A modified model for predicting mortality after transjugular intrahepatic portosystemic shunt: A multicentre study

The transjugular intrahepatic portosystemic shunt has controversial survival benefits; thus, patient screening should be performed preoperatively. In this study, we aimed to develop a model to predict post‐transjugular intrahepatic portosystemic shunt mortality to aid clinical decision making.


Decompensated cirrhosis can lead to ascites and gastrointestinal
bleeding, for which transjugular intrahepatic portosystemic shunt (TIPS) has been used as a treatment. 1Among all available treatments, TIPS is the only minimal invasive treatment that decompresses the portal venous system.6][7] Therefore, the guideline emphasizes the importance of patient selection before TIPS. 8e reasonable solution for patient selection is to construct a preoperative mortality prediction model to identify patients who can truly obtain survival benefits from TIPS.0][11][12] However, to enhance performance, some aspects still need attention.First, the models should be developed specifically for patients undergoing TIPS, rather than relying on models built for patients with cirrhosis or hepatocellular carcinoma (HCC). 13ditionally, external validation should be emphasized upon to avoid overfitting and ensure model performance among different populations. 14,15Finally, most models lack attention to hypersplenism, a crucial manifestation of portal hypertension that can provide additional information for patient prediction. 16,17 establishing a specific model of TIPS, controlling the influence of overfitting, as well as extracting additional information from hypersplenism, in this study, we aimed to construct a refined model for the prediction of post-TIPS mortality and, thus, improve the predictive performance and robustness across different populations.The model will assist in patient selection for TIPS.Moreover, we also compared the efficacy of our model with those of previous models.

| Patients
This study utilized multicentre data from five hospitals located in China, namely, Nanfang Hospital (NF), Shenzhen People's Hospital (SZ), Anhui Provincial Hospital (AH), the Third Affiliated Hospital of Sun Yat-sen University (SY) and Zhuhai People's Hospital (ZH).rebleeding after the initial pharmacological and endoscopic therapy (early TIPS); (2) had HCC that did not meet Milan Criteria; 18 (3) had recurrent hepatic encephalopathy before TIPS; and (4) underwent liver transplantation candidates within 1 year of TIPS.
Between January 2012 and July 2021, patients who underwent TIPS due to refractory ascites or variceal rebleeding were screened for eligibility.Ultimately, after screening, a total of 811 patients were included in the study (Figure 1).
The study protocol was approved by the ethical review committee of Zhuhai People's Hospital (approval number: 2021[67]).The need for informed consent was waived because patient data were collected retrospectively.All patient data were anonymized during analysis.

| Preoperative treatment
According to the guidelines, 2 the following preoperative treatments were performed when necessary: (1) primary antibiotic prophylaxis was administered to the patient pre-TIPS if prolonged surgery was anticipated; (2) non-selective beta-blockers or carvedilol were used to prevent recurrent variceal bleeding; and (3) preoperative largevolume paracentesis was performed with albumin supplementation for patients with ascites.

| Candidate factors
We collected data on 19 variables, categorized into five groups: epidemiology, liver function, renal function, coagulation function and hypersplenism.All laboratory test results were obtained within 1 week prior to TIPS.These variables are listed as follows.We included Child-Pugh score as a candidate factor because it was the most commonly used score for predicting the prognosis of liver disease.To avoid collinearity, we did not include the indicators used to calculate the Child-Pugh score, such as prothrombin time and total bilirubin, as candidate variables.

| Follow-up and outcome
During the follow-up period, patients were initially seen in the outpatient department on a weekly basis for the first month and subsequently every 4 weeks.The follow-up visits consisted of telephone interviews, outpatient visits or hospital visits.
The outcome in this study was 1-year mortality after TIPS.By the end date in October 2022, all included patients had completed at least 1 year of follow-up, allowing for analysis of the outcome.S2).Additionally, overt hepatic encephalopathy occurred in 200 postoperative patients, and 18 died as a result (25.00%) (Table S3); rebleeding of varying degrees occurred in 144 individuals, and four ultimately died (5.56%) (Table S3).

