External Validation of Laboratory-Based Models to Assess Portal Hypertension Severity in Patients with Compensated Chronic Liver Disease

External Validation of Laboratory-Based Models to Assess Portal Hypertension Severity in Patients with Compensated Chronic Liver Disease | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article External Validation of Laboratory-Based Models to Assess Portal Hypertension Severity in Patients with Compensated Chronic Liver Disease Fahad Mohammed, Minzhi Xing, Michael Mohnasky, Andrew Caddell, and 8 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8845244/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 10 You are reading this latest preprint version Abstract Purpose Non-invasive laboratory-based models have been proposed to estimate portal hypertension severity in patients with compensated advanced chronic liver disease (cACLD), but external validation in diverse populations remains limited. We aimed to externally validate two such models, originally developed in European cohorts, for predicting clinically significant portal hypertension (CSPH; hepatic venous pressure gradient [HVPG] ≥ 10 mmHg) and severe portal hypertension (HVPG ≥ 16 mmHg) in a U.S.-based cACLD cohort. Methods We conducted a retrospective single-center study of adults with cACLD who underwent HVPG measurement between 2014 and 2024. Patients with active hepatic decompensation, hepatocellular carcinoma, or prior transjugular intrahepatic portosystemic shunt were excluded. Model performance of the Vienna laboratory-based model and the FIB-4 plus albumin (FIB4+) model was evaluated using discrimination (area under the receiver operating characteristic curve [AUROC]) and calibration metrics, including calibration intercepts, slopes, and Brier scores. Results The cohort included 143 patients (median age 56 years; 54% female), predominantly with metabolic dysfunction–associated steatotic liver disease or alcohol-related liver disease. Median HVPG was 7 mmHg, with CSPH present in 33% and HVPG ≥ 16 mmHg in 11%. Discrimination was modest for both models. The Vienna model achieved AUROCs of 0.71 for HVPG ≥ 10 mmHg and 0.77 for HVPG ≥ 16 mmHg, while the FIB4 + model achieved AUROCs of 0.69 and 0.71, respectively. Both models systematically overestimated risk, demonstrating poor calibration across thresholds. Negative predictive values for HVPG ≥ 16 mmHg exceeded 95% for both models. Predictive performance was weaker in non-viral etiologies. Conclusion In this U.S.-based external validation, laboratory-based models for portal hypertension showed modest discrimination and poor calibration, with systematic risk overestimation. While high negative predictive values suggest potential utility as rule-out tools for severe portal hypertension, recalibration and prospective validation are required before clinical implementation. Hepatic Venous Pressure Gradient Portal Hypertension Compensated Cirrhosis External Validation Figures Figure 1 Figure 2 Figure 3 Introduction Portal hypertension is a major driver of disease progression in patients with compensated advanced chronic liver disease (cACLD) and significantly impacts clinical outcomes. The hepatic venous pressure gradient (HVPG) is the gold standard for assessing the severity of portal hypertension and serves as a crucial prognostic marker in chronic liver disease. An HVPG of ≥ 10 mmHg defines clinically significant portal hypertension (CSPH), which is associated with an increased risk of hepatic decompensation, including the development of ascites, variceal bleeding, and hepatic encephalopathy [ 1 ]. Furthermore, an HVPG of ≥ 16 mmHg is predictive of more severe complications including variceal bleeding and refractory ascites and is associated with increased mortality risk [ 2 ]. Despite its utility, HVPG measurement is an invasive procedure requiring specialized expertise and is not widely available in many clinical settings. Consequently, there is a need for reliable, non-invasive methods to accurately assess portal hypertension severity and facilitate risk stratification. Several predictive models have been developed to estimate portal hypertension severity using widely available laboratory parameters. Image-based fibrosis assessment (e.g. transient elastography) has substantial data for predicting decompensation risk and has been adopted by the Baveno VII guidelines [ 3 ]. However, transient elastography is not available in all settings, and reliance on image-based tools may worsen disparities in liver disease care. Machine learning algorithms have increasingly been utilized to develop non-invasive models that predict CSPH and severe portal hypertension with high accuracy. One such machine learning model, the Vienna Model, was designed to predict HVPG ≥ 10 mmHg and ≥ 16 mmHg using three laboratory parameters: platelet count, total bilirubin, and international normalized ratio (INR) [ 4 ]. This model was developed and validated in a European cohort, demonstrating robust predictive performance. Another non-invasive model, the FIB4 + model, incorporates the Fibrosis-4 (FIB-4) index, which includes age, aspartate aminotransferase (AST), alanine aminotransferase (ALT), and platelet count, along with serum albumin, to predict CSPH. The FIB4 + model was proposed as an alternative to liver stiffness measurement (LSM) and transient elastography, which, although widely used, are not always accessible in routine clinical practice [ 5 ]. While these models have shown promising results in their initial validation studies, their performance has not been extensively evaluated in independent, real-world patient populations. Machine learning models are at particular risk of overfitting and demonstrating poor external validity. External validation is essential to determine whether these models can be generalized across different patient cohorts, particularly in populations with different demographic characteristics, liver disease etiologies, and clinical practices. Prior studies have demonstrated that the performance of non-invasive predictive models may vary depending on factors such as the prevalence of CSPH, obesity, and liver disease etiology. Therefore, external validation in a US-based cohort is critical to assess the generalizability and clinical applicability of these models. However, transportability of non-invasive portal hypertension models across populations remains uncertain. Differences in disease etiology, portal hypertension prevalence, and referral patterns may substantially affect model performance and calibration. In particular, U.S. cohorts are increasingly dominated by metabolic dysfunction–associated steatotic liver disease (MASLD), in which portal hypertension may include a presinusoidal component that is incompletely captured by HVPG. External validation in contemporary U.S. populations is therefore critical, not only to assess discrimination but also to evaluate calibration and clinical reliability. The primary objective of this study was to externally validate two laboratory-based models, the Vienna Model and the FIB4 + Model, for assessing portal hypertension severity in a cohort of patients with cACLD who underwent HVPG measurement at UNC Health. Specifically, we aimed to evaluate the ability of these models to predict HVPG ≥ 10 mmHg and ≥ 16 mmHg using non-invasive laboratory parameters. Model performance was assessed in terms of discrimination, calibration, and overall predictive accuracy. Additionally, we conducted subgroup analyses to determine whether the predictive performance of these models varied by demographic and clinical characteristics, including age, sex, and etiology of liver disease. By validating these models in an independent cohort, this study seeks to provide insight into their clinical utility and potential role in guiding non-invasive risk stratification for patients with cACLD. Methods Study Design and Patient Selection This was a retrospective cohort study conducted at UNC Health to externally validate two non-invasive models for assessing portal hypertension severity in patients with compensated advanced chronic liver disease (cACLD). The study included patients who underwent percutaneous transjugular hepatic venous pressure gradient (HVPG) measurement between May 2014 and April 2024. HVPG measurements were performed via a transjugular approach using a balloon-occlusion catheter, with wedged hepatic venous pressure obtained following balloon inflation to achieve complete hepatic vein occlusion and free hepatic venous pressure measured in the same vein; HVPG was calculated as the difference between wedged and free pressures. Eligible patients were identified through the electronic medical record system, and data were extracted from clinical notes, procedural reports, and laboratory results. Transient elastography data were only available for a small minority of patients and were therefore not included in the analysis. To ensure a clinically relevant cohort, inclusion criteria required that patients had a diagnosis of cACLD and underwent HVPG measurement as part of their clinical evaluation. cACLD was defined by histologic evidence of F3–F4 fibrosis in the absence of active hepatic decompensation at the time of HVPG measurement. Patients were excluded if they had current hepatic decompensation, defined as the presence of ascites, gastroesophageal variceal bleeding, or hepatic encephalopathy at the time of HVPG measurement. Patients with active hepatic decompensation at the time of HVPG measurement were excluded; patients with prior remote decompensation without active manifestations were not excluded. Patients with current or prior hepatic malignancy, as well as those with a history of transjugular intrahepatic portosystemic shunt (TIPS), were also excluded. Additionally, patients were required to have available laboratory data, including platelet count, bilirubin, international normalized ratio (INR), AST, ALT, and albumin, within 30 days of HVPG measurement. Data Collection and Variables Patient demographic information, including age, sex, race, and etiology of liver disease, was extracted from the electronic health record system. Clinical data included the presence of metabolic dysfunction-associated steatotic liver disease (MASLD), alcohol-related liver disease (ALD), viral hepatitis, autoimmune hepatitis, or other chronic liver diseases. Indications for the patient’s biopsy were also recorded. Laboratory parameters collected included platelet count (×10⁹/L), total bilirubin (mg/dL), INR, AST (U/L), ALT (U/L), and albumin (g/dL), as these were integral to the predictive models being validated. HVPG values were obtained from interventional radiology procedural reports, with 11 different radiologists performing the procedures. HVPG was measured using a transjugular catheter technique, in which the free hepatic venous pressure (FHVP) and wedged hepatic venous pressure (WHVP) were recorded. HVPG was calculated as the difference between WHVP and FHVP. Portal hypertension severity was categorized as clinically significant portal hypertension (CSPH) if HVPG was ≥ 10 mmHg and as severe portal hypertension if HVPG was ≥ 16 mmHg. HVPG measurements were obtained selectively, most commonly as part of liver transplant evaluation, diagnostic clarification of portal hypertension severity, or enrollment in clinical research protocols. HVPG was not performed routinely in all patients with cACLD, and the study cohort therefore represents a selected subset of patients evaluated at our center during the study period. Use of non-selective beta-blockers at the time of HVPG measurement was abstracted from the medical record where available. Predictive Models for External Validation Two predictive models for assessing portal hypertension severity were externally validated: 1. Vienna Model – This model was originally developed using a European cohort and utilizes three laboratory parameters: platelet count, total bilirubin, and INR [ 4 ]. It was designed to predict HVPG ≥ 10 mmHg and ≥ 16 mmHg based on a machine-learning approach that optimizes classification performance. The formula for the prediction of severe PH is: \(\:{\text{H}\text{V}\text{P}\text{G}}_{\text{p}\text{r}\text{o}\text{b}\text{a}\text{b}\text{i}\text{l}\text{i}\text{t}\text{y}}={\sigma\:}({{\beta\:}}_{0}+{{\beta\:}}_{1}\times\:\text{p}\text{l}\text{a}\text{t}\text{e}\text{l}\text{e}\text{t}\:\text{c}\text{o}\text{u}\text{n}\text{t}+{{\beta\:}}_{2}\times\:\text{s}\text{e}\text{r}\text{u}\text{m}\:\text{b}\text{i}\text{l}\text{i}\text{r}\text{u}\text{b}\text{i}\text{n}+{{\beta\:}}_{3}\times\:\text{I}\text{N}\text{R}\) ) where σ is a sigmoid function calculated as σ(x)= \(\:\frac{1}{1+{e}^{-x}}\) , e=Euler’s number, platelet count units are 10 9 /L, and serum bilirubin units are mg/dl. The extracted coefficients β n both for HVPG > 10 and > 16 mmHg predictions are available in the supplementary materials. 