Predictive target of portal pressure reduction and clinical significance of the PVT subgroup for hepatic encephalopathy after TIPS in decompensated cirrhosis

Introduction

Since its clinical application in the 1980s, transjugular intrahepatic portosystemic shunt (TIPS) has become an important interventional technique for managing complications of portal hypertension in cirrhosis (1). For esophageal and gastric variceal bleeding refractory to medical therapy and refractory ascites, TIPS can significantly reduce portal venous pressure (PVP), control acute bleeding, and improve ascites status. However, the long-term clinical benefit of this therapeutic modality is constrained by the high incidence of post-TIPS hepatic encephalopathy (HE) (1-3). The reported incidence of post-TIPS HE ranges from 30% to 55% (2,4,5), which not only severely impacts patient quality of life but also increases readmission rates and healthcare burden; in some severe cases, shunt reduction or even occlusion is required (5,6). Balancing the portal decompression efficacy of TIPS with effective control of HE risks remains a core clinical challenge in this field.

Current risk assessment for post-TIPS HE primarily relies on preoperative indicators of liver function reserve, such as Child-Pugh class, MELD score, and a history of prior HE (7,8). Predictive models based on these variables, such as the Freeman scoring system, have achieved a certain degree of risk stratification, but their discriminatory ability is generally limited, and they are difficult to directly guide intraoperative procedural decisions (8-10). In recent years, increasing attention has been paid to the association between the magnitude of post-procedural portal pressure reduction (ΔPVP) or residual portal pressure gradient (PPG) and HE (11,12). Theoretically, excessive portal decompression may allow gut-derived neurotoxins to enter the systemic circulation without adequate hepatic metabolism, thereby precipitating HE (5,13). Some retrospective studies suggest that a post-TIPS PPG below a certain threshold (e.g., <5 mmHg or <8 mmHg) is associated with an increased risk of HE. However, considerable heterogeneity exists across studies regarding the timing of measurement, pressure units, and threshold criteria, and a universally accepted optimal safety interval for ΔPVP has yet to be established (1,12). More importantly, whether the relationship between ΔPVP and HE is simply linear or exhibits threshold effects and plateau phases has not been systematically investigated using non-linear regression methods such as restricted cubic splines (RCS).

Concurrently, the understanding of the impact of portal vein thrombosis (PVT) on TIPS outcomes is evolving. Early viewpoints considered PVT a relative contraindication due to significantly increased procedural difficulty and risk of shunt dysfunction (14-16). However, with advancements in interventional techniques, the technical success rate of TIPS in patients with PVT has substantially improved. Current evidence generally suggests that PVT itself is not an independent determinant of poor prognosis post-TIPS; its influence may be indirect, potentially reflecting the severity of underlying liver disease or affecting procedural complexity and long-term shunt patency (17-19). Nevertheless, whether PVT modifies the association pattern between ΔPVP and HE (acts as an effect modifier influencing the safety threshold of ΔPVP remains to be systematically analyzed.

Addressing these questions, this study leverages a single-center cohort of 263 patients with decompensated cirrhosis who underwent TIPS. The objectives are to develop and validate a predictive model for post-TIPS HE risks that integrates preoperative clinical variables with the intraoperative parameter ΔPVP, thereby supplementing existing models lacking intraoperative hemodynamic data. Furthermore, RCS will be employed to precisely analyze the non-linear association between ΔPVP and HE, aiming to identify an optimal safety interval for ΔPVP that can serve as a quantitative reference for intraoperative pressure management. The study further focuses on the PVT subgroup, systematically comparing differences in the HE risks factor profile, the effect size of ΔPVP, and the applicability of a simplified preoperative predictive model between the PVT and non-PVT subgroups. This analysis seeks to address the clinically pertinent question of whether patients with PVT require distinct portal decompression strategies. The findings are expected to provide evidence-based support for individualized PVP management during TIPS and to lay the groundwork for the design of future prospective interventional trials. The analysis workflow is presented in Figure 1. We present this article in accordance with the TRIPOD reporting checklist (available at https://tgh.amegroups.com/article/view/10.21037/tgh-2026-0092/rc).

Figure 1 Study flow diagram of the analytical framework. HE, hepatic encephalopathy; HL, Hosmer-Lemeshow; MELD, Model for End-Stage Liver Disease; PVP, portal venous pressure; ROC, receiver operating characteristic; TIPS, transjugular intrahepatic portosystemic shunt.

Methods

Study design and population

This single-center retrospective cohort study consecutively enrolled patients with decompensated cirrhosis who underwent TIPS at our hospital between January 2014 and January 2026.

Inclusion criteria were: (I) age 18–80 years; (II) diagnosis of cirrhosis confirmed by imaging, laboratory tests, or pathology; (II) presence of indications for TIPS (esophagogastric variceal bleeding or refractory ascites); (IV) complete clinical and follow-up data. All enrolled 263 patients had complete baseline clinical data, intraoperative hemodynamic data and long-term follow-up information. No missing data remained in the final analytical dataset; therefore, no multiple imputation procedures were performed in this study.

Exclusion criteria were: (I) concurrent hepatocellular carcinoma or other malignancies; (II) previous liver transplantation or TIPS; (III) presence of overt HE (West-Haven grade ≥2) prior to TIPS; (IV) severe cardiac, pulmonary, or renal insufficiency. Patients with preoperative West-Haven I HE were not excluded to better reflect the real-world TIPS population. A total of 263 patients were ultimately included in the analysis. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This study was approved by the Institutional Ethics Committee of the hospital (approval No 2025362). Due to the retrospective nature of the study, the requirement for informed consent was waived.