| Model construction
The model construction process involved the use of univariable logistic regression for preliminary screening, followed by multivariate regression to identify the most significant prognostic factors  2A) and calibration (Figure 2B).The decision curve was also displayed (Figure 2C).
Based on multivariate regression coefficients, the equation corresponding to Model MT was derived as follows:

| Patient risk stratification
The risk ratio of the entire study population was .08.To distinguish low-, medium-, high-and extremely high-risk groups, we used .5, 1 and 3 times the risk ratio as cut-offs.These corresponded to scores of ≤160, 160-  For model ease of use, we created an applet to calculate each patient's score and grouping patient (F).
103, 86, 41 and 32 patients in the validation data set were classified as low-, medium-, high-and extremely high-risk subgroups, respectively.The mortality of each subgroup was .63%versus 9.09% versus 15.07%versus 36.00% in the training set and 1.94% versus 8.14% versus 12.20% versus 53.13% in the validation set (Figure 3D,E).A statistically significant difference was found in mortality between all groups (all p < .001)(Supplementary Table S5).
Accordingly, we developed an applet for Model MT .The applet was tested on a Windows system and is used as follows: unzip the file and enter the corresponding clinical data in the input box (Figure 3F, Download URL: https:// github.com/ FuSir ui123/ modelfor-predi cting -morta lity-after -TIPS).