2. FIB4 + Model – This model incorporates the Fibrosis-4 (FIB4) index, which is calculated using age, AST, ALT, and platelet count, along with serum albumin [ 5 ]. The model was developed as an alternative to elastography-based methods for predicting CSPH in patients with compensated liver disease. The formula for the final calibrated model is: $$\:\text{L}\text{o}\text{g}\:\text{o}\text{d}\text{d}\text{s}\left(\text{C}\text{S}\text{P}\text{H}\right)=0.7207-\left(0.6729\times\:\text{a}\text{l}\text{b}\text{u}\text{m}\text{i}\text{n}\right)+(0.4408\times\:\text{F}\text{I}\text{B}4)$$ where FIB4= \(\:\frac{\text{A}\text{g}\text{e}\times\:\text{A}\text{S}\text{T}}{\text{P}\text{l}\text{a}\text{t}\text{e}\text{l}\text{e}\text{t}\text{s}\times\:\surd\:\text{A}\text{L}\text{T}}\) , age units are years, AST/ALT units are U/L, platelets units are10 9 /L, and albumin units are g/ds. Both models were applied to the UNC Health cohort to assess their predictive performance in an independent, real-world US population. Statistical Analysis Descriptive statistics were used to summarize baseline patient characteristics, with continuous variables reported as median with interquartile range (IQR) and categorical variables presented as frequencies and percentages. The predictive performance of the Vienna Model and FIB4 + Model was assessed by calculating the area under the receiver operating characteristic (AUROC) curve for HVPG thresholds of ≥ 10 mmHg and ≥ 16 mmHg. Confidence intervals were estimated using bootstrap resampling with 1000 repetitions. AUROC values between 0.7 and 0.8 were considered indicative of acceptable discrimination, while values above 0.8 were considered excellent. To evaluate model calibration, calibration plots were generated comparing predicted vs. observed probabilities of HVPG ≥ 10 mmHg and ≥ 16 mmHg. We fit logistic recalibration models regressing the observed outcome on the logit of the predicted probability to estimate the calibration intercept (ideal value 0) and calibration slope (ideal value 1). These parameters describe whether the model systematically over- or underestimates risk and whether predictions are overly extreme or too narrow. Brier scores were computed (with bootstrap 95% CIs) to assess the overall accuracy of probability estimates, with lower scores indicating better calibration. Additionally, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV), with exact binomial CIs, were calculated for both models at optimal cutoff thresholds determined by the Youden index. All analyses were conducted in RStudio (version 2024.12.1) using the pROC, rms, boot, and binom packages. Subgroup analyses were performed to assess model performance across different patient populations, including stratification by age, sex, race, and etiology of liver disease. Logistic regression models were used to examine the association between individual laboratory parameters and HVPG thresholds, controlling for potential confounders such as age and sex. Results Cohort characteristics We identified 143 adults with compensated chronic liver disease who underwent HVPG measurement. Baseline characteristics are included in Table 1 . Median age was 56 years (IQR 50–64) and 53.9% were female. Most patients had a liver disease etiology of MASLD 31.5%, ALD 16.8%, or viral hepatitis 13.3%. Median HVPG was 7 mmHg (IQR 4–11). The prevalence of HVPG ≥ 10 mmHg was 32.9% and of HVPG ≥ 16 mmHg was 11.2%. Table 1 Patient Characteristics Variable Total Patients, n 143 Age, years (median, IQR) 56 (50–64) Sex, n (%) Female 77 (53.9%) Male 66 (46.2%) Race, n (%) American Indian or Alaska Native 3 (2.1%) Asian 1 (0.7%) Black 44 (31.0%) White 83 (58.5%) Other 11 (7.8%) Ethnicity, n (%) Hispanic/Latino 9 (6.3%) Non-Hispanic 134 (93.7%) Chronic Liver Disease Etiology ALD 24 (16.8%) Viral Hepatitis 19 (13.3%) ALD + Viral Hepatitis 3 (2.1%) Autoimmune Hepatitis 19 (13.3%) MASLD 45 (31.5%) Cryptogenic 22 (15.4%) Other 24 (16.7%) HVPG, median (IQR) 7 (4–11) Platelet Count, G/L, median (IQR) 131 (88–193) Total Bilirubin, mg/dL, median (IQR) 0.8 (0.5–1.7) INR, median (IQR) 1.2 (1.1–1.4) AST, U/L, median (IQR) 47 (30–79) ALT, U/L, median (IQR) 40 (24–69) Albumin, g/dL, median (IQR) 3.7 (3.0-4.1) Non-selective beta blockers 49 (34.3%) Indications for HVPG measurement Elevated liver enzymes 40 (28.0%) Cirrhosis of unknown etiology 28 (19.6%) Transplant evaluation 37 (25.9%) Acute liver injury of unknown etiology 8 (5.6%) Evaluate for cirrhosis/fibrosis 57 (39.9%) Graft-versus-host disease concern 0 (0.0%) Transplant rejection concern 2 (1.4%) Ascites of unknown etiology 0 (0.0%) Other 9 (6.3%) Upper endoscopy/imaging findings for portal hypertension complications Gastric varices (non-bleeding) 3 (2.1%) Esophageal varices (non-bleeding) 34 (23.8%) Rectal varices (non-bleeding) 0 (0.0%) Portal hypertensive gastropathy 15 (10.5%) Gastrorenal shunt 0 (0.0%) Splenorenal shunt 2 (1.4%) Gastroesophageal varices (non-bleeding) 1 (0.7%) Perisplenic varices (non-bleeding) 6 (4.2%) Peripancreatic varices (non-bleeding) 0 (0.0%) Unspecified varices (non-bleeding) 0 (0.0%) History of hepatic decompensation 24 (16.8%) Form of prior decompensation (if yes to above) Ascites 16 (66.7%) Variceal bleed 6 (25.0%) Hepatic encephalopathy 7 (29.2%) ALD, alcohol-related liver disease; ALT, alanine transaminase; AST, aspartate aminotransferase; IQR, inter-quartile range; INR, international normalized ratio; MASLD, metabolic dysfunction-associated steatotic liver disease. IQR is represented as values Q1 to Q3. At the time of HVPG measurement, 49 patients (34.3%) were receiving non-selective beta-blocker therapy. Indications for HVPG measurement were heterogeneous and included evaluation for cirrhosis or fibrosis (39.9%), transplant evaluation (25.9%), elevated liver enzymes (28.0%), and cirrhosis of unclear etiology (19.6%), reflecting selective clinical use rather than routine assessment. Endoscopic or imaging evidence of portal hypertension complications was present in a subset of patients, including non-bleeding esophageal varices in 23.8% and portal hypertensive gastropathy in 10.5%. A history of prior hepatic decompensation was present in 24 patients (16.8%), most commonly ascites (66.7%) and variceal bleeding (25.0%). Model discrimination and overall accuracy Performance of model results is included in Table 2 and Fig. 1 . For HVPG ≥ 10 mmHg, the Vienna model achieved AUROC 0.71 (95% CI 0.61–0.80) and Brier score 0.32 (95% CI 0.27–0.37). The FIB4 + model achieved AUROC 0.69 (95% CI 0.60–0.79) and Brier score 0.24 (95% CI 0.20–0.28). The percent difference in AUC was 1.7 percent in favor of Vienna. Table 2 Summary of predictive performance for Models M1 vs. M2. AUROC (95% CI) Brier Score (95% CI), HVPG ≥ 10 mmHg AUROC (95% CI) Brier Score (95% CI), HVPG ≥ 16 mmHg Vienna/Machine Learning Model (M1) 0.71 (0.61, 0.80) 0.32 (0.27, 0.37) 0.77 (0.66, 0.88) 0.18 (0.14, 0.22) Fib4 + Model (M2) 0.69 (0.60, 0.79) 0.24 (0.20, 0.28) 0.71 (0.59, 0.83) 0.28 (0.24, 0.33) Percent difference in AUC (M1 vs. M2, %) 1.7% 7.7% AUROC, area under the receiver-operator characteristic curve. Table 3 Operating Characteristics of Vienna and FIB-4 + Models Model Threshold (Youden) Sensitivity (95% CI) Specificity (95% CI) PPV (95% CI) NPV (95% CI) Vienna ≥ 10 0.75 0.62 (0.46, 0.76) 0.73 (0.63, 0.82) 0.53 (0.39, 0.66) 0.80 (0.70, 0.87) Vienna ≥ 16 0.46 0.69 (0.41, 0.89) 0.81 (0.73, 0.88) 0.31 (0.17, 0.49) 0.95 (0.90, 0.99) FIB-4 + ≥ 10 0.68 0.57 (0.42, 0.72) 0.83 (0.74, 0.90) 0.63 (0.47, 0.77) 0.80 (0.71, 0.87) FIB-4 + ≥ 16 0.47 0.81 (0.54, 0.96) 0.56 (0.47, 0.65) 0.19 (0.10, 0.30) 0.96 (0.89, 0.99) For HVPG ≥ 16 mmHg, the Vienna model achieved AUROC 0.77 (95% CI 0.66–0.88) and Brier score 0.18 (95% CI 0.14–0.22). The FIB4 + model achieved AUROC 0.71 (95% CI 0.59–0.83) and Brier score 0.28 (95% CI 0.24–0.33). The percent difference in AUC was 7.7 percent in favor of Vienna. Summary metrics mirrored these findings. For Vienna ≥ 10: AUROC 0.705 (95 percent CI 0.611 to 0.798), raw Brier 0.322, null Brier 0.221, and scaled Brier − 0.458. For Vienna ≥ 16: AUROC 0.767 (95 percent CI 0.656 to 0.878), raw Brier 0.177, null Brier 0.099, and scaled Brier − 0.783. For FIB4 + ≥ 10: AUROC 0.693 (95 percent CI 0.601 to 0.786), raw Brier 0.242, null Brier 0.221, and scaled Brier − 0.095. For FIB4 + ≥ 16: AUROC 0.708 (95 percent CI 0.590 to 0.825), raw Brier 0.279, null Brier 0.099, and scaled Brier − 1.81. Negative scaled Brier values indicate performance worse than a noninformative model under this scaling. Calibration All models systematically overestimated risk. For Vienna ≥ 10, the calibration intercept was − 1.13 and slope 0.25, with the nonparametric calibration curve showing observed risks consistently below predicted across the range. For Vienna ≥ 16, the calibration intercept was − 1.96 and slope 0.36, again indicating overprediction with overly extreme probabilities. For FIB4 + ≥ 10, the calibration intercept was − 0.81 and slope 0.21, and for FIB4 + ≥ 16 the intercept was − 2.20 and slope 0.17. Calibration curves confirmed substantial miscalibration. The FIB4 + ≥ 16 model showed the largest calibration error, with Eavg 0.39 and Emax 0.79. Subgroup analyses by liver disease etiology Among patients with nonviral disease (n = 121), Vienna ≥ 10 had AUROC 0.691 (95% CI 0.59–0.792) and Brier 0.331, while Vienna ≥ 16 had AUROC 0.728 (95% CI 0.609–0.848) and Brier 0.199. FIB4 + ≥ 10 had AUROC 0.696 (95% CI 0.596–0.795) and Brier 0.242, and FIB4 + ≥ 16 had AUROC 0.693 (95% CI 0.567–0.819) and Brier 0.286 (Fig. 2 ) Among patients with viral disease (n = 22), Vienna ≥ 10 had AUROC 0.776 (95% CI 0.441-1.00) and Brier 0.270, while FIB4 + ≥ 10 had AUROC 0.671 (95% CI 0.361–0.980) and Brier 0.240. Estimates in viral disease were imprecise given the small sample (Fig. 3 ). Associations between individual laboratory parameters and HVPG thresholds Unadjusted logistic regressions showed graded associations between several laboratory parameters and higher portal pressure categories. Relative to the lowest platelet quartile reference, higher platelet quartiles were associated with lower odds of HVPG ≥ 10, for example Q4 OR 0.28 (95% CI 0.10–0.78). For