A creatinine-based model to predict recurrent bleeding after modi ed percutaneous transhepatic variceal embolization in cirrhotic patients

Zhen Li (  lzlyct620@163.com ) Kun Ji Zhengzhou University First A liated Hospital Han-Long Zhu Nanjing Medical University Second A liated Hospital Si Zhao Nanjing Medical University Second A liated Hospital Xin Li Zhengzhou University First A liated Hospital Peng-Chao Zhan Zhengzhou University First A liated Hospital Yang Shi Zhengzhou University First A liated Hospital Shu-Wen Ye Zhengzhou University First A liated Hospital Bing-Can Xie Zhengzhou University First A liated Hospital Yu-Yuan Zhang Zhengzhou University First A liated Hospital Peng Yu Zhengzhou University First A liated Hospital Zhi-Gang Ren Zhengzhou University First A liated Hospital Juan Ding Zhengzhou University First A liated Hospital Xin-Wei Han Zhengzhou University First A liated Hospital

have been widely recognized as factors predictive of rebleeding after initial EVB 9 . However, the Child-Pugh classi cation only divides patients into high, intermediate and low risk but fails to quantify an expected probability of rebleeding; and the MELD calculation is too complicated to easily obtain the score. To the best of knowledge, this study is the rst attempt to generate a predictive nomogram for the unselected cirrhotic patients with EVB treated by PTVE, which can simply predict the speci c probability of rebleeding.
The present study aimed to identify the risk factors and to establish a model to predict 3-month rebleeding for cirrhotic patients with EVB treated by modi ed PTVE as a rescue treatment after failing endoscopic therapy. Additionally, high-risk patients for rebleeding was discerned based on the nomogram to consider administering TIPS or liver transplantation earlier and intensive follow-up plan.

Patients
Patients who were diagnosed with EVB and received modi ed PTVE in our hospital from January 2015 to March 2020 were enrolled in this study. The inclusion criteria were as follows: (1) diagnosis of liver cirrhosis by clinical examination and imaging techniques, including ultrasounds, CT, or MRI; (2) gastroscopy con rmed that esophageal varices were the only potential source of bleeding; (3) PTVE was conducted as a rescue therapy in patients who had uncontrolled severe bleeding or recurrent episode of bleeding from varices after pharmacological and endoscopic therapy, such as sclerotherapy and band ligation; (4) patients without catheterizable gastrorenal shunts who could not be treated by balloonoccluded retrograde transvenous obliteration. The exclusion criteria were as follows: (1) severe hypertension, coronary heart disease, cardiopulmonary insu ciency, or chronic renal insu ciency; (2) a history of TIPS creation, pericardial devascularization or splenectomy for varices; (3) patients who died during the current hospitalization; (4) incomplete data on medical records or follow-up. The owchart of the enrolled patients is shown in Fig. 1 On admission, all of the patients were treated by an infusion of vasoactive drugs for 72 h and prophylactic antibiotics for 7 days, and endoscopy within the rst 24 h of hospitalization. PTVE with or without partial splenic embolization (PSE) was performed as a rescue therapy in patients who failed the pharmacological and endoscopic treatment. The procedures were conducted by two interventional radiologists with 15 and 25 years of experience for all cases.
After a percutaneous transhepatic puncture of a branch of the portal vein under digital subtraction angiographic guidance (Artis zeego, Siemens), a 5-F Cobra catheter (Cook) was introduced into the portal venous system and a splenoportography was conducted to evaluate the location of the index varices, the feeding vessels, draining veins and the possible presence of gastrorenal shunt. Then the catheter was advanced into each of main feeding vessel (e.g., the left gastric vein, short gastric vein, or coronary gastric vein) to embolize the vascular trunk using coils (3-10 mm × 5-12 cm; Cook) or microcoils (2-3 mm × 2-3 cm; Cook). Subsequently, cyanoacrylate was slowly injected to occlude the vascular bed until angiography veri ed that the blood ow in the varices were completely obstructed. Splenoportography was repeated to assess the extent of variceal obliteration. If other feeding veins were available, the above steps would be repeated until the blood ow in the varices ceased absolutely. Ultimately, the catheter was withdrawn and the puncture tract was embolized with microcoils.
PSE was performed 5-7 days after PTVE in patients who were willing to receive PSE. Brie y, using a transfemoral approach, splenic arterial angiography was implemented using a 5-F RH catheter (Cook) to demonstrate the distribution of splenic arteries. The catheter was inserted into the middle and lower brands of the splenic arteries, and 350-560 µm of Polyvinyl alcohol particles (ALICON Pharm SCI & TECH) mixed with contrast media were carefully injected into the splenic arteries through the catheter. Splenic arterial angiography was repeated to estimate the degree of splenic infarction which was limited to approximately 50-70% of the original splenic volume 10 .