Data collection and variable definitions

Baseline patient data were collected from the electronic medical record system, including: demographic characteristics (age, sex, body mass index); etiology of cirrhosis (viral, alcoholic, other); comorbidities and medical history (diabetes, prior history of HE, PVT, prior endoscopic treatment); severity of liver disease (ascites grade, Child-Pugh class and score, MELD score, ALBI score); laboratory parameters (blood cell count, hemoglobin, platelet count, alanine aminotransferase, aspartate aminotransferase, total bilirubin, albumin, blood ammonia, prothrombin time, international normalized ratio, blood urea nitrogen, creatinine; all laboratory results were the values obtained closest to the surgery date within one week preoperatively); and procedure-related information [indication for TIPS (variceal bleeding/ascites), pre-TIPS PVP, post-TIPS PVP, magnitude of ΔPVP, stent type (covered stent/hybrid stent)]. The pressure reduction rate was calculated as (preoperative pressure - postoperative pressure)/preoperative pressure × 100%. Ascites grade was classified as none/mild/moderate to severe. PVT was diagnosed based on preoperative contrast-enhanced CT or MRI, independently assessed by two radiologists.

TIPS procedure and PVP measurement

All TIPS procedures were performed by the same interventional radiology team. PVP was measured before stent placement (immediately after successful portal vein puncture) and 5 minutes after stent deployment, with patients in a resting state and under stable breathing conditions. Two consecutive measurements were performed by the same operator for each patient, and the mean value was recorded. All pressure measurements were referenced to the right atrial pressure as the zero point, with units expressed in cmH₂O. All measured values represent direct PVP, not the PPG. Pressure tracings were independently reviewed by a second operator; if the difference between the two measurements exceeded 10%, a third measurement was performed and the mean of the three values was used. Based on intraoperative findings, either covered stents (Viabahn, Gore) or hybrid stents (bare stent with covered stent salvage) were selected, with stent diameters ranging from 8-10 mm. PVP was measured again after stent placement.

Outcome definitions

The primary outcome was the occurrence of post-TIPS HE. HE was diagnosed according to the West-Haven criteria, independently assessed by two attending physicians. Any new onset of grade ≥1 HE or an increase of ≥1 grade in pre-existing HE was recorded as an event. Secondary outcomes included: (I) shunt dysfunction (in-stent thrombosis, stenosis ≥50%, or occlusion confirmed by ultrasound or CT); (II) composite outcome (concurrent occurrence of HE and shunt dysfunction); (III) all-cause mortality. Survival time was calculated from the date of TIPS procedure to the date of event occurrence or the last follow-up visit.

Statistical analysis

Normality of continuous variables was assessed using the Kolmogorov-Smirnov test. Normally distributed data were expressed as mean ± standard deviation, and comparisons between groups were performed using the independent samples t-test. Non-normally distributed data were expressed as median (interquartile range), and comparisons between groups were performed using the Mann-Whitney U test. Categorical variables were expressed as counts (%), and comparisons between groups were performed using the chi-square test or Fisher’s exact test. Standardized mean differences (SMD) were used to evaluate inter-group balance, with SMD <0.15 considered indicative of good balance.

Survival analyses were performed using the Kaplan-Meier method to estimate overall survival and cumulative incidence of HE, with comparisons between groups made using the log-rank test. Univariate and multivariate Cox proportional hazards models were used to identify independent risk factors for post-TIPS HE and shunt dysfunction. Forward stepwise regression was used for variable selection in the multivariate model. The proportional hazards assumption was verified using Schoenfeld residuals.

The CatBoost algorithm was employed to construct a machine learning prediction model. Model performance was evaluated using 5-fold cross-validation, expressed as the mean area under the curve (AUC) and standard deviation. Feature importance was ranked using built-in Shapley values. Consistency between the two methods was assessed by comparing the overlapping variables and the overlap rate between variables significantly associated with outcomes in the Cox regression (P<0.05) and the top five features identified by CatBoost.

A preoperative risk prediction model for HE was constructed using multivariate logistic regression with stepwise selection. In this study, we constructed two separate analytical systems. The preoperative HE risks prediction model was built on variables available before surgery. Intraoperative hemodynamic parameters (ΔPVP and ΔPVP rate) were not included in this model. Separately, we performed RCS and stratified analyses on intraoperative ΔPVP. Variables with a P value <0.05 in univariate analysis were eligible for entry into the stepwise selection. Model discrimination was evaluated using the C-index and the area under the AUC. Internal validation was performed using the bootstrap method (1,000 resamples) to calculate the bias-corrected C-index. Calibration was assessed using calibration curves, the Hosmer-Lemeshow goodness-of-fit test, and the Brier score. Clinical utility was evaluated using decision curve analysis (DCA), and the net benefit of the model was compared with that of the Child-Pugh score and the MELD score. Patients were stratified into low-, medium-, and high-risk groups based on tertiles of the total nomogram score, and the trend in HE incidence across groups was tested.

RCS combined with logistic regression (for HE) or Cox regression (for shunt dysfunction) were used to fit the association curve between ΔPVP and outcomes. The test for non-linearity was performed using the likelihood ratio test. ΔPVP was categorized into groups based on quartiles, and adjusted odds ratios (OR) or hazard ratios (HR) and P values for trend were calculated with reference to the lowest category. Interaction analyses were conducted by testing interaction terms between ΔPVP and blood ammonia level, as well as between ΔPVP and stent type. The effect size of ΔPVP per 1 cmH₂O increase was calculated within each subgroup. Sensitivity analyses included: (I) refitting the RCS models using either 3 or 5 knots; (II) repeating the analyses after excluding extreme ΔPVP values (below the 5th or above the 95th percentile). Interaction effects were tested using the likelihood ratio test comparing the goodness-of-fit of models with and without the interaction term. All statistical analyses were performed using R software (version 4.3.2). A two-sided P<0.05 was considered statistically significant.

