Portal hypertension and its complications, particularly high-risk esophageal varices (HRV), remain life-threatening challenges for patients with compensated cirrhosis. While traditional surveillance using invasive endoscopy is effective, it faces limitations in terms of accessibility, cost, and patient compliance. Non-invasive diagnostic tools are particularly advantageous in well-defined populations. Compensated cirrhosis patients are the primary target group, though consensus is needed on whether sub-clinical decompensation events (e.g., minimal ascites or hepatic encephalopathy) should reclassify patients as “non-acute decompensation” for non-invasive screening [
1]. While current guidelines prioritize invasive monitoring in decompensation patients, emerging evidence suggests potential utility of non-invasive tools for longitudinal risk assessment [
2].
The emergence of non-invasive tools, such as models based on liver stiffness measurement (LSM) and spleen stiffness measurement (SSM), has shown promising diagnostic performance [
3-
5]. These tools provide dual benefits for both patients and healthcare systems. For patients, the primary benefit is the exemption from invasive procedures. Non-invasive models, such as the Baveno criteria, already allow for the safe avoidance of endoscopy in up to 70% of compensated patients, with a high negative predictive value (>95%) [
3,
5]. Preliminary studies suggest that similar potential exists for replacing hepatic venous pressure gradient (HVPG) measurements in select patient cohorts [
6]. For system-level benefits, non-invasive screening optimizes resource allocation by identifying low-risk patients (e.g., those with LSM <20 kPa and platelet count >150×109/L, otherwise SSM <40 kPa) [
7]. This approach reduces the need for endoscopic screening and surveillance, thereby conserving endoscopy resources for high-risk patients. Additionally, it proves to be cost-effective, lowering per-patient costs and enhancing compliance, particularly in resource-limited settings [
8].
Recently, the machine learning-based FIB-4plus scoring system (CHESS2004 research) has made important progress in this area [
9]. The FIB-4plus score is an innovative, machine learning-based tool designed to improve the accuracy of HRV predictions by integrating LSM, SSM, age, liver enzymes and platelet count [
9]. By outperforming conventional FIB-4 and other serum indices, it highlights how artificial intelligence (AI) can enhance existing tools and continuously improve diagnostic efficiency. Provided that this non-invasive approach with high accessibility, it would be especially suitable for quick clinical decision-making in the clinic. Other models such as LSM/SSM (VCTE) [
3,
5], the platelet count-to-spleen ratio [
10], and radiomic models [
11]—further demonstrate that multi-parametric, algorithm-driven approaches can achieve diagnostic accuracy comparable to invasive methods. The indications for these tools are expanding, reserving endoscopy for high-risk varices, and have also transitioned to risk stratification for clinically significant portal hypertension [
6,
12] and hepatic decompensation (
Fig. 1) [
2,
13,
14].
While non-invasive screening has made significant progress, critical gaps remain in guiding dynamic clinical decisions: 1) Initiating therapy: Should non-selective beta-blocker (NSBB) be started based solely on non-invasive thresholds? Establishing the thresholds of high-risk patients for NSBB initiation is very important, and current guidelines recommend NSBB prophylaxis for patients with HRV or clinically significant portal hypertension (CSPH, HVPG ≥10 mmHg) [
7]. However, non-invasive models (e.g., LSM >25 kPa or FIB-4plus >2.5) do not directly correlate with decompensation risks equivalent to HRV/CSPH populations. 2 We suggest that NSBB therapy should be initiated only when decompensation rates predicted with non-invasive risk scores are comparable to those of HRV or CSPH. Defining low-risk populations discharge from intensive portal hypertension monitoring is important as well. While noninvasive tools (e.g., LSM and SSM) can identify patients with a minimal short-term risk of decompensation, the exact threshold for defining “low-risk” remains unclear. Observational studies suggest that a decompensation rate of <1.0 event per 1,000 person-years as a potential cutoff for safely reducing the intensity of surveillance [
2]. 2) Therapy Response: A major challenge is the lack of non-invasive methods to monitor the real-time effects of NSBB therapy. The current gold standard, repeated HVPG measurement during NSBB therapy, is highly invasive. Furthermore, dynamic monitoring remains difficult, with limited data correlating current non-invasive indicators with clinical outcomes, such as reductions in variceal bleeding [
15-
17]. While some non-invasive methods are still being explored for their diagnostic potential, further validation is necessary for their application in treatment monitoring. 3) Guiding withdrawal: currently, there is no consensus regarding the use of non-invasive test to assess resolution of CSPH and to stop NSBB treatment. While a reduction in HVPG to <12 mmHg (or a >10% decrease from baseline) is associated with a reduced risk of bleeding [
18], the optimal HVPG threshold for safely discontinuing NSBB therapy remains controversial, with debates about whether it should be ≤6 mmHg or ≤8 mmHg [
19]. Prospective validation is urgently needed, if validated, tools like FIB-4plus or serial LSM/SSM measurements could potentially replace repeated HVPG assessments. These gaps highlight the need for longitudinal, outcome-driven studies to establish non-invasive endpoints that correlate with clinical events. As AI advances in this field, it is also essential that machine learning-based models need substantial improvements in both interpretability and clinical accessibility.