Data Collection And Follow Up
All of the patients underwent complete medical assessment at admission, including collection of demographic information, medical history, physical examination, clinical symptoms, ascites, concomitant HCC, degree of splenomegaly, encephalopathy and portal vein thrombosis. Blood for laboratory testing [complete blood count, international normalized ratio (INR), total bilirubin, albumin, aspartate aminotransferase (AST) and serum creatinine] was also collected. The Child-Pugh and MELD scores 11 were calculated from data recorded upon patient admission.
Rebleeding was de ned according to the Baveno criteria 12 as recurrent hemorrhage proven by new melena or hematemesis, requirement for > 2 units of packed red blood cell in a 24 h period and hemodynamic instability after a 24 h period of stable vital signs and hemoglobin after PTVE. Time to rebleeding was referred to the duration from the eradication of varices hemorrhage to recurrent bleeding.
Patients who suffered from recurrent bleeding within 3 months after PTVE were classi ed as the rebleeding group, and remaining patients as the non-rebleeding group. The patients were followed up in the outpatient clinic or by telephone at 1, 3, 6 and 12 months or until death, loss to follow-up, or the cutoff date for data analysis (June 2020).

Nomogram Construction And Validation
The univariate and multivariate logistic regression analyses were conducted to screen out the independent risk factors that signi cantly affected rebleeding. Based on the identi ed independent risk factors, a nomogram was constructed to predict the probability of 3-month rebleeding after PTVE.
The nomogram's performance was evaluated using the concordance index (C-index), receiver operating characteristic (ROC) curve, calibration curve and decision curve analysis (DCA). The C-index is de ned as the proportion of all evaluable patient pairs whose predictions are consistent with the actual results 13 .
The nomogram was subjected to bootstrapping validation (1,000 bootstrap resamples) to compute a relatively corrected C-index 14 . The nomogram's discrimination and clinical application value were measured utilizing the area under ROC curve (AUC) and DCA 15 to certify whether the nomogram is superior to the MELD and Child-Pugh models. The calibration of the nomogram was assessed by a calibration curve that compares nomogram-predicted and actually-observed estimates of rebleeding probability. Kaplan-Meier (K-M) curve was constructed to analyze the difference in rebleeding between the high-and low-risk groups based on the nomogram.

Statistical Analysis
All of the statistical analyses were conducted employing the SPSS software (version 21, IBM Corporation, Armonk, NY, USA) and the programming language R (version 3.6.2) for Windows. The Student's t-test and the Chi-square test, for the continuous and categorical variables respectively, were used to evaluate the association between the rebleeding and variables, by comparing the rebleeding and non-rebleeding groups. Subsequently, the variables with a P-value below 0.05 in univariate analysis were selected for entering multivariate logistic regression analysis to ascertain the independent risk factors for rebleeding adopting the forward stepwise selection method. The optimized cut-off values for equally important sensitivity and speci city of INR, creatinine, aspartate aminotransferase, albumin, bilirubin and the total scores calculated from the nomogram were determined using ROC curve. The nomogram, C-index, ROC curve, calibration curve, DCA curve and K-M curve were generated in R with packages "rms", "Hmisc", "ROCR", "Survminer", "survival" and "rmda". Statistical signi cance was set as P 0.05 in a two-sided test.

Patient Characteristics
Among 122 patients, more than two-thirds had virally induced cirrhosis. HCC and portal vein thrombosis were present in 22.1% and 15.6% of the patients, respectively, and PSE was carried out in 48.4% of the patients. Hematemesis and/or melena were the main clinical symptoms of hemorrhage in all cases. There were 32 patients in the rebleeding group and 90 in the non-rebleeding group. The median follow-up time was 6.7 months (3-12 months). More details on patient characteristics are shown in Table 1. A total of were 32 (26.2%) patients experienced adverse effects following PTVE or PSE including transient upper abdominal pain (n = 23), low fever (n = 6), and both (n = 3), which were minor and alleviated by pharmacologic therapy.