Results

Baseline characteristics of the study population

A total of 263 patients were enrolled in this study, of whom 89 (33.84%) developed HE after TIPS. As shown in Table 1, patients who developed post-TIPS HE exhibited poorer baseline liver function reserve and distinct intraoperative hemodynamic responses. Disease severity prior to TIPS was significantly higher in the HE, as indicated by a significantly higher proportion of Child-Pugh class C (28.09% vs. 6.90%, P<0.001, SMD =0.581) and significantly worse ALBI scores (−1.67±0.62 vs. −1.96±0.53, P<0.001, SMD =0.522) compared to the non-HE groups. The degree of ascites was also more severe (P=0.03, SMD =0.358). However, blood ammonia levels (P=0.16, SMD =0.198) and the prevalence of PVT (P=0.30, SMD =−0.160), which are traditionally considered to be strongly associated with HE risks, showed no significant differences between groups in this cohort. Furthermore, post-TIPS HE was independently associated with a significantly smaller magnitude of ΔPVP. The ΔPVP and the PVP rate in the HE group were 21.24% and 18.96% lower than those in the non-HE groups, respectively (P<0.001). Concurrently, the proportion of patients receiving covered stents in the HE group was only 61.34% of that in the non-HE (35.96% vs. 58.62%, P=0.001, SMD =0.466), suggesting that the use of bare stents might be a risk factor.

Analysis of the PVT subgroup (Table S1) further refined these findings. Patients with PVT had more severe disease preoperatively (higher Child-Pugh and MELD scores, P<0.05), but the achieved intraoperative portal decompression (ΔPVP and pressure reduction rate, P>0.05) was comparable to that of patients without PVT. This suggests that PVT status itself may not directly influence the immediate hemodynamic outcome of the procedure, and its potential risk needs to be assessed through other pathways.

Postoperative occurrence of multiple outcomes

Based on long-term follow-up data (30.90 months, IQR: 12.00–62.00 months), the risk profile of post-TIPS complications and their determinants exhibited distinct patterns. The cumulative incidence rates of HE and shunt dysfunction were 33.8% and 18.3%, respectively, with 5.7% of patients experiencing both complications concurrently (Table 2).

The risk of postoperative complications demonstrated a clear pattern of modifiability. The incidence of HE showed a significant negative correlation with the intraoperative magnitude of ΔPVP (test for trend P<0.001). As shown in Table 2, among patients with ΔPVP >20 cmH₂O, the HE incidence was only 12.96%; whereas when ΔPVP was <10 cmH₂O, the incidence increased to 45.31%. No significant difference in the incidence of shunt dysfunction was observed across different ΔPVP groups (P=0.76). PVT status was not independently associated with the occurrence of major postoperative complications. The HE incidence in PVT-positive patients was 26.53%, which was not statistically different from the 35.51% observed in PVT-negative patients (P=0.30). Survival analysis further corroborated these findings. The 1-year and 2-year overall survival rates were 87.58% and 78.66%, respectively (Figure 2A). Survival curves stratified by ΔPVP groups showed significant differences (log-rank P<0.05), whereas those stratified by PVT status were not statistically significant (Figure 2B,2C).

Figure 2 Post-TIPS outcomes, survival analysis, and machine learning feature importance. (A) Kaplan-Meier curve showing overall survival of the entire cohort. The 1-year and 2-year survival rates were 87.58% and 78.66%, respectively. (B) Kaplan-Meier curves for overall survival stratified by ΔPVP groups (log-rank P<0.05). (C) Kaplan-Meier curves for overall survival stratified by PVT status (log-rank P>0.05). (D) ROC curves of the CatBoost model for predicting hepatic encephalopathy based on 5-fold cross-validation (mean AUC =0.604, SD =0.070). (E) ROC curves of the CatBoost model for predicting shunt dysfunction based on 5-fold cross-validation (mean AUC =0.688, SD =0.036). (F) Feature importance ranking for HE prediction. ΔPVP rate, age, and blood ammonia were the top three predictors. (G) Feature importance ranking for shunt dysfunction prediction. Blood ammonia, age, and ΔPVP rate were the top three predictors. AUC, area under the curve; CV, clinical variant; HE, hepatic encephalopathy; PVP, portal venous pressure; PVT, portal vein thrombosis; ROC, receiver operating characteristic; SD, standard deviation; TIPS, transjugular intrahepatic portosystemic shunt.

Independent risk factors for postoperative multiple outcomes

Multivariate Cox proportional hazards models were used to identify independent risk factors for postoperative HE and shunt dysfunction. The proportional hazards assumption test indicated that both the HE models (global χ²=4.49, df=6, P=0.61) and the shunt dysfunction model (global χ²=0.09, df=1, P=0.76) satisfied the proportional hazards assumption.

Multivariate Cox regression analysis revealed distinct risk factor profiles for postoperative HE and shunt dysfunction (Table 3). For postoperative HE, age (HR =1.75, 95% CI: 1.39–2.20, P<0.001) and preoperative blood ammonia level (HR =1.19, 95% CI: 1.08–1.45, P=0.008) were identified as independent risk factors, while the ΔPVP rate was an independent protective factor (HR =0.54, 95% CI: 0.34–0.86, P=0.009). Notably, the absolute magnitude of ΔPVP failed to achieve statistical significance after full adjustment (P=0.25). This inconsistency between absolute ΔPVP and ΔPVP rate is attributable to their different mathematical properties: ΔPVP rate is a normalized relative index that corrects variations in baseline PVP among patients, making it more suitable for population-level risk factor analysis. In contrast, absolute ΔPVP is an intuitive intraoperative metric for real-time procedural guidance. This non-significance is also consistent with the nonlinear, plateau-shaped relationship between absolute ΔPVP and HE demonstrated by the RCS analysis, which is better captured by non-linear approaches than by a linear Cox model. PVT showed no significant association with HE in univariate analysis (P=0.35) and was not entered into the multivariate model. For shunt dysfunction, no statistically significant independent risk factors were identified in either univariate or multivariate analyses. Regression analysis was not performed for the composite outcome (HE concurrent with shunt dysfunction) due to the small number of events (n=15).

