Development and validation of a predictive model for acute-on-chronic liver failure after transjugular intrahepatic portosystemic shunt

Abstract

BACKGROUND

Transjugular intrahepatic portosystemic shunt (TIPS) is a cause of acute-on-chronic liver failure (ACLF).

AIM

To investigate the risk factors of ACLF within 1 year after TIPS in patients with cirrhosis and construct a prediction model.

METHODS

In total, 379 patients with decompensated cirrhosis treated with TIPS at Nanjing Drum Tower Hospital from 2017 to 2020 were selected as the training cohort, and 123 patients from Nanfang Hospital were included in the external validation cohort. Univariate and multivariate logistic regression analyses were performed to identify independent predictors. The prediction model was established based on the Akaike information criterion. Internal and external validation were conducted to assess the performance of the model.

RESULTS

Age and total bilirubin (TBil) were independent risk factors for the incidence of ACLF within 1 year after TIPS. We developed a prediction model comprising age, TBil, and serum sodium, which demonstrated good discrimination and calibration in both the training cohort and the external validation cohort.

CONCLUSION

Age and TBil are independent risk factors for the incidence of ACLF within 1 year after TIPS in patients with decompensated cirrhosis. Our model showed satisfying predictive value.

Key Words: Acute-on-chronic liver failure, Transjugular intrahepatic portosystemic shunt, Influencing factor analysis, Risk prediction model, Nomogram

Core Tip: Previous studies have proposed several models for predicting the prognosis of patients with acute-on-chronic liver failure (ACLF). However, to date, no such prediction model exists for forecasting the occurrence of ACLF following transjugular intrahepatic portosystemic shunt (TIPS). This study provides an internally and externally validated nomogram model, as well as an easy-to-use risk score scale for predicting the risk of ACLF within 1 year after TIPS. This information could enable physicians to effectively communicate the risks and benefits of the procedure to patients, facilitating shared decision-making.

INTRODUCTION

Acute-on-chronic liver failure (ACLF) is a clinical syndrome characterized by acute liver failure on underlying chronic liver disease. It manifests as jaundice, coagulation dysfunction, and hepatic encephalopathy. Currently, medical therapy, including pharmacological treatment and artificial liver therapy, is the main treatment for patients with ACLF. For those who do not respond to routine medical treatment, liver transplantation is the only curative treatment[1]. However, the use of liver transplantation is limited by the number of donors and the high cost of the procedure. Despite aggressive therapy, the short-term mortality remains very high among patients with ACLF, usually over 30%[2,3]. Therefore, it is crucial to identify the risk factors of ACLF, identify high-risk patients in the early stages, and improve the management of high-risk patients to delay or prevent the progression of liver damage to ACLF[4]. In recent years, several risk score models have been developed to predict the prognosis of patients with ACLF, such as the model for end-stage liver disease (MELD) score, chronic liver failure-sequential organ failure assessment score, and the Asian Pacific Association for the Study of the Liver (APASL) ACLF research consortium score[5]; however, studies predicting the incidence of ACLF are relatively rare.

Transjugular intrahepatic portosystemic shunt (TIPS) is an interventional strategy that reduces portal venous pressure. It is mostly used to treat the complications of portal hypertension, such as esophagogastric variceal bleeding and refractory ascites, in patients with cirrhosis[6]. Although the TIPS procedure can effectively reduce portal venous pressure and relieve the complications of portal hypertension, it may also deteriorate liver function and induce ACLF in some patients[7]. A previous study showed that within 3 months after TIPS, liver failure occurred in 9.2% of patients with cirrhosis with MELD ≤ 12 points[8].

Currently, there are no studies on ACLF after TIPS. Predicting the risk of ACLF after TIPS in patients with cirrhosis may help clinicians make more accurate treatment decisions. Therefore, we investigated independent predictors of ACLF within 1 year after TIPS and developed an effective predictive model to predict the risk of ACLF after TIPS. This predictive model can help physicians identify patients at high risk of postoperative ACLF before conducting the TIPS procedure.

