypocoagulable and hypercoagulable features in the hemostatic system in patients with liver disease. Laboratory and clinical studies have demonstrated that patients have a relatively well-preserved hemostatic system even though routine diagnostic tests of hemostasis (prothrombin time, platelet count) suggest a bleeding tendency. Routine diagnostic tests of hemostasis are unsuitable to assess the hemostatic status of patients with liver disease, as these tests are insensitive for the concurrent prohemostatic and antihemostatic changes in these patients. These tests are, however, frequently requested in patients with liver disease, as they are well established indicators of severity of liver disease. This paper will discuss commonly used diagnostic and research-type hemostatic tests and will outline how test results should be interpreted in patients with liver disease.

Hepatic venous pressure gradient: Inaccurately estimates portal venous pressure gradient in alcoholic cirrhosis and portal hypertension

BACKGROUND

Portal hypertension (PHT) in patients with alcoholic cirrhosis causes a range of clinical symptoms, including gastroesophageal varices and ascites. The hepatic venous pressure gradient (HVPG), which is easier to measure, has replaced the portal venous pressure gradient (PPG) as the gold standard for diagnosing PHT in clinical practice. Therefore, attention should be paid to the correlation between HVPG and PPG.

AIM

To explore the correlation between HVPG and PPG in patients with alcoholic cirrhosis and PHT.

METHODS

Between January 2017 and June 2020, 134 patients with alcoholic cirrhosis and PHT who met the inclusion criteria underwent various pressure measurements during transjugular intrahepatic portosystemic shunt procedures. Correlations were assessed using Pearson's correlation coefficient to estimate the correlation coefficient (r) and determination coefficient (R2). Bland-Altman plots were constructed to further analyze the agreement between the measurements. Disagreements were analyzed using paired t tests, and P values < 0.05 were considered statistically significant.

RESULTS

In this study, the correlation coefficient (r) and determination coefficient (R2) between HVPG and PPG were 0.201 and 0.040, respectively (P = 0.020). In the 108 patients with no collateral branch, the average wedged hepatic venous pressure was lower than the average portal venous pressure (30.65 ± 8.17 vs. 33.25 ± 6.60 mmHg, P = 0.002). Hepatic collaterals were identified in 26 cases with balloon occlusion hepatic venography (19.4%), while the average PPG was significantly higher than the average HVPG (25.94 ± 7.42 mmHg vs 9.86 ± 7.44 mmHg; P < 0.001). The differences between HVPG and PPG < 5 mmHg in the collateral vs no collateral branch groups were three cases (11.54%) and 44 cases (40.74%), respectively.

CONCLUSION

In most patients, HVPG cannot accurately represent PPG. The formation of hepatic collaterals is a vital reason for the strong underestimation of HVPG.

Deep Neural Network Regression to Assist Non-Invasive Diagnosis of Portal Hypertension

Portal hypertension is a complex medical condition characterized by elevated blood pressure in the portal venous system. The conventional diagnosis of such disease often involves invasive procedures such as liver biopsy, endoscopy, or imaging techniques with contrast agents, which can be uncomfortable for patients and carry inherent risks. This study presents a deep neural network method in support of the non-invasive diagnosis of portal hypertension in patients with chronic liver diseases. The proposed method utilizes readily available clinical data, thus eliminating the need for invasive procedures. A dataset composed of standard laboratory parameters is used to train and validate the deep neural network regressor. The experimental results exhibit reasonable performance in distinguishing patients with portal hypertension from healthy individuals. Such performances may be improved by using larger datasets of high quality. These findings suggest that deep neural networks can serve as useful auxiliary diagnostic tools, aiding healthcare professionals in making timely and accurate decisions for patients suspected of having portal hypertension.

Oesophageal varices predict complications in compensated advanced non-alcoholic fatty liver disease

Background & Aims

We aimed to evaluate the impact of oesophageal varices (OV) and their evolution on the risk of complications of compensated advanced chronic liver disease (cACLD) caused by non-alcoholic fatty liver disease (NAFLD). We also assessed the accuracy of non-invasive scores for predicting the development of complications and for identifying patients at low risk of high-risk OV.

Methods

We performed a retrospective assessment of 629 patients with NAFLD-related cACLD who had baseline and follow-up oesophagogastroduodenoscopy and clinical follow-up to record decompensation, portal vein thrombosis (PVT), and hepatocellular carcinoma.

Results

Small and large OV were observed at baseline in 30 and 15.9% of patients, respectively. The 4-year incidence of OV from absence at baseline, and that of progression from small to large OV were 16.3 and 22.4%, respectively. Diabetes and a ≥5% increase in BMI were associated with OV progression. Multivariate Cox regression revealed that small (hazard ratio [HR] 2.24, 95% CI 1.47-3.41) and large (HR 3.86, 95% CI 2.34-6.39) OV were independently associated with decompensation. When considering OV status and trajectories, small (HR 2.65, 95% CI 1.39-5.05) and large (HR 4.90, 95% CI 2.49-9.63) OV at baseline and/or follow-up were independently associated with decompensation compared with the absence of OV at baseline and/or follow-up. The presence of either small (HR 2.8, 95% CI 1.16-6.74) or large (HR 5.29, 95% CI 1.96-14.2) OV was also independently associated with incident PVT.

Conclusion

In NAFLD-related cACLD, the presence, severity, and evolution of OV stratify the risk of developing decompensation and PVT.

Impact and implications

Portal hypertension is the main driver of liver decompensation in chronic liver diseases, and its non-invasive markers can help risk prediction. The presence, severity, and progression of oesophageal varices stratify the risk of complications of non-alcoholic fatty liver disease. Easily obtainable laboratory values and liver stiffness measurement can identify patients at low risk for whom endoscopy may be withheld, and can also stratify the risk of liver-related complications.

CADUCEO: A Platform to Support Federated Healthcare Facilities through Artificial Intelligence

Data-driven algorithms have proven to be effective for a variety of medical tasks, including disease categorization and prediction, personalized medicine design, and imaging diagnostics. Although their performance is frequently on par with that of clinicians, their widespread use is constrained by a number of obstacles, including the requirement for high-quality data that are typical of the population, the difficulty of explaining how they operate, and ethical and regulatory concerns. The use of data augmentation and synthetic data generation methodologies, such as federated learning and explainable artificial intelligence ones, could provide a viable solution to the current issues, facilitating the widespread application of artificial intelligence algorithms in the clinical application domain and reducing the time needed for prevention, diagnosis, and prognosis by up to 70%. To this end, a novel AI-based functional framework is conceived and presented in this paper.