The CatBoost machine learning model demonstrated a mean AUC of 0.604 (SD =0.070) for the HE prediction model and 0.688 (SD =0.036) for the shunt dysfunction prediction model, based on 5-fold cross-validation (Figure 2D,2E). In the feature importance ranking, ΔPVP rate, age, and blood ammonia ranked as the top three predictors for HE; blood ammonia, age, and ΔPVP rate ranked as the top three predictors for shunt dysfunction (Figure 2F,2G). Consistency assessment between the two methods (Table S2) showed that for the HE outcome, the overlapping variable between the significant variables identified by Cox regression (age, blood ammonia, ΔPVP rate) and the top five features from CatBoost was ΔPVP rate, age, and blood ammonia, yielding an overlap rate of 60%. For the shunt dysfunction outcome, consistency comparison could not be performed as no significant variables were identified by Cox regression.

Preoperative HE risks prediction model

A preoperative risk prediction model for HE was constructed using multivariate logistic regression with stepwise selection, ultimately incorporating two variables: age and Child-Pugh class. In the model, age (OR =1.04, 95% CI: 1.01−1.06, P<0.001) and Child-Pugh class (OR =2.81, 95% CI: 1.75−4.67, P<0.001) were identified as independent risk factors for postoperative HE (Table 4).

The model demonstrated moderate discrimination, with an AUC of 0.69 (95% CI: 0.62−0.76) and a C-index of 0.69 (Figure 3A). Calibration analysis indicated good agreement between predicted probabilities and observed outcomes: the calibration intercept was 0.02, the calibration slope was 0.98, the Hosmer-Lemeshow test yielded P=0.89, and the Brier score was 0.199 (Figure 3B). Decision curve analysis showed that within the threshold probability range of 10–50%, the preoperative model provided higher clinical net benefit compared to both the Child-Pugh score and the MELD score (Figure 3C). at a threshold probability of 30%, the net benefit was 0.114 for the preoperative model, 0.080 for the Child-Pugh score, and 0.064 for the MELD score; at a threshold probability of 40%, the corresponding net benefits were 0.068, 0.065, and 0.038, respectively.

Figure 3 Performance of the preoperative HE prediction model. (A) ROC curve of the model in the derivation cohort. AUC =0.69 (95% CI: 0.62–0.76). C-index =0.69. (B) Calibration plot comparing predicted versus observed HE risk. The diagonal dashed line represents perfect calibration. Calibration intercept: 0.02; calibration slope: 0.98. Hosmer-Lemeshow test: P=0.89. Brier score: 0.199. (C) Decision curve analysis comparing the net benefit of the preoperative model, Child-Pugh score, and MELD score across threshold probabilities from 10% to 50%. Gray lines represent “treat all” and “treat none” strategies. At a 30% threshold probability, net benefit was 0.114 for the preoperative model, 0.080 for Child-Pugh, and 0.064 for MELD; at 40%, the corresponding values were 0.068, 0.065, and 0.038. (D) Nomogram for predicting post-TIPS HE based on age and Child-Pugh grade. Points are assigned to each variable, and the total points correspond to the predicted probability of HE. Patients were stratified into low, medium, and high-risk groups based on tertiles of total points. AUC, area under the curve; CI, confidence interval; HE, hepatic encephalopathy; MELD, Model for End-Stage Liver Disease; ROC, receiver operating characteristic; TIPS, transjugular intrahepatic portosystemic shunt.

Based on the nomogram (Figure 3D), a total score was calculated for each patient, and patients were stratified into three risk groups according to tertiles. The low-risk group (total score ≤38.7, n=89) had a postoperative HE incidence of 21.35% (19/89); the medium-risk group (total score 38.7–53.2, n=90) had an incidence of 28.89% (26/90); and the high-risk group (total score >53.2, n=84) had an incidence of 52.38% (44/84), with a statistically significant difference across groups (P for trend <0.001).

The association and stratified analyses between ΔPVP and postoperative HE

The association between the intraoperative magnitude of ΔPVP and the risk of postoperative HE exhibited a nonlinear pattern. RCS analysis, adjusted for age, sex, Child-Pugh class, MELD score, blood ammonia, stent type, and ascites grade, with the median ΔPVP of 15 cmH₂O as the reference, revealed a curve descriptively consistent with a reverse L-shaped pattern, with a steep decline at lower ΔPVP values and a flattening above 20 cmH₂O (P for nonlinearity =0.46). The formal test for nonlinearity was not statistically significant, which we interpret as limited power to detect curvature rather than evidence of true linearity. Notably, the categorical analysis (HE incidence 12.96% in the >20 cmH₂O group vs. 45.31% in the <10 cmH₂O group, P for trend <0.001) provides stronger support for the clinical threshold. The lowest HE risk was observed at ΔPVP =29.7 cmH₂O, corresponding to an OR of 0.18 (95%CI 0.05–0.66); the HR approached unity within the ΔPVP range of 14.3–15.8 cmH₂O (Figure 4A). When ΔPVP was categorized into quartiles (<11, 11–15, 15–20, >20 cmH₂O), the incidence of HE was 45.71%, 46.97%, 26.03%, and 12.96%, respectively. Using the >20 cmH₂O group as the reference, multivariate logistic regression showed significantly elevated HE risks in the <11 cmH₂O group (OR =4.27, 95% CI: 1.58–11.53, P=0.004) and the 11–15 cmH₂O group (OR =6.65, 95% CI: 2.40–18.44, P<0.001); the 15–20 cmH₂O group showed a borderline increase (OR =2.64, 95% CI: 0.94–7.40, P=0.07), with a test for trend P=0.11 (Table 5).