MATERIALS AND METHODS

Patients

In total, 828 consecutive patients with cirrhosis who underwent TIPS at Nanjing Drum Tower Hospital and Nanfang Hospital, Southern Medical University, between January 2017 and December 2020, were screened based on the inclusion and exclusion criteria. Inclusion criteria were as follows: (1) Age more than or equal to 18 years; (2) meeting the diagnostic criteria of cirrhosis based on medical history, laboratory, and imaging studies; and (3) successful TIPS procedure. Exclusion criteria were as follows: (1) Confirmed APASL-ACLF before TIPS; (2) undergoing recanalization for occluded TIPS; (3) lost to follow-up within 1 year after TIPS; (4) absence of preoperative clinical data; (5) concomitant cancer; and (6) heart, kidney, and other major organ failure. In total, 502 patients were included in the final analysis. First, 379 patients at Nanjing Drum Tower Hospital, including 42 patients in the ACLF group and 337 patients in the non-ACLF group, were enrolled as the training cohort and were enrolled to develop the model. Then, 123 patients at Nanfang Hospital, Southern Medical University, including 12 patients in the ACLF group and 111 patients in the non-ACLF group, were enrolled in the external validation group. This study complies with the Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD) statement. The flow chart is shown in Figure 1. This study followed the ethical principles reported in the Declaration of Helsinki and Istanbul. The protocol of this study was approved by the Ethics Committee of The Affiliated Drum Tower Hospital of Nanjing University Medical School (approval No. 2022-596-02). Verbal informed consent was obtained from all participants via telephone.

Figure 1 Patients’ flowchart. TIPS: Transjugular intrahepatic portosystemic shunt; APASL: Asian Pacific Association for the Study of the Liver; ACLF: Acute-on-chronic liver failure.

Data collection

Preoperative and intraoperative variables were recorded for each patient. Preoperative variables included gender, age, history of diabetes mellitus, history of portal vein thrombosis, history of splenectomy, etiology of cirrhosis, and TIPS indication. Four liver function scores, including the Child-Pugh score, MELD score, MELD-Na score, and CLIF-C acute decompensation (AD) score, were recorded. Preoperative laboratory indicators included white blood cell (WBC) count, platelet count, alanine transaminase (ALT), serum total bilirubin (TBil), serum albumin (Alb), serum creatinine, serum sodium (Na), international normalized ratio (INR), and fibrinogen, which were measured within 1 wk before surgery. Intraoperative variables included stent diameter and puncture site.

Study endpoints and follow-up

The primary endpoint was the incidence of ACLF within 1 year after TIPS. Postoperative death, postoperative rebleeding, and postoperative stent stenosis were secondary endpoints. All patients were followed in clinic or via telephone at 1, 3, and 6 months after TIPS and every 6 months thereafter until August 31, 2022.

Definition of ACLF

There is no universal definition of ACLF globally; therefore, we used the definition of APASL because its study population was similar to ours. According to the APASL ACLF consensus of 2019, ACLF is defined as an acute hepatic injury manifesting as jaundice (serum bilirubin ≥ 5 mg/dL or 85 umol/L) and coagulopathy (INR ≥ 1.5 or prothrombin activity < 40%) developing within 4 wk and accompanied by clinical ascites and/or encephalopathy in a patient with previously diagnosed or undiagnosed chronic liver disease/cirrhosis. Previously, it was shown that ACLF is associated with a high 28-d mortality rate.

Statistical analysis

R 4.2.1 was used for data processing and development and validation of the nomogram model. Continuous variables are expressed as mean ± SD or median (interquartile range), and statistical differences were estimated using the independent sample t-test or the Mann-Whitney U test. Count data are expressed as the number of cases and percentage (%). We used the χ² test or Fisher’s exact probability method to compare groups. Univariate regression analysis was conducted using binary logistic regression, and variables with P < 0.10 in univariate analysis were included in multivariate logistic regression analysis. We used the Akaike Information Criterion (AIC) as a stopping criterion and selected the model with the lowest AIC using the backward stepwise method. The nomogram model was then plotted. We performed a bootstrap internal validation procedure with 1000 bootstrap resamples. Additionally, a geographically independent cohort was used for external validation. Model performance was measured using discrimination and calibration. The model’s discrimination ability was assessed by calculating the area under the receiver operating characteristic (ROC) curve, and calibration was assessed using the Hosmer-Lemeshow test and calibration plots. Using ROC curves in the entire cohort, we assessed the predictive performance of the nomogram model compared to four commonly used liver function scoring systems (i.e., MELD score, Child-Pugh score, CLIF-C AD score, and MELD-Na score). The ROC curves were compared using DeLong’s test. To facilitate clinical application, we integrated the regression coefficients of each variable in the model and assigned values to these variables to develop a risk score model. Survival analysis was conducted using the Kaplan-Meier method, and differences between groups were measured using the log-rank test. A P value < 0.05 was considered statistically significant.