HVPG ≥ 16, Q4 platelets had OR 0.09 (95% CI 0.01–0.77). Per unit platelet effects were small on the continuous scale. Higher bilirubin was strongly associated with HVPG ≥ 10, with Q4 OR 4.30 (95% CI 1.59–11.64) and per 1 mg/dL increase OR 1.15 (95% CI 1.05–1.27). For HVPG ≥ 16, Q4 bilirubin had OR 17.27 (95% CI 2.07-144.11). Higher INR was associated with both thresholds. For HVPG ≥ 10, Q4 INR had OR 3.29 (95% CI 1.21–8.97). For HVPG ≥ 16, Q3 and Q4 INR had OR 10.50 (95% CI 1.21–90.85) and 9.00 (95% CI 1.05–77.46), respectively. Transaminase levels showed weaker and less consistent associations. For HVPG ≥ 10, AST Q3 and Q4 had OR 3.10 (95% CI 1.12–8.58) and 3.66 (95% CI 1.32–10.16), and ALT Q2 to Q4 were each associated with higher odds with OR 2.92 to 3.88. For HVPG ≥ 16, AST Q3 had OR 14.61 (95% CI 1.76-121.17) with wide uncertainty, and ALT Q3 had OR 6.06 (95% CI 1.21–30.43). Higher FIB-4 was associated with HVPG ≥ 10 in the highest quartile, OR 8.46 (95% CI 2.78–25.75), with a per unit OR of 1.09 (95% CI 0.99–1.19). Associations with HVPG ≥ 16 were smaller and imprecise. Discussion In this external validation study of two non-invasive predictive models for portal hypertension severity, we found that both the Vienna and FIB4 + models demonstrated only modest discrimination for HVPG ≥ 10 mmHg and ≥ 16 mmHg, with AUROCs ranging from 0.69–0.77. Importantly, both models systematically overestimated risk across thresholds, as reflected by negative calibration intercepts and shallow calibration slopes. Among the two, the Vienna model for HVPG ≥ 16 mmHg performed best overall, achieving the highest AUROC (0.77) and lowest Brier score (0.18), though calibration remained poor. These findings underscore the challenges of generalizing machine-learning models for portal hypertension beyond their derivation cohorts. Our results contrast with the robust discrimination and calibration initially reported for both models in European cohorts [ 4 , 5 ]. The Vienna model was originally shown to predict CSPH and severe portal hypertension with AUROCs > 0.80, while the FIB4 + model demonstrated strong performance as an alternative to elastography-based methods. The attenuation of predictive accuracy in our cohort likely reflects differences in patient population and disease context. Specifically, the prevalence of CSPH and severe portal hypertension in our UNC cohort (33% and 11%, respectively) was considerably lower than in the derivation cohorts, leading to systematic risk overestimation. This “spectrum effect” is well recognized in risk prediction and emphasizes the need for calibration when applying non-invasive models to new populations [ 6 ]. Despite modest discrimination, both models maintained high negative predictive values, particularly at the HVPG ≥ 16 mmHg threshold (Vienna model NPV 0.95, FIB4 + NPV 0.96). These findings suggest that the models may retain clinical utility as rule-out tools, helping to identify patients at low likelihood of severe portal hypertension who may be spared invasive HVPG measurement. However, their poor calibration diminishes confidence in absolute risk estimates and limits their use for individualized risk stratification or clinical decision-making without further adjustment. These findings indicate that recalibration is required before application of these models in U.S. or MASLD-predominant populations. The subgroup analyses also suggested variability in model performance by liver disease etiology. While limited by small sample sizes, discrimination appeared higher in viral compared to non-viral etiologies, consistent with prior studies showing that underlying disease mechanism influences the relationship between laboratory parameters and portal pressure. This heterogeneity reinforces the importance of validating predictive tools across diverse patient populations, especially as MASLD becomes the leading indication for liver transplantation in the United States. Model performance was notably weaker in non-viral etiologies, particularly MASLD. In MASLD-related cACLD, portal hypertension may include a substantial presinusoidal component, which is incompletely reflected by HVPG measurements [ 7 ]. As a result, HVPG-based thresholds may underestimate portal hypertension severity in this population, potentially contributing to apparent miscalibration and attenuated discrimination. This limitation highlights the complexity of applying HVPG-derived prediction models to metabolically driven liver disease and underscores the need for etiology-specific validation. Our exploratory analyses of individual laboratory predictors highlighted expected associations between low platelet count, elevated bilirubin, and elevated INR with higher portal pressures, aligning with the biological underpinnings of portal hypertension. These results support the continued centrality of synthetic function markers and platelet count in non-invasive prediction. This study has several strengths. It represents the first independent, US-based external validation of the Vienna and FIB4 + models, providing important insight into their generalizability beyond European derivation cohorts. The cohort used in this analysis was well-characterized, with rigorously measured HVPG values obtained by experienced interventional radiologists following standardized protocols. This ensured high internal validity and accurate phenotyping of portal pressure severity In addition, this study evaluated both discrimination and calibration: two complementary aspects of model performance that are often not assessed together in external validations. The inclusion of calibration analysis, Brier scoring, and subgroup stratification offers a comprehensive assessment of each model’s clinical utility and robustness. We also examined model behavior across clinically relevant HVPG thresholds (≥ 10 mmHg and ≥ 16 mmHg), capturing the prognostic spectrum of compensated liver disease and severe portal hypertension. Finally, our cohort reflects the evolving epidemiology of liver disease in the United States, with a large representation of patients with MASLD and mixed etiology cACLD. This provides an early benchmark for how laboratory-based predictive models derived from European viral-predominant populations perform in a contemporary US setting. Together, these features position this study as an important step toward refining and contextualizing noninvasive risk prediction tools for broader clinical application in diverse patient populations. However, limitations must be acknowledged. First, the retrospective single-center design introduces potential for selection bias, as HVPG was obtained primarily in patients undergoing transplant evaluation or with strong clinical suspicion for portal hypertension. Despite this, the prevalence of CSPH in our cohort was lower than that reported in other validation studies, likely reflecting differing referral patterns and inclusion of patients with compensated disease who underwent HVPG for transplant workup rather than clear decompensation. This limits generalizability to broader cACLD populations. Second, our cohort size was modest (n = 143), reducing power for subgroup analyses and precision of performance estimates. Third, transient elastography data were largely unavailable, preventing head-to-head comparison of these models with image-based modalities that are increasingly standard in clinical practice. Finally, the relatively low prevalence of CSPH and severe portal hypertension in our sample may have inflated low calibration. Given that HVPG is typically reserved for patients with a high clinical suspicion of CSPH, we acknowledge that this cohort likely represents a population with an elevated pre-test probability of portal hypertension. As such, there is potential for selection bias, and findings may not fully generalize to broader populations in which clinical prediction scores would be applied more routinely. This limitation is important when interpreting model performance and planning prospective validation in lower-risk populations. Our findings highlight the need for recalibration of existing non-invasive models when applied to US-based or MASLD-predominant populations. Prospective multicenter studies with larger and more diverse cohorts will be critical to refine and validate predictive models for portal hypertension. Integration of novel biomarkers, imaging data, and advanced machine learning techniques may further enhance predictive accuracy. Ultimately, achieving accurate, widely available non-invasive risk stratification tools is essential to reducing reliance on HVPG and addressing disparities in portal hypertension care. Declarations Ethical Considerations This study was conducted in compliance with ethical standards and was approved by the Institutional Review Board (IRB) at the University of North Carolina at Chapel Hill (IRB #24-1162). Given the retrospective nature of the study and the use of de-identified clinical data, a waiver of informed consent was granted Author Contribution FM, manuscript writing and critical review, decision to publish; MX, statistical analyses and data interpretation; MM, data collection and extraction; AC, data collection and extraction; JF, data collection and extraction; TT, data collection and extraction; AJ, data collection and extraction; NK, manuscript writing and critical review; HY, manuscript writing and critical review; CA, data analysis and critical review; AB, senior supervision and critical review; AM, manuscript writing and critical review, statistical analysis, and decision to publish. References Mandorfer M, Aigner E, Cejna M, et al. Austrian consensus on the diagnosis and management of portal hypertension in advanced chronic liver disease (Billroth IV). Wiener Klinische Wochenschrift . 2023;135(S3):493-523. doi:10.1007/s00508-023-02229-w Mandorfer M, Kozbial K, Schwabl P, et al. Changes in hepatic venous pressure gradient predict hepatic decompensation in patients who achieved sustained virologic response to Interferon‐Free therapy. Hepatology . 2019;71(3):1023-1036. doi:10.1002/hep.30885 De Franchis R, Bosch J, Garcia-Tsao G, et al. Baveno VII – Renewing consensus in portal hypertension. Journal of Hepatology . 2021;76(4):959-974. doi:10.1016/j.jhep.2021.12.022 Reiniš J, Petrenko O, Simbrunner B, et al. Assessment of portal hypertension severity using machine learning models in patients with compensated cirrhosis. Journal of Hepatology . 2022;78(2):390-400. doi:10.1016/j.jhep.2022.09.012 Rabiee A, Deng Y, Ciarleglio M, et al. Noninvasive predictors of clinically significant portal hypertension in NASH cirrhosis: Validation of ANTICIPATE models and development of a lab‐based model. Hepatology Communications . 2022;6(12):3324-3334. doi:10.1002/hep4.2091 Usher-Smith JA, Sharp, SJ, Griffin SJ. The spectrum effect in tests for risk prediction, screening, and diagnosis. BMJ . 2016; 353. doi:10.1136/bmj.i3139 Madir A, Arienzo A, Helmy A, et al. Portal hypertension in patients with nonalcoholic fatty liver disease: Current knowledge and challenges. World Journal of Gastroenterology . 