Risk Factors For Rebleeding
Rebleeding occurred in 32 patients (26.2%) within the rst 3 months after cessation of EVB. The recurrent bleeding was controlled by further pharmacological treatment (n = 6), endoscopic therapy (n = 13) and TIPS (n = 4), while the remaining 9 patients died of massive bleeding (n = 7) and hepatic or multiorgan failure (n = 2). The cumulative probabilities of presence of variceal rebleeding at 1, 2, 6 and 12 months were 9.0%, 21.3%, 37.4% and 48.6%, respectively. As uncovered by the univariable analysis, 12 variables of gender, the history of endoscopic therapy, INR, creatinine, AST, albumin, bilirubin, ascites, PSE, MELD score, and Child-Pugh score/classi cation were markedly correlated with the risk of rebleeding (all p < 0.05). Of note, the Child-Pugh classi cation and MELD were not further analyzed due to the dependence from Child-Pugh score and the complicated calculation, respectively. In multivariable logistic analysis, a history of endoscopic therapy (OR = 4.125; 95% CI: 1.208-14.084; P = 0.024), Child-Pugh score (OR = 1.792; 95% CI: 1.332-2.411; P 0.001), non-PSE (OR = 0.258; 95% CI: 0.085-0.777; P = 0.016) and a creatinine ≥ 78 µmol/L (OR = 7.960; 95% CI: 2.492-25.425; P 0.001) were independent risk factors for rebleeding within 3 months (Table 2). A nomogram was developed based on the independent predictors (Fig. 2). In the nomogram, each category of the variables is assigned a score on the Points scale. The sum of these scores is located on the Total Points scale and a line is drawn downward to determine the speci c probability of 3-month rebleeding. The score assignment for variables included in the nomogram is summarized in Table 3. The nomogram's C-index was 0.85 (95% CI: 0.76-0.94) for the cohort, and was con rmed to be 0.83 through bootstrapping validation. Meanwhile, the calibration curve for the probability of rebleeding at 3 months exhibited an excellent agreement between the actual and predicted outcomes (Fig. 3A). The nomogram had higher AUC (0.850 vs 0.723 and 0.767) and better clinical applicability than the Child-Pugh and MELD models ( Fig. 3B-C). In addition, the patients were divided into two sets depending on the nomogram scores: low-risk group (0-95 points) and high-risk group (96-219 points). As displayed in the K-M curves, the patients at high risk were more likely to experience rebleeding (P < 0.001). (Fig. 3D).

Development Of Webserver
To facilitate the nomogram's clinical use, an online version is provided at https://jikun.shinyapps.io/rebleeding_of_ev (Fig. 4), which can not only predict personalized rebleeding, but also avoid the manual measurement errors.