RESULTS

In the entire cohort, 828 patients were successfully treated with TIPS from January 2017 to December 2020. Of them, 6 patients who were diagnosed with APASL-ACLF before TIPS procedure, 99 patients who underwent TIPS for shunt dysfunction, 156 patients who were lost to follow-up within 1 year after TIPS, and 65 patients with notable missing data were excluded based on the exclusion criteria. In total, 502 patients were enrolled in the final analysis, including 379 patients in the training cohort and 123 patients in the external validation cohort. The training cohort consisted of 219 men and 160 women, with a median age of 59 years (range: 18-86 years). The median follow-up time was 629 d in the ACLF group and 722 d in the non-ACLF group. The external validation cohort consisted of 100 men and 23 women, with a median age of 51 years (range: 22-74 years). The median follow-up time was 364 d in the ACLF group and 542 d in the non-ACLF group.

Overall, 54 patients (10.76%) developed ACLF within 1 year after TIPS. Among those who developed ACLF within 1 year after TIPS, 23 (42.60%) developed ACLF within 28 d and 17 (31.50%) developed ACLF between day 29 and day 180. In the entire cohort, the incidence of ACLF within 28 d, 180 d, and 1 year after the TIPS procedure was 4.68%, 7.97%, and 10.76%, respectively. The baseline characteristics of the training cohort and the external validation cohort are shown in Supplementary Table 1.

Univariate and multivariate analysis for the incidence of ACLF within 1 year after the TIPS procedure. In total, 17 variables were included in a univariate regression analysis to investigate the predictors of ACLF within 1 year after TIPS. Variables with a P value of less than 0.10 in the univariate analysis were selected for multivariate logistic regression analysis. The model was built using a backward conditional method, which identified age [age ≥ 65 years, odds ratio (OR): 2.649, 95% confidence interval (95%CI): 1.263-5.558, P = 0.010] and TBil (TB: 17.1-34.2 umol/L, OR: 2.944, 95%CI: 1.151-7.528, P = 0.024; TB: 34.2-51.2 umol/L, OR: 11.632, 95%CI: 4.068-33.259, P < 0.001; TB ≥ 51.3 umol/L, OR: 28.746, 95%CI: 6.969-118.579, P < 0.001) as independent risk factors for ACLF within 1 year after TIPS procedure (Table 1).

Table 1 Univariate and multivariate logistic regression analyses of variables in predicting the incidence of acute-on-chronic liver failure within 1 year after transjugular intrahepatic portosystemic shunt in the training cohort.

Variables	OR comparison	Univariate analysis			Multivariate analysis
		OR	95%CI	P value	OR	95%CI	P value
Gender	Male vs female	0.922	0.471-1.802	0.811
Age	≥ 65 yr vs < 65 yr	2.086	1.074-4.055	0.030	2.649	1.263-5.558	0.010a
DM	Yes vs No	1.847	0.905-3.769	0.092
Portal vein thrombosis	Yes vs No	0.809	0.411-1.590	0.539
Splenectomy	Yes vs No	0.426	0.146-1.240	0.117
Etiology of cirrhosis	Viral hepatitis vs other	1.174	0.609-2.265	0.631
TIPS indication	EGVB vs refractory ascites	1.304	0.477-3.563	0.605
WBC	< 4.0 × 109/L vs ≥ 4.0 × 109/L	0.578	0.297-1.123	0.106
PLT	< 100 × 109/L vs ≥ 100 × 109/L	2.085	0.930-4.674	0.075
ALT	> 40 U/L vs ≤ 40 U/L	2.000	0.856-4.673	0.109
TBil	17.1-34.1 μmol/L vs < 17.1 μmol/L	3.064	1.212-7.742	0.018	2.944	1.151-7.528	0.024a
	34.2-51.2 μmol/L vs < 17.1 μmol/L	12.357	4.433-34.443	0	11.632	4.068-33.259	< 0.001a
	≥ 51.3 μmol/L vs < 17.1 μmol/L	24.714	6.342-96.315	0	28.746	6.969-118.579	< 0.001a
Alb	≥ 35.0 g/L vs < 35.0 g/L	1.651	0.758-3.596	0.206
Na	< 135 mmol/L vs ≥ 135 mmol/L	3.606	1.402-9.275	0.008	2.741	0.960-7.821	0.0601
Scr	≥ 133 μmol/L vs < 133 μmol/L	1.368	0.295-6.349	0.689
INR	< 1.5 vs ≥ 1.5	2.363	1.001-5.575	0.050
FIB	< 2.0 g/L vs ≥ 2.0 g/L	1.542	0.726-3.273	0.259
Stent diameter	≥ 8 mm vs < 8 mm	0.787	0.392-1.581	0.501
Puncture site	Right branch vs left branch	1.000	0.373-2.680	1.000
	Bifurcation vs left branch	1.067	0.437-2.602	0.887
	Trunk vs left branch	0.889	0.352-2.246	0.803