2024;30(4):290-307. doi:10.3748/wjg.v30.i4.290 Additional Declarations No competing interests reported. Supplementary Files Supplementaryfigure.docx Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: Revision requested 08 Apr, 2026 Reviews received at journal 08 Apr, 2026 Reviews received at journal 16 Mar, 2026 Reviewers agreed at journal 16 Mar, 2026 Reviewers agreed at journal 16 Mar, 2026 Reviewers agreed at journal 16 Feb, 2026 Reviewers invited by journal 11 Feb, 2026 Editor assigned by journal 11 Feb, 2026 Submission checks completed at journal 11 Feb, 2026 First submitted to journal 10 Feb, 2026 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8845244","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":592243724,"identity":"f20f3028-e1ab-49d1-bd2b-dca67f6fa7f0","order_by":0,"name":"Fahad Mohammed","email":"","orcid":"","institution":"University of North Carolina School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Fahad","middleName":"","lastName":"Mohammed","suffix":""},{"id":592243725,"identity":"3082166c-ded5-493d-b374-18ad7ce939e7","order_by":1,"name":"Minzhi Xing","email":"","orcid":"","institution":"University of North Carolina at Chapel Hill","correspondingAuthor":false,"prefix":"","firstName":"Minzhi","middleName":"","lastName":"Xing","suffix":""},{"id":592243726,"identity":"e9a518ad-ebcf-4449-aba5-6623a1c5ad36","order_by":2,"name":"Michael Mohnasky","email":"","orcid":"","institution":"University of North Carolina School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Michael","middleName":"","lastName":"Mohnasky","suffix":""},{"id":592243727,"identity":"87fd3c41-2e37-4a6e-90fd-a8812e3f1ef4","order_by":3,"name":"Andrew Caddell","email":"","orcid":"","institution":"University of North Carolina School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Andrew","middleName":"","lastName":"Caddell","suffix":""},{"id":592243728,"identity":"d5ce283e-b85c-41f3-9428-a62f712ca9d3","order_by":4,"name":"Jack Felkner","email":"","orcid":"","institution":"University of North Carolina School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Jack","middleName":"","lastName":"Felkner","suffix":""},{"id":592243729,"identity":"2e0e80b4-0ddf-4d3d-8026-152d2d19fd17","order_by":5,"name":"Thomas Turner","email":"","orcid":"","institution":"University of North Carolina School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Thomas","middleName":"","lastName":"Turner","suffix":""},{"id":592243730,"identity":"c43d8e94-6d63-4b33-9a54-a72b3c3a80c1","order_by":6,"name":"Arjun Juneja","email":"","orcid":"","institution":"University of North Carolina School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Arjun","middleName":"","lastName":"Juneja","suffix":""},{"id":592243731,"identity":"48400f65-e127-405d-b48e-386f48015ad3","order_by":7,"name":"Nima Kokabi","email":"","orcid":"","institution":"University of North Carolina School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Nima","middleName":"","lastName":"Kokabi","suffix":""},{"id":592243732,"identity":"c937848f-03a9-4f53-9100-5c773408663c","order_by":8,"name":"Hyeon Yu","email":"","orcid":"","institution":"University of North Carolina School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Hyeon","middleName":"","lastName":"Yu","suffix":""},{"id":592243733,"identity":"91aa6a2d-595c-49d7-a68c-e11014454ebc","order_by":9,"name":"Chelsea Anderson","email":"","orcid":"","institution":"University of North Carolina School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Chelsea","middleName":"","lastName":"Anderson","suffix":""},{"id":592243734,"identity":"64956a32-e6e8-4ec2-aaa8-040f7703e9b2","order_by":10,"name":"A. Sidney Barritt","email":"","orcid":"","institution":"University of North Carolina at Chapel Hill","correspondingAuthor":false,"prefix":"","firstName":"A.","middleName":"Sidney","lastName":"Barritt","suffix":""},{"id":592243735,"identity":"46deba8a-ac45-4d46-bf1f-39fa6f15fe8b","order_by":11,"name":"Andrew M. Moon","email":"data:image/png;base64,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","orcid":"","institution":"University of North Carolina at Chapel Hill","correspondingAuthor":true,"prefix":"","firstName":"Andrew","middleName":"M.","lastName":"Moon","suffix":""}],"badges":[],"createdAt":"2026-02-10 21:08:25","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8845244/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8845244/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":102892513,"identity":"8b5ef1e1-c603-44ad-ab02-15ec6497a44a","added_by":"auto","created_at":"2026-02-18 05:25:37","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":28701,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eReceiver Operating Characteristic Curves for patients overall with liver disease (\u003c/strong\u003eA) Vienna≥10, (B) Vienna≥16, (C) FIB4+≥10, (D) FIB4+≥16.\u003c/p\u003e","description":"","filename":"Binder21.png","url":"https://assets-eu.researchsquare.com/files/rs-8845244/v1/403ea92a2bf7b0bbc8e960bf.png"},{"id":102892447,"identity":"4f462549-7a07-4048-b3a4-74b4314d27f0","added_by":"auto","created_at":"2026-02-18 05:25:30","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":28634,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eReceiver Operating Characteristic Curves for patients with nonviral liver disease.\u003c/strong\u003e (A)Vienna≥10, (B) Vienna≥16, (C) FIB4+≥10, (D) FIB4+≥16.\u003c/p\u003e","description":"","filename":"Binder22.png","url":"https://assets-eu.researchsquare.com/files/rs-8845244/v1/8ceb3bfc0e517a0680d837bb.png"},{"id":102892467,"identity":"7e14d990-b4bd-48c4-ae82-78499b2b831e","added_by":"auto","created_at":"2026-02-18 05:25:34","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":15799,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eReceiver Operating Characteristic Curves for patients with viral liver disease.\u003c/strong\u003e (A)Vienna≥10, (B) FIB4+≥10.\u003c/p\u003e","description":"","filename":"Binder23.png","url":"https://assets-eu.researchsquare.com/files/rs-8845244/v1/1e1a1ca2f1210b0b3c2e394b.png"},{"id":102892536,"identity":"752d76ed-b555-4060-be1c-37c4b621641c","added_by":"auto","created_at":"2026-02-18 05:25:42","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1031851,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8845244/v1/ea1f83ea-39ad-48f6-ab53-d6225ec7a49d.pdf"},{"id":102892438,"identity":"4af88cc4-c085-4359-97ae-031c4cfe63a7","added_by":"auto","created_at":"2026-02-18 05:25:26","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":201717,"visible":true,"origin":"","legend":"","description":"","filename":"Supplementaryfigure.docx","url":"https://assets-eu.researchsquare.com/files/rs-8845244/v1/2d499fc1b729a67036cdca24.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"External Validation of Laboratory-Based Models to Assess Portal Hypertension Severity in Patients with Compensated Chronic Liver Disease","fulltext":[{"header":"Introduction","content":"\u003cp\u003ePortal hypertension is a major driver of disease progression in patients with compensated advanced chronic liver disease (cACLD) and significantly impacts clinical outcomes. The hepatic venous pressure gradient (HVPG) is the gold standard for assessing the severity of portal hypertension and serves as a crucial prognostic marker in chronic liver disease. An HVPG of \u0026ge;\u0026thinsp;10 mmHg defines clinically significant portal hypertension (CSPH), which is associated with an increased risk of hepatic decompensation, including the development of ascites, variceal bleeding, and hepatic encephalopathy [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. Furthermore, an HVPG of \u0026ge;\u0026thinsp;16 mmHg is predictive of more severe complications including variceal bleeding and refractory ascites and is associated with increased mortality risk [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. Despite its utility, HVPG measurement is an invasive procedure requiring specialized expertise and is not widely available in many clinical settings.\u003c/p\u003e \u003cp\u003eConsequently, there is a need for reliable, non-invasive methods to accurately assess portal hypertension severity and facilitate risk stratification. Several predictive models have been developed to estimate portal hypertension severity using widely available laboratory parameters. Image-based fibrosis assessment (e.g. transient elastography) has substantial data for predicting decompensation risk and has been adopted by the Baveno VII guidelines [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. However, transient elastography is not available in all settings, and reliance on image-based tools may worsen disparities in liver disease care. Machine learning algorithms have increasingly been utilized to develop non-invasive models that predict CSPH and severe portal hypertension with high accuracy. One such machine learning model, the Vienna Model, was designed to predict HVPG\u0026thinsp;\u0026ge;\u0026thinsp;10 mmHg and \u0026ge;\u0026thinsp;16 mmHg using three laboratory parameters: platelet count, total bilirubin, and international normalized ratio (INR) [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. This model was developed and validated in a European cohort, demonstrating robust predictive performance. Another non-invasive model, the FIB4\u0026thinsp;+\u0026thinsp;model, incorporates the Fibrosis-4 (FIB-4) index, which includes age, aspartate aminotransferase (AST), alanine aminotransferase (ALT), and platelet count, along with serum albumin, to predict CSPH. The FIB4\u0026thinsp;+\u0026thinsp;model was proposed as an alternative to liver stiffness measurement (LSM) and transient elastography, which, although widely used, are not always accessible in routine clinical practice [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eWhile these models have shown promising results in their initial validation studies, their performance has not been extensively evaluated in independent, real-world patient populations. Machine learning models are at particular risk of overfitting and demonstrating poor external validity. External validation is essential to determine whether these models can be generalized across different patient cohorts, particularly in populations with different demographic characteristics, liver disease etiologies, and clinical practices. Prior studies have demonstrated that the performance of non-invasive predictive models may vary depending on factors such as the prevalence of CSPH, obesity, and liver disease etiology. Therefore, external validation in a US-based cohort is critical to assess the generalizability and clinical applicability of these models.\u003c/p\u003e \u003cp\u003eHowever, transportability of non-invasive portal hypertension models across populations remains uncertain. Differences in disease etiology, portal hypertension prevalence, and referral patterns may substantially affect model performance and calibration. In particular, U.S. cohorts are increasingly dominated by metabolic dysfunction\u0026ndash;associated steatotic liver disease (MASLD), in which portal hypertension may include a presinusoidal component that is incompletely captured by HVPG. External validation in contemporary U.S. populations is therefore critical, not only to assess discrimination but also to evaluate calibration and clinical reliability.\u003c/p\u003e \u003cp\u003eThe primary objective of this study was to externally validate two laboratory-based models, the Vienna Model and the FIB4\u0026thinsp;+\u0026thinsp;Model, for assessing portal hypertension severity in a cohort of patients with cACLD who underwent HVPG measurement at UNC Health. Specifically, we aimed to evaluate the ability of these models to predict HVPG\u0026thinsp;\u0026ge;\u0026thinsp;10 mmHg and \u0026ge;\u0026thinsp;16 mmHg using non-invasive laboratory parameters. Model performance was assessed in terms of discrimination, calibration, and overall predictive accuracy. Additionally, we conducted subgroup analyses to determine whether the predictive performance of these models varied by demographic and clinical characteristics, including age, sex, and etiology of liver disease. By validating these models in an independent cohort, this study seeks to provide insight into their clinical utility and potential role in guiding non-invasive risk stratification for patients with cACLD.