Figure 4 The association between ΔPVP and post-TIPS outcomes, sensitivity analysis, and model performance in the PVT subgroup. (A) RCS analysis of the association between ΔPVP and post-TIPS HE, adjusted for age, sex, Child-Pugh class, MELD score, blood ammonia, stent type, and ascites grade. (B) RCS analysis of the association between ΔPVP and shunt dysfunction, adjusted for the same covariates. (C) Sensitivity analysis of the ΔPVP-HE association after excluding extreme ΔPVP values (below the 5th or above the 95th percentile). (D) ROC curve of the simplified prediction model (age + Child-Pugh) in the overall cohort (without PVT) and PVT subgroup. (E,F) Calibration plots of the simplified prediction model in the overall cohort (blue) and PVT subgroup (red). The diagonal dashed line represents perfect calibration. For the overall cohort: Hosmer-Lemeshow P=0.67, Brier score =0.196. For the PVT subgroup: Hosmer-Lemeshow P=0.88, Brier score =0.175. AUC, area under the curve; CI, confidence interval; HE, hepatic encephalopathy; MELD, Model for End-Stage Liver Disease; PVP, portal venous pressure; PVT, portal vein thrombosis; RCS, restricted cubic splines; ROC, receiver operating characteristic; TIPS, transjugular intrahepatic portosystemic shunt.

No significant association was observed between ΔPVP and shunt dysfunction. RCS analysis, with the median ΔPVP of 15 cmH₂O as the reference, demonstrated a flat fitted curve (P for nonlinearity =0.49), with the lowest risk point at 13.2 cmH₂O corresponding to an HR of 0.97 (95% CI: 0.68–1.40); the HR approached unity within the ΔPVP range of 11.5–16.4 cmH₂O (Figure 4B). After stratification by the same quartiles, no significant differences in the incidence of shunt dysfunction were observed across ΔPVP groups (test for trend P=0.39) (Table S3). The HRs for each 1, 5, and 10 cmH₂O increase in ΔPVP as a continuous variable were 1.06 (95% CI: 0.99–1.13), 1.31 (95% CI: 0.94–1.84), and 1.73 (95% CI: 0.88–3.37), respectively.

The interaction between ΔPVP and blood ammonia level was not statistically significant (P=0.59), nor was the interaction between ΔPVP and stent type (P=0.23). In stratified analysis by stent type, each 1 cmH₂O increase in ΔPVP was associated with a 12% reduction in HE risk in the covered stent group (OR =0.88, 95% CI: 0.81–0.96, P=0.003) and a 6% reduction in the hybrid stent group (OR =0.94, 95% CI: 0.88–1.00, P=0.050) (Table 5). Sensitivity analyses yielded robust results. Refitting the RCS curves using either 3 or 5 knots yielded P values for nonlinearity of 0.237 and 0.064, respectively. After excluding extreme ΔPVP values (below the 5th or above the 95th percentile), the lowest risk point shifted to 24.7 cmH₂O, corresponding to an OR of 0.15 (95% CI: 0.05–0.51); the HR approached unity within the ΔPVP range of 10.5–15.6 cmH₂O, and the negative association trend between ΔPVP and HE remained substantially unchanged (Figure 4C).

PVT subgroup analysis results

No significant differences were observed between the PVT group and the non-PVT group in the incidence of postoperative HE or shunt dysfunction (Table S4). The HE incidence was 26.53% (13/49) in the PVT group and 35.51% (76/214) in the non-PVT group (P=0.30); the incidence of shunt dysfunction was 20.41% (10/49) and 17.76% (38/214), respectively (P=0.82). A formal interaction analysis between PVT status and ΔPVP in the multivariate logistic regression model demonstrated a significant interaction (P=0.02). The PVT subgroup included 49 patients with 13 HE events.

Multivariate logistic regression revealed a modifying effect of PVT status on risk factors for HE. In the overall cohort, age (OR =1.04, 95% CI: 1.02−1.07, P=0.001) and Child-Pugh class C (OR =8.21, 95% CI: 3.25−22.20, P<0.001) were independent risk factors for HE. Within the PVT subgroup, age showed borderline significance (OR =1.07, 95% CI: 1.00−1.15, P=0.056), while Child-Pugh class C did not reach statistical significance (OR =3.12, 95% CI: 0.21−54.11, P=0.40), with the wide confidence interval consistent with the small subgroup sample size (n=49, 13 events); in the non-PVT subgroup, both age (OR =1.04, 95% CI: 1.01−1.06, P=0.005) and Child-Pugh class C (OR =9.30, 95% CI: 3.44−27.28, P<0.001) were significant risk factors. The interaction between age and PVT status was not statistically significant (P for interaction =0.44) (Table 6). Regarding shunt dysfunction, age was identified as a risk factor only in the overall cohort. Multivariate analysis in the overall cohort showed that age was an independent protective factor for shunt dysfunction (OR =0.96, 95% CI: 0.93−0.99, P=0.007), while PVT status and other liver function indicators showed no significant associations (Table S5). Within the PVT subgroup, no variables were found to be independently associated with shunt dysfunction (P>0.05) (Table S6).

