Biliary atresia (BA) is the leading cause of neonatal cholestasis. Because the disease progresses very rapidly to cirrhosis and portal hypertension, prompt diagnosis is essential, as the chances of a successful Kasai procedure are highest when the surgery is performed early in life. That said, the native liver survival rate following a Kasai procedure is only 20% at the age of 20 years (1). Consequently, patients progress to biliary cirrhosis and those developing liver failure must undergo liver transplantation (often very early in life). Although awaiting a transplant, patients run the risk of gastrointestinal (GI) bleeding because of esophageal varices (EV). In cases of portal hypertension, it is established that bleeding risk is correlated with EV characteristics (grade III varices; red color signs) in adults and children alike (2,3). Although recommended in adults (3–5), prophylactic treatment remains controversial in children, despite being suggested by several authors (3,6,7). The procedure is indeed a delicate one and not devoid of complications. Moreover, living donor–related liver transplantation, an increasingly common solution in recent years, has significantly reduced waiting times, leading to doubts about interest of prophylactic treatment. Our study was designed to estimate bleeding risk and risk factors in a population of patients followed up for BA in our hospital by analyzing their clinical, radiological, endoscopic, and laboratory data.
This was a retrospective analysis of patients born between January 1, 2003 and June 1, 2010 and treated for BA in our department. Diagnoses were confirmed by cholangiography during Kasai portoenterostomy or, if such a procedure was not possible because of late diagnosis and disease progression, by cholangiography of the diseased liver during transplantation.
Kasai procedures were rated as completely successful, partially successful, or failed. Interventions were considered successful if bilirubin levels normalized with no subsequent portal hypertension or cholangitis. If bilirubin levels normalized but portal hypertension or repeated cholangitis occurred, the intervention was considered partially successful. If bilirubin levels did not normalize or increased again during the 12 months following surgery, the intervention was considered a failure.
In cases of Kasai surgery failure, patients were rapidly placed on the waiting list for liver transplantation from a deceased donor or referred for living related donor transplantation. Patient data were assessed at each endoscopic examination from the start of medical management up to the occurrence of upper GI bleeding. If no bleeding occurred, data were collected until final follow-up, including death, or until liver transplantation or endoscopic treatment of EV was performed.
The following parameters were analyzed: clinical (ascites or splenomegaly), laboratory (plasma levels of total bilirubin, albumin, platelets, and coagulation factors), ultrasonographic (ascites, direction of portal flow, hepatic artery resistance index, and portal vein dimensions), and endoscopic (8,9). Data from patients with and without GI bleeding were then compared.
Gastroscopy was performed under general anesthesia with intubation. Because most of the patients were referred, the first elective gastroscopy was performed during the pretransplant checkup of the patients or in the event of clinical or radiological signs of portal hypertension or GI bleeding. The severity and localization of esophageal and gastric varices were described using standard classifications (8,9). Any red color signs were also reported.
In the event of bleeding, EV were treated either by ligation or sclerotherapy. Variceal ligation was performed using a multiband ligator (Saeed Six Shooter Multi-Band Ligator, Cook Medical, Bloomington, IN) depending on patient size. In younger patients, 2% ethoxysclerol was injected (0.5–2 mL per injection site). Concurrent intravenous somatostatin treatment was initiated rapidly and continued until 5 days after bleeding cessation. A β-blocker was then administered variably. Endoscopy was repeated weekly until complete eradication of varices or sooner if bleeding recurred. Follow-up took place after 6 months and then annually. Prophylactic endoscopic variceal ligation was only considered in patients for whom the Kasai procedure proved partially successful and long-term native liver survival was expected.