\u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eStudy Design and Patient Selection\u003c/h2\u003e \u003cp\u003eThis was a retrospective cohort study conducted at UNC Health to externally validate two non-invasive models for assessing portal hypertension severity in patients with compensated advanced chronic liver disease (cACLD). The study included patients who underwent percutaneous transjugular hepatic venous pressure gradient (HVPG) measurement between May 2014 and April 2024. HVPG measurements were performed via a transjugular approach using a balloon-occlusion catheter, with wedged hepatic venous pressure obtained following balloon inflation to achieve complete hepatic vein occlusion and free hepatic venous pressure measured in the same vein; HVPG was calculated as the difference between wedged and free pressures. Eligible patients were identified through the electronic medical record system, and data were extracted from clinical notes, procedural reports, and laboratory results. Transient elastography data were only available for a small minority of patients and were therefore not included in the analysis.\u003c/p\u003e \u003cp\u003eTo ensure a clinically relevant cohort, inclusion criteria required that patients had a diagnosis of cACLD and underwent HVPG measurement as part of their clinical evaluation. cACLD was defined by histologic evidence of F3\u0026ndash;F4 fibrosis in the absence of active hepatic decompensation at the time of HVPG measurement. Patients were excluded if they had current hepatic decompensation, defined as the presence of ascites, gastroesophageal variceal bleeding, or hepatic encephalopathy at the time of HVPG measurement. Patients with active hepatic decompensation at the time of HVPG measurement were excluded; patients with prior remote decompensation without active manifestations were not excluded. Patients with current or prior hepatic malignancy, as well as those with a history of transjugular intrahepatic portosystemic shunt (TIPS), were also excluded. Additionally, patients were required to have available laboratory data, including platelet count, bilirubin, international normalized ratio (INR), AST, ALT, and albumin, within 30 days of HVPG measurement.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eData Collection and Variables\u003c/h3\u003e\n\u003cp\u003ePatient demographic information, including age, sex, race, and etiology of liver disease, was extracted from the electronic health record system. Clinical data included the presence of metabolic dysfunction-associated steatotic liver disease (MASLD), alcohol-related liver disease (ALD), viral hepatitis, autoimmune hepatitis, or other chronic liver diseases. Indications for the patient\u0026rsquo;s biopsy were also recorded. Laboratory parameters collected included platelet count (\u0026times;10⁹/L), total bilirubin (mg/dL), INR, AST (U/L), ALT (U/L), and albumin (g/dL), as these were integral to the predictive models being validated.\u003c/p\u003e \u003cp\u003eHVPG values were obtained from interventional radiology procedural reports, with 11 different radiologists performing the procedures. HVPG was measured using a transjugular catheter technique, in which the free hepatic venous pressure (FHVP) and wedged hepatic venous pressure (WHVP) were recorded. HVPG was calculated as the difference between WHVP and FHVP. Portal hypertension severity was categorized as clinically significant portal hypertension (CSPH) if HVPG was \u0026ge;\u0026thinsp;10 mmHg and as severe portal hypertension if HVPG was \u0026ge;\u0026thinsp;16 mmHg.\u003c/p\u003e \u003cp\u003eHVPG measurements were obtained selectively, most commonly as part of liver transplant evaluation, diagnostic clarification of portal hypertension severity, or enrollment in clinical research protocols. HVPG was not performed routinely in all patients with cACLD, and the study cohort therefore represents a selected subset of patients evaluated at our center during the study period. Use of non-selective beta-blockers at the time of HVPG measurement was abstracted from the medical record where available.\u003c/p\u003e\n\u003ch3\u003ePredictive Models for External Validation\u003c/h3\u003e\n\u003cp\u003eTwo predictive models for assessing portal hypertension severity were externally validated:\u003c/p\u003e \u003cp\u003e1. \u003cb\u003eVienna Model\u003c/b\u003e \u0026ndash; This model was originally developed using a European cohort and utilizes three laboratory parameters: platelet count, total bilirubin, and INR [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. It was designed to predict HVPG\u0026thinsp;\u0026ge;\u0026thinsp;10 mmHg and \u0026ge;\u0026thinsp;16 mmHg based on a machine-learning approach that optimizes classification performance. The formula for the prediction of severe PH is:\u003c/p\u003e \u003cp\u003e \u003cspan class=\"InlineEquation\"\u003e \u003cspan class=\"mathinline\"\u003e\\(\\:{\\text{H}\\text{V}\\text{P}\\text{G}}_{\\text{p}\\text{r}\\text{o}\\text{b}\\text{a}\\text{b}\\text{i}\\text{l}\\text{i}\\text{t}\\text{y}}={\\sigma\\:}({{\\beta\\:}}_{0}+{{\\beta\\:}}_{1}\\times\\:\\text{p}\\text{l}\\text{a}\\text{t}\\text{e}\\text{l}\\text{e}\\text{t}\\:\\text{c}\\text{o}\\text{u}\\text{n}\\text{t}+{{\\beta\\:}}_{2}\\times\\:\\text{s}\\text{e}\\text{r}\\text{u}\\text{m}\\:\\text{b}\\text{i}\\text{l}\\text{i}\\text{r}\\text{u}\\text{b}\\text{i}\\text{n}+{{\\beta\\:}}_{3}\\times\\:\\text{I}\\text{N}\\text{R}\\)\u003c/span\u003e \u003c/span\u003e)\u003c/p\u003e \u003cp\u003ewhere σ is a sigmoid function calculated as σ(x)=\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\frac{1}{1+{e}^{-x}}\\)\u003c/span\u003e\u003c/span\u003e, e=Euler\u0026rsquo;s number, platelet count units are 10\u003csup\u003e9\u003c/sup\u003e/L, and serum bilirubin units are mg/dl. The extracted coefficients β\u003csub\u003en\u003c/sub\u003e both for HVPG \u0026gt;\u0026thinsp;10 and \u0026gt;\u0026thinsp;16 mmHg predictions are available in the supplementary materials.\u003c/p\u003e \u003cp\u003e2. \u003cb\u003eFIB4\u0026thinsp;+\u0026thinsp;Model\u003c/b\u003e \u0026ndash; This model incorporates the Fibrosis-4 (FIB4) index, which is calculated using age, AST, ALT, and platelet count, along with serum albumin [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. The model was developed as an alternative to elastography-based methods for predicting CSPH in patients with compensated liver disease. The formula for the final calibrated model is:\u003cdiv id=\"Equa\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equa\" name=\"EquationSource\"\u003e\n$$\\:\\text{L}\\text{o}\\text{g}\\:\\text{o}\\text{d}\\text{d}\\text{s}\\left(\\text{C}\\text{S}\\text{P}\\text{H}\\right)=0.7207-\\left(0.6729\\times\\:\\text{a}\\text{l}\\text{b}\\text{u}\\text{m}\\text{i}\\text{n}\\right)+(0.4408\\times\\:\\text{F}\\text{I}\\text{B}4)$$\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003ewhere FIB4=\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\frac{\\text{A}\\text{g}\\text{e}\\times\\:\\text{A}\\text{S}\\text{T}}{\\text{P}\\text{l}\\text{a}\\text{t}\\text{e}\\text{l}\\text{e}\\text{t}\\text{s}\\times\\:\\surd\\:\\text{A}\\text{L}\\text{T}}\\)\u003c/span\u003e\u003c/span\u003e, age units are years, AST/ALT units are U/L, platelets units are10\u003csup\u003e9\u003c/sup\u003e/L, and albumin units are g/ds.\u003c/p\u003e \u003cp\u003eBoth models were applied to the UNC Health cohort to assess their predictive performance in an independent, real-world US population.\u003c/p\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003eStatistical Analysis\u003c/h2\u003e \u003cp\u003eDescriptive statistics were used to summarize baseline patient characteristics, with continuous variables reported as median with interquartile range (IQR) and categorical variables presented as frequencies and percentages.\u003c/p\u003e \u003cp\u003eThe predictive performance of the Vienna Model and FIB4\u0026thinsp;+\u0026thinsp;Model was assessed by calculating the area under the receiver operating characteristic (AUROC) curve for HVPG thresholds of \u0026ge;\u0026thinsp;10 mmHg and \u0026ge;\u0026thinsp;16 mmHg. Confidence intervals were estimated using bootstrap resampling with 1000 repetitions. AUROC values between 0.7 and 0.8 were considered indicative of acceptable discrimination, while values above 0.8 were considered excellent.\u003c/p\u003e \u003cp\u003eTo evaluate model calibration, calibration plots were generated comparing predicted vs. observed probabilities of HVPG\u0026thinsp;\u0026ge;\u0026thinsp;10 mmHg and \u0026ge;\u0026thinsp;16 mmHg. We fit logistic recalibration models regressing the observed outcome on the logit of the predicted probability to estimate the calibration intercept (ideal value 0) and calibration slope (ideal value 1). These parameters describe whether the model systematically over- or underestimates risk and whether predictions are overly extreme or too narrow. Brier scores were computed (with bootstrap 95% CIs) to assess the overall accuracy of probability estimates, with lower scores indicating better calibration. Additionally, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV), with exact binomial CIs, were calculated for both models at optimal cutoff thresholds determined by the Youden index. All analyses were conducted in RStudio (version 2024.12.1) using the pROC, rms, boot, and binom packages.\u003c/p\u003e \u003cp\u003eSubgroup analyses were performed to assess model performance across different patient populations, including stratification by age, sex, race, and etiology of liver disease. Logistic regression models were used to examine the association between individual laboratory parameters and HVPG thresholds, controlling for potential confounders such as age and sex.\u003c/p\u003e \u003c/div\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eCohort characteristics\u003c/h2\u003e \u003cp\u003eWe identified 143 adults with compensated chronic liver disease who underwent HVPG measurement. Baseline characteristics are included in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. Median age was 56 years (IQR 50\u0026ndash;64) and 53.9% were female. Most patients had a liver disease etiology of MASLD 31.5%, ALD 16.8%, or viral hepatitis 13.3%. Median HVPG was 7 mmHg (IQR 4\u0026ndash;11). The prevalence of HVPG\u0026thinsp;\u0026ge;\u0026thinsp;10 mmHg was 32.9% and of HVPG\u0026thinsp;\u0026ge;\u0026thinsp;16 mmHg was 11.2%.