The future of liver disease management lies in integrating non-invasive data to provide clinic guidance. AI could enhance predictive capabilities by combining LSM, biomarkers, imaging, and electronic health record data, enabling individualized risk assessments. Wearable biosensors, such as implantable Doppler probes or skin-adherent patches, are in early trials and promise real-time monitoring of hepatic venous pressure gradients, alerting clinicians to acute changes [
20]. Bloodless biomarkers, including liquid biopsies for miRNA signatures or gut microbiome-derived metabolites, offer dynamic risk profiling without the need for venipuncture. Lastly, decentralized care models, utilizing remote AI platforms to analyze home-based LSM devices or smartphone-captured vital signs, could personalize access to specialist-level monitoring, particularly in resource-limited settings.
The FIB-4plus score represents a pioneering step in the integration of AI-driven tools into hepatology. However, while simpler algorithms based on LSM and SSM have shown excellent performance for HRV, integrating this machine learning model into clinical workflows requires the development of a plug-in for rapid FIB-4plus score calculation at the doctor's workstation.
From a clinical application perspective, it is important to consider the diagnostic efficacy of HRV, attention should also be given to the miss rate of HRV (<5%) and the endoscopic spared rate. However, the missed diagnosis rate and endoscopy-sparing rate for the FIB-4plus score have not been mentioned in this study.
Achieving a true “non-invasive revolution”, we must approach the management of cirrhosis as an integrated process from screening to therapeutic intervention. This requires rigorous validation of clinical endpoints and the establishment of regulatory frameworks for AI as a medical device. By addressing these challenges, the integration of non-invasive diagnostics, wearable monitoring technologies, and tele-medicine will not only reduce the invasiveness of diagnostic procedures but also propel hepatology into a new era of predictive and personalized medicine.
New technology must solve unmet clinical needs—not merely replace old tools with new ones.
FOOTNOTES
-
Authors’ contribution
Jinjun Chen: review and edit; Haiyu Wang: writing of the article, review and edit.
-
Acknowledgements
Prof Jinjun Chen is supported by National Key Research and Development Program of China (2022YFC2304800), National Science and Technology Major Project (2018ZX-10723203), National Natural Science Foundation of China (82070650, 82370614), Local Innovative and Research Teams Project of Guangdong Pearl River Talents Program (2017BT01S131), Clinical Research Program of Nanfang Hospital, Southern Medical University (2018CR037, 2020CR026), Clinical Research Start-up Program of Southern Medical University by High-level University Construction Funding of Guangdong Provincial Department of Education (LC2019ZD006), President Foundation of Nanfang Hospital, Southern Medical University (2019Z003) and Key-Area Research and Development Program of Guangdong Province (2019B020227004). Dr Haiyu Wang is supported by National Natural Science Foundation of China (82200674), National Postdoctoral Program for Innovative Talents of China (BX20220144) and Postdoctoral Science Foundation of China (2022M711518).
-
Conflicts of Interest
The authors have no conflicts to disclose.
Figure 1.Progress and challenges in non-invasive approaches to liver cirrhosis. CSPH, clinically significant portal hypertension.
Abbreviations
clinically significant portal hypertension
hepatic venous pressure gradient
liver stiffness measurement
non-selective beta-blocker
spleen stiffness measurement
REFERENCES
- 1. D’Amico G, Bernardi M, Angeli P. Towards a new definition of decompensated cirrhosis. J Hepatol 2022;76:202-207.
- 2. Wang H, Liang W, Zhou L, Song J, Wen B, Wu Q, et al. Baveno VI-SSM stratifies the risk of portal hypertension-related events in patients with HBV-related cirrhosis. Clin Mol Hepatol 2025;31:866-880.
- 3. Wang H, Wen B, Chang X, Wu Q, Wen W, Zhou F, et al. Baveno VI criteria and spleen stiffness measurement rule out high-risk varices in virally suppressed HBV-related cirrhosis. J Hepatol 2021;74:584-592.