Discussion
Conventional PTVE was introduced in the 1970s for the treatment of EVB, but it has not attracted su cient attention because of the high rebleeding rate. However, modi ed PTVE with cyanoacrylate started a new chapter in the management of EVB, which has the high hemostasis and low rebleeding rate due to a more extensive obliteration area and permanent embolization of variceal veins than conventional PTVE 16,17 . A study which compared the modi ed PTVE and TIPS revealed that variceal rebleeding rates were comparable (30.2% in TIPS group vs. 20.8% in PTVE group), and PTVE offered lower incidence of encephalopathy than TIPS 18 , and the similar result was a rmed by Zhang et al 16 .
In the present study, 122 EVB patients who failed endoscopic therapy underwent modi ed PTVE with cyanoacrylate as rescue therapy. Because variceal embolization of PTVE only occludes portosystemic shunts but fails to eliminate cirrhosis, new esophagogastric varices are then established as a result of worsening portal hypertension 19 . Our research showed that 32 patients (26.2%) developed variceal rebleeding within 3 months after the initial cessation of hemorrhage, which is similar to a 6-week rebleeding rate of 20.8% after PTVE reported by Zhao et al 7 .
Most previous studies regarded 6 weeks as an interval for rebleeding according to the Baveno criteria, and excluded seriously ill patients with advanced cirrhosis, HCC and portal thrombosis 12,20 . In our research, a period of 3 months for rebleeding was chosen in view of studies regarding longer-term e cacy of PTVE remain lacking. Further, average waiting time on a list for liver transplantation is approximately 3 months 21 . Risk factors such as advanced cirrhosis, HCC, and portal thrombosis may result in a continuous increase in portal hypertension, making patients more susceptible to early rebleeding 22 . Therefore, we analyzed a group of unselected cirrhotic patients to quantify the risk of rebleeding to identify which patients may need early application of TIPS or liver transplantation and intensive follow-up plan.
Our nding revealed that the rebleeding was noticeably related to the history of endoscopic therapy, Child-Pugh score, PSE and creatinine, consistent with the previous reports 3, 17 . The Child-Pugh classi cation only strati es risk for high, intermediate and low levels, so we selected the Child-Pugh score to quantify an expected probability of rebleeding. The Child-Pugh score is a vital factor re ecting the severity of liver disease in patients with cirrhosis, with every 1-point increase in the Child-Pugh score conferring an 79% increased risk of rebleeding at 3 months as we discovered. The present study also disclosed that patients with a history of endoscopic therapy were more likely to suffer from rebleeding within 3 months after PTVE. Coincidentally, a retrospective study conducted by Lee et al 23 found that 58.3% patients had early recurrent hemorrhage after banding ligation in cirrhotic patients as well. This relationship could be explained by the extensive mucosal injury surface area and post-banding ulcers caused by the endoscopic therapy 23 . This makes sense since a patient with a history of endoscopic therapy, as a signal of previous EVB, would be at a higher risk of repeated bleeding in such a setting.
Acute renal failure is a severe complication of cirrhosis and a possible harbinger of death 24 , and serum creatinine is a sensitive marker of renal function which acts as a key component of the MELD model. As exhibited in our study, a creatinine ≥ 78 µmol/L is an independent predictor for rebleeding. Augustin S et al proposed creatinine as a crucial marker to identify patients at high risk after an acute variceal hemorrhage 25 . It is reported that 24-55% of patients with end-stage renal disease have increased bleeding complications attributing to the inhibitory in uence of uremia on platelet function 26 . It has been demonstrated that PSE appears to be e cacious in reducing episodes of variceal bleeding by reducing the increased portal pressure, as well as improving leukocytopenia and thrombocytopenia caused by hypersplenism 10,27 . In accordance with a controlled study 8 , our ndings indicated that rebleeding occurred more frequently in those treated by only PTVE than a combination of PSE and PTVE (38.1% vs 13.6%, p < 0.05), and no severe complication was observed because we achieved a limited embolization part of middle or lower pole of the spleen, and a limited ratio of 50-70% of the original splenic volume. Besides, a meta-analysis also demonstrated the dramatic superiority of PSE in preventing recurrent variceal hemorrhage and prolonging overall survival 28 .
To date, only a few studies have identi ed the value of risk factors in predicting rebleeding after cessation of initial EVB in cirrhotic patients treated by PTVE 7,8 . Nevertheless, those studies failed to establish a simple model that could conveniently and accurately predict the probability of rebleeding after PTVE. To the best of our knowledge, this study was the rst attempt to develop a predictive nomogram based on the unselected cirrhotic patients with EVB treated by PTVE. Although the MELD model has been proved to be superior to the Child-Pugh model as an index of liver disease severity 11 , the MELD calculation is complex to obtain a score in clinical practice. Therefore, we did not take it into account for the nomogram construction. The nomogram exhibited an accurate prediction for rebleeding with a high C-index of 0.85, and had superior predictive accuracy and clinical applicability compared with the Child-Pugh and MELD models.
One concern worth highlighting is that our proposed nomogram could be used for early identi cation of EVB patients who are at a high risk of rebleeding. The rational therapeutic approach should adapt to the different expected risk of rebleeding for each patient; in other words, more aggressive treatments should be administered early for high-risk patients, and unnecessary procedures should be reduced for low-risk patients. As presented in K-M curves, there was a signi cant correlation between the high-risk patients and the increased probability of rebleeding. We recommend a high-risk patient could be a candidate of an early TIPS or liver transplantation and of an active follow-up plan 29 .
The present study's limitations include its retrospective nature and its single center design. Additionally, HVPG greater than 20 mm Hg has been demonstrated previously to be predictive of rebleeding 30 , whereas HVPG was not available in our study since direct measurement of portal hypertension is invasive and inconvenient. Fortunately, Child-Pugh classi cation could be regarded as an alternative factor because an HVPG above 20 mm Hg is equivalent to Child-Pugh grade C in more than 85% of patients 30 . The third limitation was a lack of a large cohort of patients from other institutions to verify our nomogram. Although the model requires continued re nement and improvement, its current form may be useful in assisting clinicians to identify the high-risk patients, select optimal treatment protocols, as well as to make clinical decisions and follow-up strategy.

Conclusion
In conclusion, we developed and validated a creatinine-based model to individually predict the 3-month rebleeding probability in unselected cirrhotic patients with EVB after modi ed PTVE. Compared with the Child-Pugh and MELD, our nomogram model showed preferable prediction ability and clinical applicability that can help clinicians to discern the high-risk patients so as to perform TIPS or liver transplantation earlier and make intensive follow-up plan for them according to expected probability of recurrent bleeding.