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003ePatient Characteristics\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"2\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVariable\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTotal\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePatients, n\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e143\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAge, years (median, IQR)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e56 (50\u0026ndash;64)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSex, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFemale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e77 (53.9%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e66 (46.2%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRace, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAmerican Indian or Alaska Native\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3 (2.1%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAsian\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1 (0.7%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBlack\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e44 (31.0%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWhite\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e83 (58.5%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOther\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e11 (7.8%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eEthnicity, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHispanic/Latino\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e9 (6.3%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNon-Hispanic\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e134 (93.7%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eChronic Liver Disease Etiology\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eALD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e24 (16.8%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eViral Hepatitis\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e19 (13.3%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eALD\u0026thinsp;+\u0026thinsp;Viral Hepatitis\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3 (2.1%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAutoimmune Hepatitis\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e19 (13.3%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMASLD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e45 (31.5%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCryptogenic\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e22 (15.4%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOther\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e24 (16.7%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHVPG, median (IQR)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e7 (4\u0026ndash;11)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePlatelet Count, G/L, median (IQR)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e131 (88\u0026ndash;193)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTotal Bilirubin, mg/dL, median (IQR)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.8 (0.5\u0026ndash;1.7)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eINR, median (IQR)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1.2 (1.1\u0026ndash;1.4)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAST, U/L, median (IQR)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e47 (30\u0026ndash;79)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eALT, U/L, median (IQR)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e40 (24\u0026ndash;69)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAlbumin, g/dL, median (IQR)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3.7 (3.0-4.1)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNon-selective beta blockers\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e49 (34.3%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIndications for HVPG measurement\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eElevated liver enzymes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e40 (28.0%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCirrhosis of unknown etiology\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e28 (19.6%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTransplant evaluation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e37 (25.9%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAcute liver injury of unknown etiology\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e8 (5.6%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eEvaluate for cirrhosis/fibrosis\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e57 (39.9%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGraft-versus-host disease concern\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0 (0.0%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTransplant rejection concern\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2 (1.4%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAscites of unknown etiology\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0 (0.0%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOther\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e9 (6.3%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUpper endoscopy/imaging findings for portal hypertension complications\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGastric varices (non-bleeding)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3 (2.1%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eEsophageal varices (non-bleeding)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e34 (23.8%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRectal varices (non-bleeding)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0 (0.0%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePortal hypertensive gastropathy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e15 (10.5%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGastrorenal shunt\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0 (0.0%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSplenorenal shunt\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2 (1.4%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGastroesophageal varices (non-bleeding)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1 (0.7%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePerisplenic varices (non-bleeding)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e6 (4.2%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePeripancreatic varices (non-bleeding)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0 (0.0%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUnspecified varices (non-bleeding)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0 (0.0%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHistory of hepatic decompensation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e24 (16.8%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eForm of prior decompensation (if yes to above)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAscites\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e16 (66.7%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVariceal bleed\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e6 (25.0%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHepatic encephalopathy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e7 (29.2%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"2\"\u003eALD, alcohol-related liver disease; ALT, alanine transaminase; AST, aspartate aminotransferase; IQR, inter-quartile range; INR, international normalized ratio; MASLD, metabolic dysfunction-associated steatotic liver disease.\u003c/td\u003e\u003c/tr\u003e \u003ctr\u003e\u003ctd colspan=\"2\"\u003eIQR is represented as values Q1 to Q3.\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eAt the time of HVPG measurement, 49 patients (34.3%) were receiving non-selective beta-blocker therapy. Indications for HVPG measurement were heterogeneous and included evaluation for cirrhosis or fibrosis (39.9%), transplant evaluation (25.9%), elevated liver enzymes (28.0%), and cirrhosis of unclear etiology (19.6%), reflecting selective clinical use rather than routine assessment. Endoscopic or imaging evidence of portal hypertension complications was present in a subset of patients, including non-bleeding esophageal varices in 23.8% and portal hypertensive gastropathy in 10.5%. A history of prior hepatic decompensation was present in 24 patients (16.8%), most commonly ascites (66.7%) and variceal bleeding (25.0%).\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eModel discrimination and overall accuracy\u003c/h3\u003e\n\u003cp\u003ePerformance of model results is included in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e and Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. For HVPG\u0026thinsp;\u0026ge;\u0026thinsp;10 mmHg, the Vienna model achieved AUROC 0.71 (95% CI 0.61\u0026ndash;0.80) and Brier score 0.32 (95% CI 0.27\u0026ndash;0.37). The FIB4\u0026thinsp;+\u0026thinsp;model achieved AUROC 0.69 (95% CI 0.60\u0026ndash;0.79) and Brier score 0.24 (95% CI 0.20\u0026ndash;0.28). The percent difference in AUC was 1.7 percent in favor of Vienna.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eSummary of predictive performance for Models M1 vs. M2.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAUROC (95% CI)\u003c/p\u003e \u003cp\u003eBrier Score (95% CI),\u003c/p\u003e \u003cp\u003eHVPG\u0026thinsp;\u0026ge;\u0026thinsp;10 mmHg\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAUROC (95% CI)\u003c/p\u003e \u003cp\u003eBrier Score (95% CI),\u003c/p\u003e \u003cp\u003eHVPG\u0026thinsp;\u0026ge;\u0026thinsp;16 mmHg\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVienna/Machine Learning Model (M1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.71 (0.61, 0.80)\u003c/p\u003e \u003cp\u003e0.32 (0.27, 0.37)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.77 (0.66, 0.88)\u003c/p\u003e \u003cp\u003e0.18 (0.14, 0.22)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFib4\u0026thinsp;+\u0026thinsp;Model (M2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.69 (0.60, 0.79)\u003c/p\u003e \u003cp\u003e0.24 (0.20, 0.28)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.71 (0.59, 0.83)\u003c/p\u003e \u003cp\u003e0.28 (0.24, 0.33)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePercent difference in AUC (M1 vs. M2, %)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1.7%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e7.7%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"3\"\u003eAUROC, area under the receiver-operator characteristic curve.\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eOperating Characteristics of Vienna and FIB-4\u0026thinsp;+\u0026thinsp;Models\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eModel\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eThreshold (Youden)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSensitivity (95% CI)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSpecificity (95% CI)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePPV (95% CI)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eNPV (95% CI)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVienna\u0026thinsp;\u0026ge;\u0026thinsp;10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.75\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.62 (0.46, 0.76)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.73 (0.63, 0.82)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.53 (0.39, 0.66)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.80 (0.70, 0.87)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVienna\u0026thinsp;\u0026ge;\u0026thinsp;16\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.46\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.69 (0.41, 0.89)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.81 (0.73, 0.88)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.31 (0.17, 0.49)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.95 (0.90, 0.99)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFIB-4\u0026thinsp;+\u0026thinsp;\u0026ge;\u0026thinsp;10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.68\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.57 (0.42, 0.72)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.83 (0.74, 0.90)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.63 (0.47, 0.77)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.80 (0.71, 0.87)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFIB-4\u0026thinsp;+\u0026thinsp;\u0026ge;\u0026thinsp;16\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.47\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.81 (0.54, 0.96)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.56 (0.47, 0.65)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.19 (0.10, 0.30)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.96 (0.89, 0.99)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eFor HVPG\u0026thinsp;\u0026ge;\u0026thinsp;16 mmHg, the Vienna model achieved AUROC 0.77 (95% CI 0.66\u0026ndash;0.88) and Brier score 0.18 (95% CI 0.14\u0026ndash;0.22). The FIB4\u0026thinsp;+\u0026thinsp;model achieved AUROC 0.71 (95% CI 0.59\u0026ndash;0.83) and Brier score 0.28 (95% CI 0.24\u0026ndash;0.33). The percent difference in AUC was 7.7 percent in favor of Vienna.