| DISCUSS ION
In light of the controversy surrounding the survival benefit of TIPS, it is crucial to perform preoperative patient selection.Model MT demonstrates significantly better performance in predicting post-TIPS mortality by incorporating information on hypersplenism and using an appropriate external validation data set.Our modified TIPSspecific model can be used to ensure survival benefits from TIPS (Figure 3).
Ascites and gastrointestinal bleeding are the two most crucial complications of portal hypertension.2][3] However, these treatments lead to a relatively higher recurrence rate due to the inability to relieve portal hypertension. 7,21 contrast, TIPS, despite its ability to control portal hypertension, and with advantages of low recurrence rate, minimal trauma and low cost, 22 is only used as an alternative therapy when first-line therapies are ineffective.One of the major challenges in utilizing TIPS is the uncertainty of whether or which patients will achieve survival benefit.Several studies, including meta-analyses, have discussed whether TIPS provides a survival advantage for patients with gastrointestinal bleeding; however, the conclusions reached were controversial. 23,24is condition is also prevalent or even more complicated in studies of ascites. 25,26Thus, careful patient selection should be performed before TIPS treatment, and a preoperative prediction model could assist doctors in making better treatment decisions.
The development of TIPS prediction models has gone through several stages, from borrowing from other models to building dedicated models.With the evolution of TIPS technology and changes in the target population, prediction models are continuously being developed.FIPS and MOTS have been created recently.liver cancer in the initial screening process.However, patients with HCC meeting the Milan criteria are not an absolute contraindication to TIPS 1,27 ; to expand the applicability of the population and match real-world data, rather than excluding these patients, we chose to include patients with HCC as a candidate factor for analysis.
Meanwhile, our subgroup analyses confirmed the stable performance of Model MT in both populations with and without HCC.Further, we also included APTT, which is the most frequently conducted test in coagulation medicine. 28APTT is included alongside INR, probably because it can complementarily reflect the endogenous coagulation system. 29Our results showed that IBIL has a higher predictive value than DBIL, although both reflect liver function.The reason for this may be because high IBIL is typically caused by excessive haemoglobin metabolism, 30  Compared with leukopenia and thrombocytopenia, anaemia can reflect hypersplenism well and indicate the prognosis of patients from more angles, such as malnutrition or chronic gastrointestinal bleeding. 32sed on our results, clinicians can classify patients into low-, medium-, high-and extremely high-risk subgroups by their scores.
For patients in the low-risk group, we believe that TIPS is feasible.Patients in the medium-risk group can also be considered for TIPS but should be followed up more closely.For patients in the high-risk group, TIPS should be avoided.Lastly, TIPS is not recommended for patients in the extremely high-risk group.However, the above classification is only a reference.In real-world practice, the actual decision should be made by the clinician on a case-bycase basis.
The limitations of this study are as follows.First, although we included multiple centres and set an external validation data set, further studies with an additional prospective cohort may still be required.Second, since an 8-mm stent is commonly used in most Chinese patients, whether our conclusions can be applied to 10-mm stents remain to be testified.Third, in this study, we used only serum parameters as indicators for hypersplenism; however, in the future, more diverse parameters measuring the change in morphology of the spleen may be added.Lastly, due to the small number of patients with ascites included in our study, the performance of our model in the ascites group was not comparable in the variceal bleeding group.
We will expand the number of cases of patients with ascites in the future and conduct additional studies.
In conclusion, we have constructed a model that is capable of accurately predicting post-TIPS mortality.The implementation of this model will enable clinicians to identify patients who can truly receive benefit from TIPS and subsequently improve their chances of survival.
The study's inclusion criteria were as follows: (1) diagnosed with portal hypertension based on clinical presentation and imaging; (2) experienced at least one incidence of variceal rebleeding or refractory ascites after first-line treatments; (3) underwent regular follow-up for at least 1 year; (4) aged 18 years or older; (5) had not undergone any previous liver-related surgical procedure; (6) underwent TIPS for the first time; (7) had no severe cardiac (stage C or D or a documented ejection fraction <50%) or severe pulmonary hypertension (mean pulmonary artery pressure >45 mmHg); and (8) had a Child-Pugh score <13 points and no liver failure.The exclusion criteria were as follows: (1) underwent TIPS to prevent failure or Results: Model MT demonstrated a satisfying predictive efficiency in both discrimination and calibration, with an area under the curve of .875 in the training set and .852 in the validation set.Compared to previous models (ALBI, BILI-PLT, MELD-Na, MOTS, FIPS, MELD, CLIF-C AD), Model MT showed superior performance in discrimination by statistical difference in the Delong test, net reclassification improvement and integrated discrimination improvement (all p < .050).Similar results were observed in calibration.Low-, medium-, high-and extremely high-risk groups were defined by scores of ≤160, 160-180, 180-200 and >200, respectively.To facilitate future clinical application, we also built an applet for Model MT .Conclusions: We successfully developed a predictive model with improved performance to assist in decision making for transjugular intrahepatic portosystemic shunt according to survival benefits.K E Y W O R D S cirrhosis, clinical decision making, hypersplenism, portal hypertension, post-TIPS mortality Lay Summary The survival benefits after transjugular intrahepatic portosystemic shunt (TIPS) are controversial; thus, patient screening should be performed preoperatively.In this study, we developed a model to predict post-TIPS mortality.Thereafter, a modified prediction model of post-TIPS mortality (Model MT ) was built.Our model demonstrated better performance with respect to discrimination and calibration compared with previous models.With the help of our high-accuracy model, clinicians can ensure a better chance of survival for patients after TIPS.
All TIPS procedures were performed by interventionalists with more than 10 years of experience.The TIPS procedures were performed as follows: (1) after general anaesthesia, the right internal jugular vein was localized under ultrasound guidance and catheterization was performed toward the hepatic vein; (2) one of the portal branches was punctured from the right hepatic vein, and portal venogram was performed; (3) after pre-dilatation of the puncture channel, a polytetrafluoroethylene-covered stent (8 mm, including Fluency stent (Bard Peripheral Vascular, USA) or Viatorr stent (W.L. Gore & Associates, USA)) was implanted, and the stent was re-expanded to ensure full expansion; (4) after stent insertion, another portal venogram was performed; and (5) finally, the portosystemic pressure gradient (PSPG) was measured.The TIPS procedures were considered successful if the PSPG was <12 mmHg or reduced to at least 20% of baseline value.All patients remained hospitalized after TIPS treatment until they met discharge criteria, such as normalization or improvement in liver function.
(i) Epidemiological indicators: age, sex, aetiology of cirrhosis, indication for TIPS, and the presence of HCC; (ii) Liver function indicators: alanine aminotransferase (ALT), aspartate aminotransferase (AST), direct bilirubin (DBIL), indirect bilirubin (IBIL) and Child-Pugh score; (iii) ReRenal function indicators: sodium, urea nitrogen and serum creatinine; (iv) Coagulation indicators: international normalized ratio (INR) and activated partial thromboplastin time (APTT); (v) Hypersplenism indicators: white blood cells (WBC), red blood cells (RBC), haemoglobin and platelets.F I G U R E 1 The inclusion and exclusion flowchart showing patient screening for this study.We screened 983 patients from five hospitals.After evaluating the inclusion and exclusion criteria, 811 patients were divided into the training (N = 549) and external validation (N = 262) data sets.| 475ZHAO et al.
External validation was performed to confirm the generalization ability of the prediction model in this study.The training data set included patients from three centres (NF, SZ and AH), whereas the remaining patients from the other two centres (SY and ZH) were used as the validation data set.The differences in the clinical characteristics between patients in different data sets were compared using the independent-samples t-test or Mann-Whitney U-test for continuous variables and Fisher's exact test or Chi-squared test for categorical variables, as appropriate.First, logistic regression was used to establish predictive models on the training data sets.The backward selection procedure was implemented to select the best subset of risk factors to avoid overfitting.Second, the performance of the predictive model was expressed using discrimination and calibration, with its net benefit displayed by decision curve analysis (DCA).Third, to test whether our model outperformed previous models, their discrimination and calibration were compared.During the process, discrimination was compared by areas under the curve (AUCs), along with Delong test, net reclassification improvement (NRI) and integrated discrimination improvement (IDI), whereas calibration was compared by calibration curve.Fourth, after testing the robustness of our model across different populations by subgroup analysis, we divided the risk score into four levels and compared the mortality among different levels using the Fisher's exact test or Chi-squared test.Finally, besides the equation corresponding to the model, we constructed an applet to facilitate future clinical application.All statistical analyses and graphical presentations were performed using R software version 4.2.1 (http:// www.r-proje ct.org/ ).Statistical significance was set at two-sided p < .050.The report of our study strictly followed the Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis (TRIPOD) statement.