^aP < 0.05.

¹Included on the basis of a minimized Akaike Information Criterion value.

OR: Odd ratio; 95%CI: 95% confidence interval; DM: Diabetes mellitus; TIPS: Transjugular intrahepatic portosystemic shunt; EGVB: Esophagogastric variceal bleeding; WBC: White blood cell; PLT: Platelet; ALT: Alanine transaminase; TBil: Total bilirubin; Alb: Albumin; Na: Sodium; Scr: Serum creatinine; INR: International normalized ratio; FIB: Fibrinogen.

Establishment and validation of the nomogram model

Serum sodium level was included in the final prediction model based on the minimum AIC. Therefore, we developed a predictive model including three variables: Age, TBil, and serum sodium. We compared the area of the curve (AUC) of this model with other models using data from the entire cohort (Supplementary Table 2). Considering the simplicity and predictive power of the model, it was determined to be the optimal choice compared to other models. Figure 2 displays a predictive nomogram based on the three variables. In addition, the nomogram is available through a free browser-based online calculator at https://jyn1212.shinyapps.io/DynNomapp/. Using this calculator, the risk of developing ACLF within 1 year after TIPS can be estimated and displayed.

Figure 2 The nomogram model for predicting the incidence of acute-on-chronic liver failure within 1 year after transjugular intrahepatic portosystemic shunt. ACLF: Acute-on-chronic liver failure; TBil: Total bilirubin; Na: Sodium.

In the training cohort, the AUC of this prediction model was 0.800 (95%CI: 0.731-0.868) with a cutoff value of 0.112, corresponding to a sensitivity and specificity of 0.700 and 0.775, respectively (Figure 3A). We conducted a bootstrap internal validation procedure with 1000 bootstrap repetitions. The AUC was 0.774, indicating good discrimination power. The calibration curve showed that the predicted probability of ACLF based on the nomogram model aligned well with the actual probability (Figure 3C), with a Brier score of 0.084 and a Hosmer-Lemeshow test χ² = 3.950, P = 0.915. These findings suggest that the model has good calibration ability. In the external validation cohort, the AUC of the nomogram prediction model was 0.761 (95%CI: 0.607-0.914), and the cutoff value of the ROC curve was 0.288, corresponding to a sensitivity and specificity of 0.545 and 0.881, respectively (Figure 3B). These findings indicate that the model showed good discrimination in the external validation cohort. The calibration in the external validation cohort was also good, with a Brier score of 0.082 and a Hosmer-Lemeshow test χ² = 5.793, P = 0.760. The calibration plot is shown in Figure 3D.

Figure 3 Receiver operating characteristic curves and the calibration plots of the model for predicting the incidence of acute-on-chronic liver failure within 1 year after transjugular intrahepatic portosystemic shunt. A: In the training cohort, the area under the receiver operating characteristic curve (AUC) of the model was 0.800 [95% confidence interval (95%CI): 0.731-0.868]; B: In the validation cohort, the AUC of the model was 0.761 (95%CI: 0.607-0.914); C and D: The calibration plots in the training and validation cohorts. The predicted probability of the model was plotted on the x-axis; the actual probability was plotted on the y-axis. An ideal calibration plot is indicated by a 45° diagonal line. AUC: Area under the receiver operating characteristic curve; 95%CI: 95% confidence interval.