\u003c/p\u003e \u003cp\u003eSummary metrics mirrored these findings. For Vienna\u0026thinsp;\u0026ge;\u0026thinsp;10: AUROC 0.705 (95 percent CI 0.611 to 0.798), raw Brier 0.322, null Brier 0.221, and scaled Brier\u0026thinsp;\u0026minus;\u0026thinsp;0.458. For Vienna\u0026thinsp;\u0026ge;\u0026thinsp;16: AUROC 0.767 (95 percent CI 0.656 to 0.878), raw Brier 0.177, null Brier 0.099, and scaled Brier\u0026thinsp;\u0026minus;\u0026thinsp;0.783. For FIB4\u0026thinsp;+\u0026thinsp;\u0026ge;\u0026thinsp;10: AUROC 0.693 (95 percent CI 0.601 to 0.786), raw Brier 0.242, null Brier 0.221, and scaled Brier\u0026thinsp;\u0026minus;\u0026thinsp;0.095. For FIB4\u0026thinsp;+\u0026thinsp;\u0026ge;\u0026thinsp;16: AUROC 0.708 (95 percent CI 0.590 to 0.825), raw Brier 0.279, null Brier 0.099, and scaled Brier\u0026thinsp;\u0026minus;\u0026thinsp;1.81. Negative scaled Brier values indicate performance worse than a noninformative model under this scaling.\u003c/p\u003e\n\u003ch3\u003eCalibration\u003c/h3\u003e\n\u003cp\u003eAll models systematically overestimated risk. For Vienna\u0026thinsp;\u0026ge;\u0026thinsp;10, the calibration intercept was \u0026minus;\u0026thinsp;1.13 and slope 0.25, with the nonparametric calibration curve showing observed risks consistently below predicted across the range. For Vienna\u0026thinsp;\u0026ge;\u0026thinsp;16, the calibration intercept was \u0026minus;\u0026thinsp;1.96 and slope 0.36, again indicating overprediction with overly extreme probabilities. For FIB4\u0026thinsp;+\u0026thinsp;\u0026ge;\u0026thinsp;10, the calibration intercept was \u0026minus;\u0026thinsp;0.81 and slope 0.21, and for FIB4\u0026thinsp;+\u0026thinsp;\u0026ge;\u0026thinsp;16 the intercept was \u0026minus;\u0026thinsp;2.20 and slope 0.17. Calibration curves confirmed substantial miscalibration. The FIB4\u0026thinsp;+\u0026thinsp;\u0026ge;\u0026thinsp;16 model showed the largest calibration error, with Eavg 0.39 and Emax 0.79.\u003c/p\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eSubgroup analyses by liver disease etiology\u003c/h2\u003e \u003cp\u003eAmong patients with nonviral disease (n\u0026thinsp;=\u0026thinsp;121), Vienna\u0026thinsp;\u0026ge;\u0026thinsp;10 had AUROC 0.691 (95% CI 0.59\u0026ndash;0.792) and Brier 0.331, while Vienna\u0026thinsp;\u0026ge;\u0026thinsp;16 had AUROC 0.728 (95% CI 0.609\u0026ndash;0.848) and Brier 0.199. FIB4\u0026thinsp;+\u0026thinsp;\u0026ge;\u0026thinsp;10 had AUROC 0.696 (95% CI 0.596\u0026ndash;0.795) and Brier 0.242, and FIB4\u0026thinsp;+\u0026thinsp;\u0026ge;\u0026thinsp;16 had AUROC 0.693 (95% CI 0.567\u0026ndash;0.819) and Brier 0.286 (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e)\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eAmong patients with viral disease (n\u0026thinsp;=\u0026thinsp;22), Vienna\u0026thinsp;\u0026ge;\u0026thinsp;10 had AUROC 0.776 (95% CI 0.441-1.00) and Brier 0.270, while FIB4\u0026thinsp;+\u0026thinsp;\u0026ge;\u0026thinsp;10 had AUROC 0.671 (95% CI 0.361\u0026ndash;0.980) and Brier 0.240. Estimates in viral disease were imprecise given the small sample (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eAssociations between individual laboratory parameters and HVPG thresholds\u003c/h2\u003e \u003cp\u003eUnadjusted logistic regressions showed graded associations between several laboratory parameters and higher portal pressure categories. Relative to the lowest platelet quartile reference, higher platelet quartiles were associated with lower odds of HVPG\u0026thinsp;\u0026ge;\u0026thinsp;10, for example Q4 OR 0.28 (95% CI 0.10\u0026ndash;0.78). For HVPG\u0026thinsp;\u0026ge;\u0026thinsp;16, Q4 platelets had OR 0.09 (95% CI 0.01\u0026ndash;0.77). Per unit platelet effects were small on the continuous scale.\u003c/p\u003e \u003cp\u003eHigher bilirubin was strongly associated with HVPG\u0026thinsp;\u0026ge;\u0026thinsp;10, with Q4 OR 4.30 (95% CI 1.59\u0026ndash;11.64) and per 1 mg/dL increase OR 1.15 (95% CI 1.05\u0026ndash;1.27). For HVPG\u0026thinsp;\u0026ge;\u0026thinsp;16, Q4 bilirubin had OR 17.27 (95% CI 2.07-144.11).\u003c/p\u003e \u003cp\u003eHigher INR was associated with both thresholds. For HVPG\u0026thinsp;\u0026ge;\u0026thinsp;10, Q4 INR had OR 3.29 (95% CI 1.21\u0026ndash;8.97). For HVPG\u0026thinsp;\u0026ge;\u0026thinsp;16, Q3 and Q4 INR had OR 10.50 (95% CI 1.21\u0026ndash;90.85) and 9.00 (95% CI 1.05\u0026ndash;77.46), respectively.\u003c/p\u003e \u003cp\u003eTransaminase levels showed weaker and less consistent associations. For HVPG\u0026thinsp;\u0026ge;\u0026thinsp;10, AST Q3 and Q4 had OR 3.10 (95% CI 1.12\u0026ndash;8.58) and 3.66 (95% CI 1.32\u0026ndash;10.16), and ALT Q2 to Q4 were each associated with higher odds with OR 2.92 to 3.88. For HVPG\u0026thinsp;\u0026ge;\u0026thinsp;16, AST Q3 had OR 14.61 (95% CI 1.76-121.17) with wide uncertainty, and ALT Q3 had OR 6.06 (95% CI 1.21\u0026ndash;30.43).\u003c/p\u003e \u003cp\u003eHigher FIB-4 was associated with HVPG\u0026thinsp;\u0026ge;\u0026thinsp;10 in the highest quartile, OR 8.46 (95% CI 2.78\u0026ndash;25.75), with a per unit OR of 1.09 (95% CI 0.99\u0026ndash;1.19). Associations with HVPG\u0026thinsp;\u0026ge;\u0026thinsp;16 were smaller and imprecise.\u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eIn this external validation study of two non-invasive predictive models for portal hypertension severity, we found that both the Vienna and FIB4\u0026thinsp;+\u0026thinsp;models demonstrated only modest discrimination for HVPG\u0026thinsp;\u0026ge;\u0026thinsp;10 mmHg and \u0026ge;\u0026thinsp;16 mmHg, with AUROCs ranging from 0.69\u0026ndash;0.77. Importantly, both models systematically overestimated risk across thresholds, as reflected by negative calibration intercepts and shallow calibration slopes. Among the two, the Vienna model for HVPG\u0026thinsp;\u0026ge;\u0026thinsp;16 mmHg performed best overall, achieving the highest AUROC (0.77) and lowest Brier score (0.18), though calibration remained poor. These findings underscore the challenges of generalizing machine-learning models for portal hypertension beyond their derivation cohorts.\u003c/p\u003e \u003cp\u003eOur results contrast with the robust discrimination and calibration initially reported for both models in European cohorts [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. The Vienna model was originally shown to predict CSPH and severe portal hypertension with AUROCs\u0026thinsp;\u0026gt;\u0026thinsp;0.80, while the FIB4\u0026thinsp;+\u0026thinsp;model demonstrated strong performance as an alternative to elastography-based methods. The attenuation of predictive accuracy in our cohort likely reflects differences in patient population and disease context. Specifically, the prevalence of CSPH and severe portal hypertension in our UNC cohort (33% and 11%, respectively) was considerably lower than in the derivation cohorts, leading to systematic risk overestimation. This \u0026ldquo;spectrum effect\u0026rdquo; is well recognized in risk prediction and emphasizes the need for calibration when applying non-invasive models to new populations [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eDespite modest discrimination, both models maintained high negative predictive values, particularly at the HVPG\u0026thinsp;\u0026ge;\u0026thinsp;16 mmHg threshold (Vienna model NPV 0.95, FIB4\u0026thinsp;+\u0026thinsp;NPV 0.96). These findings suggest that the models may retain clinical utility as rule-out tools, helping to identify patients at low likelihood of severe portal hypertension who may be spared invasive HVPG measurement. However, their poor calibration diminishes confidence in absolute risk estimates and limits their use for individualized risk stratification or clinical decision-making without further adjustment. These findings indicate that recalibration is required before application of these models in U.S. or MASLD-predominant populations.\u003c/p\u003e \u003cp\u003eThe subgroup analyses also suggested variability in model performance by liver disease etiology. While limited by small sample sizes, discrimination appeared higher in viral compared to non-viral etiologies, consistent with prior studies showing that underlying disease mechanism influences the relationship between laboratory parameters and portal pressure. This heterogeneity reinforces the importance of validating predictive tools across diverse patient populations, especially as MASLD becomes the leading indication for liver transplantation in the United States.\u003c/p\u003e \u003cp\u003eModel performance was notably weaker in non-viral etiologies, particularly MASLD. In MASLD-related cACLD, portal hypertension may include a substantial presinusoidal component, which is incompletely reflected by HVPG measurements [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. As a result, HVPG-based thresholds may underestimate portal hypertension severity in this population, potentially contributing to apparent miscalibration and attenuated discrimination. This limitation highlights the complexity of applying HVPG-derived prediction models to metabolically driven liver disease and underscores the need for etiology-specific validation.\u003c/p\u003e \u003cp\u003eOur exploratory analyses of individual laboratory predictors highlighted expected associations between low platelet count, elevated bilirubin, and elevated INR with higher portal pressures, aligning with the biological underpinnings of portal hypertension. These results support the continued centrality of synthetic function markers and platelet count in non-invasive prediction.\u003c/p\u003e \u003cp\u003eThis study has several strengths. It represents the first independent, US-based external validation of the Vienna and FIB4\u0026thinsp;+\u0026thinsp;models, providing important insight into their generalizability beyond European derivation cohorts. The cohort used in this analysis was well-characterized, with rigorously measured HVPG values obtained by experienced interventional radiologists following standardized protocols. This ensured high internal validity and accurate phenotyping of portal pressure severity\u003c/p\u003e \u003cp\u003eIn addition, this study evaluated both discrimination and calibration: two complementary aspects of model performance that are often not assessed together in external validations. The inclusion of calibration analysis, Brier scoring, and subgroup stratification offers a comprehensive assessment of each model\u0026rsquo;s clinical utility and robustness. We also examined model behavior across clinically relevant HVPG thresholds (\u0026ge;\u0026thinsp;10 mmHg and \u0026ge;\u0026thinsp;16 mmHg), capturing the prognostic spectrum of compensated liver disease and severe portal hypertension.