The strength of the association between ΔPVP and HE differed between the PVT and non-PVT subgroups. In the non-PVT subgroup, each 5 cmH₂O increase in ΔPVP was significantly associated with reduced HE risk (P<0.001), and patients with ΔPVP above the median had a 4.23-fold higher HE risk compared to those below the median (95% CI: 2.32–7.95, P<0.001); in contrast, in the PVT subgroup, no significant association was observed between ΔPVP and HE (per 5 cmH₂O increase, P=0.73; above vs. below median, OR =1.17, 95% CI: 0.33–4.29, P=0.81) (Table S7). After stratification by ΔPVP quartiles, the HE incidence in the PVT subgroup was 23.08% (3/13), 33.33% (4/12), 33.33% (4/12), and 16.67% (2/12) in quartiles Q1 through Q4, respectively, with none of the ORs reaching statistical significance when using the Q4 group as the reference; in the non-PVT subgroup, HE incidence progressively decreased from 50.88% (29/57) in Q1 to 9.76% (4/41) in Q4, with significantly elevated HE risks in the Q1 (OR =8.63, 95% CI: 2.86–26.04, P<0.001) and Q2 groups (OR =8.33, 95% CI: 2.74–25.33, P<0.001), and a borderline increase in the Q3 group (OR =2.96, 95% CI: 0.96–9.13, P=0.06) using Q4 as the reference (Table S8). The optimal ΔPVP range in the PVT subgroup (21–37 cmH₂O) largely overlapped with that in the non-PVT subgroup (20–32 cmH₂O).

The simplified predictive model maintained moderate discriminative ability in the PVT subgroup. The binary logistic regression model based on age and Child-Pugh class (without the PVT variable) yielded an AUC of 0.70 (95% CI: 0.63–0.70) in the overall cohort, with a Hosmer-Lemeshow test P=0.67 and a Brier score of 0.196. When this model was applied to the PVT subgroup, the AUC was 0.70 (95% CI: 0.54–0.71), with good calibration (HL P=0.88, Brier score 0.175) (Figure 4D-4F), indicating that the simplified model achieved predictive performance in PVT patients comparable to that in the overall cohort (Table S9).

Discussion

The occurrence of HE after TIPS is fundamentally a consequence of the imbalance between portosystemic shunting and hepatic functional reserve (5,20). This study elucidated the quantitative characteristics of this imbalance from three perspectives: the reverse L-shaped association between ΔPVP and HE with its threshold effect, the complementary roles of preoperative disease severity and intraoperative hemodynamic parameters in the predictive model, and the modifying effect of PVT status on this association.

The reverse L-shaped pattern of the ΔPVP-HE association suggests the existence of a benefit plateau for portal decompression. When ΔPVP exceeded 20 cmH₂O, the risk of HE remained at a low level below 13%, and further increasing the magnitude of decompression did not confer additional risk reduction. The threshold analysis using absolute ΔPVP was chosen to provide a directly actionable target for intraoperative pressure management, and the observed plateau in the RCS analysis supports its clinical utility despite its non-significance in the linear Cox model. This finding contrasts with concerns raised in some studies that excessive pressure reduction might precipitate HE. Previous studies have suggested maintaining post-TIPS residual pressure between 5 and 8 mmHg to avoid HE, but this recommendation was primarily based on clinical experience rather than systematic dose-effect analysis (12,21,22). A European multicenter study observed an increased risk of HE when post-TIPS residual pressure was below 12 mmHg, which is generally consistent with the threshold identified in this study (20 cmH₂O corresponds to approximately 14.7 mmHg) (22-24). However, by using ΔPVP rather than residual pressure as the variable, this study more directly reflects the effect size of the procedural intervention. Notably, the RCS curve exhibited a steep ascent in the low ΔPVP range, indicating a threshold effect rather than a continuous linear relationship for the contribution of mild pressure reduction to HE risk. This pattern aligns with the pathophysiological characteristics of HE, where the risk increases nonlinearly when portosystemic shunting exceeds the individual’s hepatic metabolic reserve capacity (25,26).

The construction strategy of the preoperative predictive model reflects the integration of static risk and dynamic intervention. The model incorporated two variables (age, Child-Pugh class) and achieved discrimination comparable to existing scoring systems while demonstrating superior clinical utility. The Freeman score requires data on prior HE history, Child-Pugh class, and MELD score, with AUC values ranging from 0.65 to 0.72 across different validation cohorts (8-10). The present model maintained equivalent discrimination while significantly reducing data collection complexity. Decision curve analysis showed that within the threshold probability range of 10% to 50%, the net benefit of the model was superior to that of using the Child-Pugh or MELD score alone, an advantage arising from the recalibration of risk distribution through multi-indicator integration. From a clinical decision-making perspective, the improved net benefit implies that, at the same intervention threshold, the model could reduce unnecessary preventive measures or earlier identify high-risk populations requiring intervention (27,28). The nomogram-based risk stratification provides a quantitative basis for clinical management, with the low-risk group suitable for routine follow-up, the intermediate-risk group requiring intensified monitoring, and the high-risk group warranting consideration of preemptive intervention.

The disappearance of the protective effect of ΔPVP in the PVT subgroup represents a core finding of this study. Previous research on the impact of PVT on TIPS outcomes has primarily focused on technical success rates and shunt patency. Berengy et al., based on multicenter data analysis, concluded that PVT does not affect post-TIPS survival or HE incidence but did not further investigate its modifying effect on hemodynamic intervention efficacy (29). The present study found no significant association between ΔPVP and HE in the PVT group, whereas in the non-PVT group, each 5 cmH₂O increment in ΔPVP corresponded to a significantly reduced HE risk, suggesting that PVT fundamentally alters the response pattern of the portal venous system to pressure intervention. Possible explanations for this difference include the following. Patients with PVT often have extensive portosystemic collateral circulation, which continues to shunt some intestinal toxins after TIPS, attenuating the protective effect of ΔPVP (5,25). Thrombus itself alters the blood flow pattern within the portal venous system, such that a decrease in main portal vein pressure may not proportionally translate into improved hepatic sinusoidal perfusion (25). The more severe underlying liver disease in the PVT group may render patients highly sensitive to any degree of portosystemic shunting, with hepatic functional reserve rather than hemodynamic parameters becoming the dominant factor in HE risks. Several studies have indicated that the prognosis of patients with PVT depends more on underlying liver disease than on the TIPS procedure itself (30-32). This study provides quantitative evidence for this perspective from the standpoint of ΔPVP effect modification. Notably, the simplified model based on age and Child-Pugh grade maintained an AUC of 0.70 in the PVT subgroup, comparable to that in the overall cohort, indicating good discriminative ability in patients with PVT.