TA B L E 1
Baseline demographics of patients in the training and validation data sets.

F I G U R E 2
The performance and applet for ModelMT  .Model MT achieved an area under the curve of .875 in the training data set and .852 in the validation data set (A).The calibration values are expressed using calibration curves (B).The decision curves are displayed to evaluate the net benefit of model-assisted decision-making (C).Low-, medium-, high-and extremely high-risk groups are defined by scores of ≤160, 160-180, 180-200 and >200, respectively (D,E).

F I G U R E 3
which is often caused by hypersplenism in patients with liver disease.Therefore, IBIL can both indicate liver function and emphasize the importance of hypersplenism, which is Study design.This study was designed to develop a model to predict post-TIPS mortality (A).For candidate factors, we collected five categories of indicators: epidemiology, liver function, renal function, coagulation function and hypersplenism (B).For model comparisons, Model MT was compared with models proposed in previous literature in terms of AUC, ROC, NRI, IDI, calibration and decision curve analysis (C).We grouped the population by risk level (D).AUC, area under the curve; IDI, integrated discrimination improvement; NRI, net reclassification improvement; ROC, receiver operator characteristic; TIPS, transjugular intrahepatic portosystemic shunt.a consequence of portal hypertension and an indicator of more advanced liver disease.31Based on these reasons, it may have a more direct relationship with mortality after TIPS.In addition to the traditional indicators, we included indicators related to hypersplenism, such as red blood cells, white blood cells, haemoglobin and platelets.31Red blood cells, which represent anaemia, are included in the final model as one of the indicators.