\u003c/p\u003e \u003cp\u003eFinally, our cohort reflects the evolving epidemiology of liver disease in the United States, with a large representation of patients with MASLD and mixed etiology cACLD. This provides an early benchmark for how laboratory-based predictive models derived from European viral-predominant populations perform in a contemporary US setting. Together, these features position this study as an important step toward refining and contextualizing noninvasive risk prediction tools for broader clinical application in diverse patient populations.\u003c/p\u003e \u003cp\u003eHowever, limitations must be acknowledged. First, the retrospective single-center design introduces potential for selection bias, as HVPG was obtained primarily in patients undergoing transplant evaluation or with strong clinical suspicion for portal hypertension. Despite this, the prevalence of CSPH in our cohort was lower than that reported in other validation studies, likely reflecting differing referral patterns and inclusion of patients with compensated disease who underwent HVPG for transplant workup rather than clear decompensation. This limits generalizability to broader cACLD populations. Second, our cohort size was modest (n\u0026thinsp;=\u0026thinsp;143), reducing power for subgroup analyses and precision of performance estimates. Third, transient elastography data were largely unavailable, preventing head-to-head comparison of these models with image-based modalities that are increasingly standard in clinical practice. Finally, the relatively low prevalence of CSPH and severe portal hypertension in our sample may have inflated low calibration.\u003c/p\u003e \u003cp\u003eGiven that HVPG is typically reserved for patients with a high clinical suspicion of CSPH, we acknowledge that this cohort likely represents a population with an elevated pre-test probability of portal hypertension. As such, there is potential for selection bias, and findings may not fully generalize to broader populations in which clinical prediction scores would be applied more routinely. This limitation is important when interpreting model performance and planning prospective validation in lower-risk populations.\u003c/p\u003e \u003cp\u003eOur findings highlight the need for recalibration of existing non-invasive models when applied to US-based or MASLD-predominant populations. Prospective multicenter studies with larger and more diverse cohorts will be critical to refine and validate predictive models for portal hypertension. Integration of novel biomarkers, imaging data, and advanced machine learning techniques may further enhance predictive accuracy. Ultimately, achieving accurate, widely available non-invasive risk stratification tools is essential to reducing reliance on HVPG and addressing disparities in portal hypertension care.\u003c/p\u003e "},{"header":"Declarations","content":"\u003ch2\u003eEthical Considerations\u003c/h2\u003e\n\u003cp\u003eThis study was conducted in compliance with ethical standards and was approved by the Institutional Review Board (IRB) at the University of North Carolina at Chapel Hill (IRB #24-1162). Given the retrospective nature of the study and the use of de-identified clinical data, a waiver of informed consent was granted\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eFM, manuscript writing and critical review, decision to publish; MX, statistical analyses and data interpretation; MM, data collection and extraction; AC, data collection and extraction; JF, data collection and extraction; TT, data collection and extraction; AJ, data collection and extraction; NK, manuscript writing and critical review; HY, manuscript writing and critical review; CA, data analysis and critical review; AB, senior supervision and critical review; AM, manuscript writing and critical review, statistical analysis, and decision to publish.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n \u003cli\u003eMandorfer M, Aigner E, Cejna M, et al. Austrian consensus on the diagnosis and management of portal hypertension in advanced chronic liver disease (Billroth IV). \u003cem\u003eWiener Klinische Wochenschrift\u003c/em\u003e. 2023;135(S3):493-523. doi:10.1007/s00508-023-02229-w\u003c/li\u003e\n \u003cli\u003eMandorfer M, Kozbial K, Schwabl P, et al. Changes in hepatic venous pressure gradient predict hepatic decompensation in patients who achieved sustained virologic response to Interferon‐Free therapy. \u003cem\u003eHepatology\u003c/em\u003e. 2019;71(3):1023-1036. doi:10.1002/hep.30885\u003c/li\u003e\n \u003cli\u003eDe Franchis R, Bosch J, Garcia-Tsao G, et al. Baveno VII \u0026ndash; Renewing consensus in portal hypertension. \u003cem\u003eJournal of Hepatology\u003c/em\u003e. 2021;76(4):959-974. doi:10.1016/j.jhep.2021.12.022\u003c/li\u003e\n \u003cli\u003eReini\u0026scaron; J, Petrenko O, Simbrunner B, et al. Assessment of portal hypertension severity using machine learning models in patients with compensated cirrhosis. \u003cem\u003eJournal of Hepatology\u003c/em\u003e. 2022;78(2):390-400. doi:10.1016/j.jhep.2022.09.012\u003c/li\u003e\n \u003cli\u003eRabiee A, Deng Y, Ciarleglio M, et al. Noninvasive predictors of clinically significant portal hypertension in NASH cirrhosis: Validation of ANTICIPATE models and development of a lab‐based model. \u003cem\u003eHepatology Communications\u003c/em\u003e. 2022;6(12):3324-3334. doi:10.1002/hep4.2091\u003c/li\u003e\n \u003cli\u003eUsher-Smith JA, Sharp, SJ, Griffin SJ. The spectrum effect in tests for risk prediction, screening, and diagnosis. \u003cem\u003eBMJ\u003c/em\u003e. 2016; 353. doi:10.1136/bmj.i3139\u003c/li\u003e\n \u003cli\u003eMadir A, Arienzo A, Helmy A, et al. Portal hypertension in patients with nonalcoholic fatty liver disease: Current knowledge and challenges. \u003cem\u003eWorld Journal of Gastroenterology\u003c/em\u003e. 2024;30(4):290-307. doi:10.3748/wjg.v30.i4.290\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"digestive-diseases-and-sciences","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"ddsj","sideBox":"Learn more about [Digestive Diseases and Sciences](http://link.springer.com/journal/10620)","snPcode":"10620","submissionUrl":"https://submission.nature.com/new-submission/10620/3","title":"Digestive Diseases and Sciences","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"Hepatic Venous Pressure Gradient, Portal Hypertension, Compensated Cirrhosis, External Validation","lastPublishedDoi":"10.21203/rs.3.rs-8845244/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8845244/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003ePurpose\u003c/h2\u003e \u003cp\u003eNon-invasive laboratory-based models have been proposed to estimate portal hypertension severity in patients with compensated advanced chronic liver disease (cACLD), but external validation in diverse populations remains limited. We aimed to externally validate two such models, originally developed in European cohorts, for predicting clinically significant portal hypertension (CSPH; hepatic venous pressure gradient [HVPG]\u0026thinsp;\u0026ge;\u0026thinsp;10 mmHg) and severe portal hypertension (HVPG\u0026thinsp;\u0026ge;\u0026thinsp;16 mmHg) in a U.S.-based cACLD cohort.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eWe conducted a retrospective single-center study of adults with cACLD who underwent HVPG measurement between 2014 and 2024. Patients with active hepatic decompensation, hepatocellular carcinoma, or prior transjugular intrahepatic portosystemic shunt were excluded. Model performance of the Vienna laboratory-based model and the FIB-4 plus albumin (FIB4+) model was evaluated using discrimination (area under the receiver operating characteristic curve [AUROC]) and calibration metrics, including calibration intercepts, slopes, and Brier scores.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eThe cohort included 143 patients (median age 56 years; 54% female), predominantly with metabolic dysfunction\u0026ndash;associated steatotic liver disease or alcohol-related liver disease. Median HVPG was 7 mmHg, with CSPH present in 33% and HVPG\u0026thinsp;\u0026ge;\u0026thinsp;16 mmHg in 11%. Discrimination was modest for both models. The Vienna model achieved AUROCs of 0.71 for HVPG\u0026thinsp;\u0026ge;\u0026thinsp;10 mmHg and 0.77 for HVPG\u0026thinsp;\u0026ge;\u0026thinsp;16 mmHg, while the FIB4\u0026thinsp;+\u0026thinsp;model achieved AUROCs of 0.69 and 0.71, respectively. Both models systematically overestimated risk, demonstrating poor calibration across thresholds. Negative predictive values for HVPG\u0026thinsp;\u0026ge;\u0026thinsp;16 mmHg exceeded 95% for both models. Predictive performance was weaker in non-viral etiologies.\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e \u003cp\u003eIn this U.S.-based external validation, laboratory-based models for portal hypertension showed modest discrimination and poor calibration, with systematic risk overestimation. While high negative predictive values suggest potential utility as rule-out tools for severe portal hypertension, recalibration and prospective validation are required before clinical implementation.\u003c/p\u003e","manuscriptTitle":"External Validation of Laboratory-Based Models to Assess Portal Hypertension Severity in Patients with Compensated Chronic Liver Disease","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-02-18 05:24:43","doi":"10.21203/rs.3.rs-8845244/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2026-04-08T16:22:30+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-04-08T13:31:45+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-03-16T08:12:16+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"21993107979421903441149295051954366354","date":"2026-03-16T07:38:11+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"198472301900184808174909921604538028131","date":"2026-03-16T07:05:53+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"223510249875327071106024300008391186798","date":"2026-02-16T15:05:02+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-02-12T00:05:27+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-02-11T20:57:22+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-02-11T13:51:57+00:00","index":"","fulltext":""},{"type":"submitted","content":"Digestive Diseases and Sciences","date":"2026-02-10T21:00:20+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"digestive-diseases-and-sciences","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"ddsj","sideBox":"Learn more about [Digestive Diseases and Sciences](http://link.springer.com/journal/10620)","snPcode":"10620","submissionUrl":"https://submission.nature.com/new-submission/10620/3","title":"Digestive Diseases and Sciences","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"450d75f9-6365-4ded-bf40-7917c969f21e","owner":[],"postedDate":"February 18th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-05-12T15:42:18+00:00","versionOfRecord":[],"versionCreatedAt":"2026-02-18 05:24:43","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8845244","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8845244","identity":"rs-8845244","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}