Comparison with previous studies should be interpreted within specific clinical contexts. The use of ΔPVP as the exposure variable in this study has the advantage of directly quantifying the intensity of the procedural intervention but the disadvantage that, compared with post-TIPS residual pressure, ΔPVP is not independent of preoperative pressure levels. Patients with high preoperative pressure may still have relatively high post-TIPS residual pressure even after achieving the same ΔPVP, which explains why the group with ΔPVP >20 cmH₂O had the lowest HE risk in this study, while RCS identified the lowest risk point at 29.7 cmH₂O. This pattern suggests that clinical practice should consider both the absolute magnitude of pressure reduction and the post-TIPS residual pressure. Regarding the effect modification by PVT, this study represents the first systematic report, and no comparable studies are currently available for direct comparison. Future multicenter data are needed to validate the generalizability of this finding and to explore the underlying hemodynamic mechanisms.

Nevertheless, this study has several limitations. The single-center retrospective design limits the generalizability of the findings. The sample size of the PVT subgroup was relatively small; although the number of covariates in multivariate regression was strictly controlled, estimation precision remained limited, as reflected in wide confidence intervals. Although we detected a significant interaction between PVT status and ΔPVP, the conclusions about PVT’s effect modification should be interpreted cautiously given the limited subgroup size, and larger patient cohorts are required to validate these findings. PVP was measured at a single intraoperative time point and could not capture the dynamic influence of postoperative vasoactive medications or volume status changes on pressure. Potential confounding factors such as nutritional status, sarcopenia, and prophylactic lactulose use were not included in the analysis. The model was validated only via internal bootstrap resampling without external validation. Its moderate discriminative performance and single-center design further limit generalizability. Additionally, the non-significant p-value for nonlinearity indicates limited power to detect curvature rather than evidence of a linear relationship; accordingly, the proposed 20 cmH₂O threshold is supported primarily by the categorical analysis and visual plateau and requires validation in larger cohorts.

Conclusions

In conclusion, ΔPVP is a predictor of post-TIPS HE, with a reverse L-shaped association and a clinically meaningful threshold at 20 cmH₂O. The preoperative model incorporating age, Child-Pugh class simplifies clinical application while maintaining adequate discrimination. Patients with PVT show significantly less dependence on ΔPVP compared to those without PVT, suggesting that HE management in this population should focus more on improving underlying liver function. Future research should conduct multicenter external validation for the preoperative prediction model, investigate whether ΔPVP-targeted intraoperative decision-making translates into improved clinical outcomes and further explore the hemodynamic basis underlying the modifying effect of PVT on the ΔPVP-HE association.

Acknowledgments

The authors would like to thank all patients and their families who participated in this study, as well as the medical and nursing staff of the Department of Interventional Radiology for their assistance in data collection and patient follow-up.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://tgh.amegroups.com/article/view/10.21037/tgh-2026-0092/rc

Data Sharing Statement: Available at https://tgh.amegroups.com/article/view/10.21037/tgh-2026-0092/dss

Peer Review File: Available at https://tgh.amegroups.com/article/view/10.21037/tgh-2026-0092/prf

Funding: None.