We also assigned scores to each variable based on the regression coefficients of the variables in the model and compiled a risk score (Table 2). The total score was 8, with a cutoff value of 4.5. In the entire cohort, the AUC of the risk score was 0.787 (95%CI: 0.729-0.845) (Figure 4), which was comparable to that of the nomogram model (P = 0.645) (Supplementary Table 2). The probability of ACLF for each score is shown in Supplementary Table 3. Patients were then stratified into three groups based on their scores: Low risk (total score of 3-4), medium risk (total score of 5-6), and high risk (total score of 7-8) group. In the whole cohort, the actual incidence of ACLF was 4.7% among patients with a total score of 3-4, 24.1% among patients with a total score of 5-6, and 50.0% among patients with a total score of 7-8.

Figure 4 Receiver operating characteristic curves of the prediction model, risk score, and four liver function scoring systems for the prediction of the incidence of acute-on-chronic liver failure within 1 year after transjugular intrahepatic portosystemic shunt in the entire cohort. MELDs: Model for end-stage liver disease score; Child-Pughs: Child-Pugh score; CLIF-C ADs: Chronic Liver Failure Consortium acute decompensation score; MELD-Nas: MELD-Na score; AUROC: Area under the receiver operating characteristic curve; 95%CI: 95% confidence interval.

Table 2 Risk score scale for prediction of the incidence of acute-on-chronic liver failure within 1 year after transjugular intrahepatic portosystemic shunt.

Variable	1 point	2 points	3 points	4 points
Age (yr)	< 65.0	≥ 65.0
TBil (μmol/L)	≤ 17.1	17.2-34.2	34.3-51.3	> 51.3
Na (mmol/L)	≥ 135.0	< 135.0

TBil: Total bilirubin; Na: Sodium.

The predictive ability of the proposed nomogram model was compared with that of four commonly used liver function scoring systems using ROC curve analysis (Figure 4) for the entire cohort of 502 patients. The discriminatory ability of the nomogram model had an AUC of 0.792 (95%CI: 0.730-0.853), which was superior to MELD score (AUC: 0.696, 95%CI: 0.623-0.769, P = 0.002), Child-Pugh score (AUC: 0.693, 95%CI: 0.611-0.776, P = 0.018), CLIF-C AD score (AUC: 0.659, 95%CI: 0.577-0.741, P = 0.002), and MELD-Na score (AUC: 0.731, 95%CI: 0.662-0.801, P = 0.022).

Survival analysis

During the follow-up period, 34 deaths occurred in the ACLF group, of which 27 were associated with liver disease. The cumulative survival rates at 28 d, 180 d, and 1 year were 88.7%, 64.2%, and 38.0%, respectively. The Kaplan-Meier survival curve showed that the cumulative survival rate in the ACLF group was significantly lower than that in the non-ACLF group (Figure 5A). Furthermore, the incidence of liver disease-related deaths was significantly higher in the ACLF group than in the non-ACLF group (both P values < 0.001) (Figure 5B), suggesting that the incidence of ACLF was associated with a poor prognosis. Regarding postoperative complications, the cumulative postoperative rebleeding rate was significantly higher in the ACLF group than in the non-ACLF group (P < 0.001) (Figure 5C), but there was no statistically significant difference in the incidence of postoperative stent stenosis between the two groups (Figure 5D).

Figure 5 The prognosis between the acute-on-chronic liver failure group and the non-acute-on-chronic liver failure group. A: The Kaplan-Meier curve shows the probability of survival; B: The Kaplan-Meier curve shows the liver disease-related mortality rates; C: The Kaplan-Meier curve shows the rebleeding rates; D: The Kaplan-Meier curve shows the probability of stent stenosis. ACLF: Acute-on-chronic liver failure.

DISCUSSION

ACLF is an acute deterioration of chronic liver disease characterized by high short-term mortality. Early diagnosis and treatment of potential precipitating events are crucial in preventing ACLF[9]. These precipitating events include hepatitis B infection, acute viral hepatitis, alcohol consumption, hepatotoxic drugs, and acute variceal bleeding[1]. Based on the CANONIC study in 2011, the TIPS procedure, as an invasive transhepatic treatment strategy, may increase the risk of ACLF[3]. After the TIPS procedure, the portal blood flow enters directly into the systemic circulation without passing through the liver, reducing blood flow to the liver. In addition, the puncture process causes direct mechanical damage to the liver. In addition, the stent compresses the surrounding liver tissue and affects bile excretion, impairing liver function.