- 4. Wang H, Xi R, Song J, Wen B, Zhang Y, Zhou L, et al. Combined model with acoustic radiation force impulse to rule out high-risk varices in HBV-related cirrhosis with viral suppression. Dig Liver Dis 2023;55:1062-1071.
- 5. Zhang X, Song J, Zhang Y, Wen B, Dai L, Xi R, et al. Baveno VII algorithm outperformed other models in ruling out high-risk varices in individuals with HBV-related cirrhosis. J Hepatol 2023;78:574-583.
- 6. Dajti E, Ravaioli F, Zykus R, Rautou PE, Elkrief L, Grgurevic I, et al. Accuracy of spleen stiffness measurement for the diagnosis of clinically significant portal hypertension in patients with compensated advanced chronic liver disease: a systematic review and individual patient data meta-analysis. Lancet Gastroenterol Hepatol 2023;8:816-828.
- 7. de Franchis R, Bosch J, Garcia-Tsao G, Reiberger T, Ripoll C; Baveno VII Faculty. Baveno VII - renewing consensus in portal hypertension. J Hepatol 2022;76:959-974.
- 8. Serra-Burriel M, Graupera I, Torán P, Thiele M, Roulot D, Wai-Sun Wong V, et al. Transient elastography for screening of liver fibrosis: cost-effectiveness analysis from six prospective cohorts in Europe and Asia. J Hepatol 2019;71:1141-1151.
- 9. Dong B, He R, Ju S, Chen Y, Grgurevic I, Ma J, et al. Fibrosis-4plus score: a novel machine learning-based tool for screening high-risk varices in compensated cirrhosis (CHESS2004): an international multicenter study. Clin Mol Hepatol 2025;31:881-898.
- 10. Vanderschueren E, Armandi A, Kwanten W, Cassiman D, Francque S, Schattenberg JM, et al. Spleen stiffness-based algorithms are superior to Baveno VI criteria to rule out varices needing treatment in patients with advanced chronic liver disease. Am J Gastroenterol 2024;119:1515-1524.
- 11. Yan Y, Li Y, Fan C, Zhang Y, Zhang S, Wang Z, et al. A novel machine learning-based radiomic model for diagnosing high bleeding risk esophageal varices in cirrhotic patients. Hepatol Int 2022;16:423-432.
- 12. Jachs M, Odriozola A, Turon F, Moga L, Téllez L, Fischer P, et al. Spleen stiffness measurement by vibration-controlled transient elastography at 100 Hz for non-invasive predicted diagnosis of clinically significant portal hypertension in patients with compensated advanced chronic liver disease: a modelling study. Lancet Gastroenterol Hepatol 2024;9:1111-1120.
- 13. Thiele M, Johansen S, Israelsen M, Trebicka J, Abraldes JG, Gines P, et al. Noninvasive assessment of hepatic decompensation. Hepatology 2025;81:1019-1037.
- 14. Wong YJ, Zhaojin C, Tosetti G, Degasperi E, Sharma S, Agarwal S, et al. Baveno-VII criteria to predict decompensation and initiate non-selective beta-blocker in compensated advanced chronic liver disease patients. Clin Mol Hepatol 2023;29:135-145.
- 15. Kim HY, So YH, Kim W, Ahn DW, Jung YJ, Woo H, et al. Noninvasive response prediction in prophylactic carvedilol therapy for cirrhotic patients with esophageal varices. J Hepatol 2019;70:412-422.
- 16. Alsaeid M, Sung S, Bai W, Tam M, Wong YJ, Cortes J, et al. Heterogeneity of treatment response to beta-blockers in the treatment of portal hypertension: a systematic review. Hepatol Commun 2024;8:e0321.
- 17. Trebicka J, von Heydebrand M, Lehmann J, Tofteng F, Busk T, Jensen HL, et al. Assessment of response to beta-blockers by expression of βArr2 and RhoA/ROCK2 in antrum mucosa in cirrhotic patients. J Hepatol 2016;64:1265-1273.
- 18. Cardenas A, Mendez-Bocanegra A. Report of the Baveno VI consensus workshop. Ann Hepatol 2016;15:289-290.
- 19. Suk KT. Hepatic venous pressure gradient: clinical use in chronic liver disease. Clin Mol Hepatol 2014;20:6-14.
- 20. Kim J, Campbell AS, de Ávila BE, Wang J. Wearable biosensors for healthcare monitoring. Nat Biotechnol 2019;37:389-406.
Citations
Citations to this article as recorded by