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tgh.amegroups.com/article/view/10.21037/tgh-2026-0092/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This study was approved by the Institutional Ethics Committee of our Hospital (approval No 2025362). Due to the retrospective nature of the study, the requirement for informed consent was waived.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. European Association for the Study of the Liver. EASL Clinical Practice Guidelines on TIPS. J Hepatol 2025;83:177-210. [Crossref] [PubMed]
2. Wang Z, Wang J, Ye J, et al. Covered TIPS Created with Viatorr Versus Fluency Stent-Grafts for the Refractory Ascites in Patients with Cirrhosis: An Observational Study. Acad Radiol 2025;32:5112-25. [Crossref] [PubMed]
3. Xu XT, Jiang MJ, Fu YL, et al. Incidence and efficacy of strategies for preventing hepatic encephalopathy following transjugular intrahepatic portosystemic shunt: A meta-analysis. World J Hepatol 2025;17:104890. [Crossref] [PubMed]
4. Rudler M, Bouzbib C, Sultanik P, et al. Overt hepatic encephalopathy after elective and preemptive TIPS: Risk factors and prognosis. JHEP Rep 2025;7:101548. [Crossref] [PubMed]
5. Li Y, Wu YT, Wu H. Management of hepatic encephalopathy following transjugular intrahepatic portosystemic shunts: Current strategies and future directions. World J Gastroenterol 2025;31:103512. [Crossref] [PubMed]
6. Kim DH, Anbari H, Gadani S, et al. Post-TIPS Encephalopathy: Clinical Presentation, Diagnosis, and Management. Tech Vasc Interv Radiol 2025;28:101063. [Crossref] [PubMed]
7. Fang Z, Liu Z, Pu Z, et al. Risk prediction models for hepatic encephalopathy following TIPS: a systematic review and meta-analysis. Front Med (Lausanne) 2026;13:1707035. [Crossref] [PubMed]
8. Xu XJ, Yin L, Zhu YJ, et al. Comparing efficacy of different scoring models to predict hepatic encephalopathy after TIPS in cirrhotic patients. Ann Med 2025;57:2514082. [Crossref] [PubMed]
9. Jiang Z, Chu J, Han Z, et al. Development and validation of a preoperative clinical parameter-based nomogram to predict overt hepatic encephalopathy within 1 year after transjugular intrahepatic portosystemic shunt. Front Med (Lausanne) 2025;12:1634368. [Crossref] [PubMed]
10. Shen MY, Li ZJ, Wu JN. Letter to the editor regarding: 'comparing efficacy of different scoring models to predict hepatic encephalopathy after TIPS in cirrhotic patients'. Ann Med 2025;57:2529574. [Crossref] [PubMed]
11. China L, Patch D. Identifying the sweet spot in portal pressure reduction with TIPS. Hepatology 2025;82:1053-5. [Crossref] [PubMed]
12. Lv Y, Ye J, Bai W, et al. Pre-emptive TIPS with 8-mm stents reduces hepatic encephalopathy without compromising efficacy in acute variceal bleeding. JHEP Rep 2025;7:101591. [Crossref] [PubMed]
13. Kabelitz MA, Hartl L, Schaub G, et al. Identification of optimal portal pressure decrease to control ascites while minimizing HE after TIPS: A multicenter study. Hepatology 2025;82:1172-86. [Crossref] [PubMed]
14. Sanyal AJ, Shiffman ML. Transjugular intrahepatic portosystemic shunt: a medical perspective. Dig Dis 1995;13:153-62. [Crossref] [PubMed]
15. Copelan A, Kapoor B, Sands M. Transjugular intrahepatic portosystemic shunt: indications, contraindications, and patient work-up. Semin Intervent Radiol 2014;31:235-42. [Crossref] [PubMed]
16. Rajesh S, Singh S, Philips CA. Transjugular Intrahepatic Portosystemic Shunt in Chronic Portal Vein Thrombosis-From Routine Recommendations to Demanding Scenarios. Diagnostics (Basel) 2022;12:3100. [Crossref] [PubMed]
17. Knott E, Kapoor B, Chansangrat J, et al. Single-Center Experience of Portal Vein Recanalization and Transjugular Intrahepatic Shunt Placement in Patients with Portovisceral Thrombosis. J Vasc Interv Radiol 2025;36:1313-20. [Crossref] [PubMed]
18. Du S, Luo X, Yang L, et al. Clinical significance of portal vein thrombosis in patients with decompensated cirrhosis: A still matter of debate. J Transl Int Med 2025;13:100-3. [Crossref] [PubMed]
19. Barranco Acosta S, Lombardo Galera MS, García Jurado PB, et al. Transjugular Intrahepatic Portosystemic Shunt for Portal Vein Thrombosis in Cirrhotic Patients: 18-Year Experience in a Tertiary Referral Hospital. Diagnostics (Basel) 2025;15:2878. [Crossref] [PubMed]
20. Xu X, Zhu T, Jing C, et al. Hepatic encephalopathy treatment after transjugular intrahepatic portosystemic shunt: a new perspective on the gut microbiota. Front Med (Lausanne) 2025;12:1423780. [Crossref] [PubMed]
21. Liu G, Xiao S, Wang X, et al. Determining the optimal portal pressure gradient after small-diameter TIPS for ascites: a retrospective study. Ther Adv Gastroenterol 2025;18:17562848251372265. [Crossref] [PubMed]
22. Roccarina D, Saltini D, Senzolo M, et al. Shunt magnitude is a key determinant of overt hepatic encephalopathy in patients undergoing TIPS. JHEP Rep 2026;8:101676. [Crossref] [PubMed]
23. Roccarina D, Saltini D, Adotti V, et al. End-procedural adherence to recommended hemodynamic targets does not improve the outcome of elective TIPS in patients with cirrhosis. Hepatology 2025;82:1155-71. [Crossref] [PubMed]
24. Kabelitz MA, Hartl L, Schaub G, et al. Identification of optimal portal pressure decrease to control ascites while minimizing HE after TIPS: A multicenter study. Hepatology 2025;82:1172-86. [Crossref] [PubMed]
25. Schindler P, Heinzow H, Trebicka J, et al. Shunt-Induced Hepatic Encephalopathy in TIPS: Current Approaches and Clinical Challenges. J Clin Med 2020;9:3784. [Crossref] [PubMed]
26. Vidal-González J, Quiroga S, Simón-Talero M, et al. Spontaneous portosystemic shunts in liver cirrhosis: new approaches to an old problem. Ther Adv Gastroenterol 2020;13:1756284820961287. [Crossref] [PubMed]
27. Vickers AJ, Holland F. Decision curve analysis to evaluate the clinical benefit of prediction models. Spine J 2021;21:1643-8. [Crossref] [PubMed]
28. Zhao L, Leng Y, Hu Y, et al. Understanding decision curve analysis in clinical prediction model research. Postgrad Med J 2024;100:512-5. [Crossref] [PubMed]
29. Berengy MS, Abd El-Hamid Hassan EM, Ibrahim AH, et al. Safety and efficacy of transjugular intrahepatic portosystemic shunts vs endoscopic band ligation plus propranolol in patients with cirrhosis with portal vein thrombosis: a systematic review and meta-analysis. J Gastrointest Surg 2024;28:316-26. [Crossref] [PubMed]
30. Cheloff AZ, Bonanni LJ, Kirschenbaum JD, et al. Treatment of portal vein thrombosis in cirrhosis is associated with no survival advantage: a retrospective controlled study. Hepatol Int 2025;19:191-8. [Crossref] [PubMed]
31. Intagliata NM, Caldwell SH, Tripodi A. Diagnosis, Development, and Treatment of Portal Vein Thrombosis in Patients With and Without Cirrhosis. Gastroenterology 2019;156:1582-1599.e1. [Crossref] [PubMed]
32. Englesbe MJ, Kubus J, Muhammad W, et al. Portal vein thrombosis and survival in patients with cirrhosis. Liver Transpl 2010;16:83-90. [Crossref] [PubMed]