Considering these mechanisms, the main purpose of this study was to establish a predictive model based on the risk factors of ACLF after TIPS to help clinicians select appropriate patients for TIPS procedures and reduce the risk of ACLF after TIPS. To date, several studies have been conducted to predict ACLF, but none of them were conducted on post-TIPS patients; thus, their results are not suitable for predicting ACLF after TIPS. Xiao et al[10] proposed the first prediction model of APASL-ACLF based on outpatients with compensated cirrhosis. The independent predictors included hazardous drinking behaviors, age, baseline Alb, INR, TBil, creatinine, and hemoglobin. The incidence of ACLF 1 year after the baseline assessment was 1.36%. Similarly, Yu et al[11] developed a prediction model for ACLF in patients with chronic hepatitis B and severe acute exacerbation. Yu et al[11] proposed a PATA model comprising prothrombin time, age, TBil, and ALT. The above-mentioned studies together confirm the predictive ability of age and TBil for ACLF.

Similar to previous studies, in our study, TBil and age were found to be independent predictors of ACLF after TIPS. TBil is often used to assess the severity of liver impairment, and higher levels of TBil indicate more severe hepatic damage. In addition, TBil is an important parameter in several liver function evaluation models, such as the Child-Pugh score and MELD score. Our study confirmed that TBil can also be used to predict the incidence of ACLF after TIPS, and patients with high TBil levels before TIPS are more likely to develop ACLF after surgery. Age is also an independent risk factor for ACLF after TIPS, which may be related to the aging of the liver. With aging, the liver undergoes significant changes at the tissue-organ and cellular levels, resulting in decreased reserve and regeneration capacity[12]. Therefore, elderly patients are more likely to exhibit abnormal liver function under internal or external stresses. Hyponatremia is common in ACLF and predicts a poor prognosis[13,14]. A previous study has shown that a lower level of serum sodium is an independent risk factor for developing severe ACLF in patients with cirrhosis and bacterial infection[15].

Previous studies have confirmed that a severe systemic inflammatory state is the major driver of extensive tissue and organ damage in patients with acute decompensated cirrhosis who develop ACLF[16]. Compared to cirrhotic patients without ACLF, patients with ACLF have higher WBC counts and plasma levels of C-reactive protein[3]. However, our study did not find a correlation between WBC count and the development of ACLF. This could be because our data were the baseline data of patients, which were collected before TIPS, and patients did not exhibit a severe systemic inflammatory state before developing ACLF.

Based on the minimum AIC, we developed a risk prediction model that can predict the incidence of ACLF within 1 year after the TIPS procedure. The model showed good discrimination and calibration in both the training cohort and the external validation cohort. Furthermore, the discriminatory ability of the nomogram model was superior to that of the MELD score, Child-Pugh score, CLIF-C AD score, and MELD-Na score. Based on this model, we also constructed other models, and after comparison, we found that there was no significant improvement in the AUC. Therefore, we finally chose the first model as it was simpler. To facilitate its clinical application, we also transformed the nomogram model into a risk score. The predictive ability of the risk score was comparable to that of the nomogram model. Patients with a score of 3-4 are at low risk of ACLF, and for such patients, TIPS can be actively performed in the presence of an indication for TIPS. In contrast, patients with a score of 7-8 are at higher risk of ACLF, and clinicians should be cautious and adopt active treatment to reduce patients’ scores before the TIPS procedure. Other treatments or liver transplantation should be considered if patients’ risk scores remain high. For patients with a score of 5-6, we should assess the benefit-risk of TIPS and fully consider patients’ willingness to undergo the procedure.

Limitations: First, selection bias may exist as this study was a retrospective study with a limited sample size. Additionally, a notable number of patients were lost to follow-up, indicating that the number of patients was possibly underestimated. However, the incidence of ACLF was comparable between the two cohorts. Second, as our findings were based on patients with APASL-ACLF, whether the model can be applied to other populations needs more studies.

CONCLUSION

In conclusion, the incidence of ACLF within 1 year after TIPS was independently associated with age and TBil. Our model and risk score can help predict the incidence of ACLF after TIPS, providing a reference for clinical decision-making. Prospective validation in larger cohorts is needed to assess the generalizability of our findings.

ACKNOWLEDGEMENTS

The authors would like to thank all the study participants for their voluntary participation and extend special thanks to Tai-shun Li for his guidance in biostatistics.