Bleeding was a binary target variable analyzed on the available patients. Two distinct approaches were used to identify risk factors for bleeding. The first was univariate analysis. In univariate analysis, each variable was tested to determine whether its value was significantly different in the 2 groups (GI bleeding, no GI bleeding). The clinically relevant variables are summarized in Table 1. Three extra features were added to basic features: WaitingTimeBirthEndo (time [days] between patient's birth and the first endoscopy, WaitingTimeBirthOLT (time [days] between patient's birth and the orthotopic liver transplant), WaitingTimeEndoOLT (time [days] between the patient's first endoscopy and the orthotopic liver transplant [approximately the waiting time before the graft in months]). Statistical testing depended on the nature of each variable. An unpaired t test with no assumption of variance equality (Welch test) was applied to the continuous variables. The χ2 test was used for categorical variables (Pearson test). The P values obtained for all variables were then adjusted using the Benjamini-Hochberg procedure for multiple testing corrections (10). Variables with a corrected P value of <0.05 were considered significantly different in the 2 groups and thus regarded as risk factors for bleeding. Multivariate analysis took into account possible interactions between variables. The multivariate approach used in the present study served 2 objectives simultaneously: first, to further develop a predictive model of bleeding and second, to thereby automatically identify the variables relevant to making such predictions. Methodology was based on machine learning and “ensemble” feature selection methods (11). Our aim was to build a large number of models (10,000) using randomly constituted subcohorts of patients. This ensemble approach is highly robust and generally offers good predictive performance. It is, however, difficult to apply to variables with missing values. The dataset was therefore reduced to 23 variables (Table 1, in italics) and 75 patients, with no missing values. Typically, 80% of the 75 patients were randomly selected for each of the 10,000 models. Each model was tested on patients who were not used in its development (ie, the remaining 20%) and was of the generalized least absolute shrinkage and selection operator type (ie, logistic regression with an L1 penalty on regression parameters). The penalty constant was moderate (cost parameter = 1 in LiblineaR). The generalized LASSO model gives direct information on the relative importance of each variable in the decision-making process leading up to prediction (12). For each of the 10,000 models, variables were classified by order of importance, average was calculated, and predominant variables for prediction were identified. All analyses described in this section were performed using R 2.12.1 (base, stats and LiblineaR 1.80–4 packages).
Patient Demographics and Clinical History
In total, 87 patients met our inclusion criteria. Of these, 4 were excluded from analysis (3 lost to follow-up and 1 prematurely deceased from polymalformative syndrome). Sixty-five patients (78%) underwent Kasai portoenterostomy at a median age of 8 weeks (1–17 weeks) (Fig. 1), with 48 of these interventions performed at the referral center.
Surgery was considered partially successful in 12 patients (18%), 4 of whom nonetheless required subsequent liver transplantation. Only 1 intervention appeared completely successful after a 16-month follow-up. The Kasai procedure failed in 52 patients (80%), who were then placed on a waiting list for liver transplantation. One patient died from liver failure while awaiting a deceased-donor transplant. Of the rest, 35 received liver transplants from live donors, and 16 from deceased donors.
The 18 patients who underwent no Kasai procedure because of advanced disease progression were also placed on the waiting list for liver transplantation. Two patients died while awaiting deceased-donor transplantations, one from GI bleeding and the other from liver failure. Eleven received live-donor liver transplants. Five received deceased-donor livers.
A total of 71 of 83 patients (85%) thus underwent liver transplants, 65% from live donors (46/71), at a median age of 11 months (6–58 months). Only 9 patients (11%) survived with their native liver. Time between birth and liver transplant was 572.0 ± 419.1 days for deceased-donor transplantation patients versus 433.1 ± 310.5 days for live-donor transplantation patients (not statistically significant); however, we observed a statistically significant difference between delay from first endoscopy and liver transplantation, approximately the waiting for liver transplant, 272.3 ± 260.0 days for deceased-donor transplantation patients and 129.4 ± 285.4 days for live-donor transplantation patients (P < 0.05).
At least 1 endoscopic examination was carried out in 80 patients at a median age of 8 months (2–62 months). In 70 patients, endoscopy was performed as part of portal hypertension assessment, and in the remaining 10, in response to GI bleeding (hematemesis or melena). In these 10 patients, 4 went to live-donor transplantation, 4 to deceased-donor transplantation, and 2 were not transplanted (1 died and 1 remained under follow-up).
All data obtained on first endoscopic examination are presented in Table 2. Out of 80 patients, 73 (91%) developed EV, of which 31% were grade I, 37% were grade II, and 23% were grade III. Red color signs were observed in 27% of patients and hypertensive gastritis in 62%. There were a total of 7% gastroesophageal varices and 2% isolated gastric varices.
Seven patients developed GI bleeding from EV during the months following their first endoscopy. Most of them were later transplanted, 3 with deceased-donor liver (median waiting time 9 months, range 5–15 months), 2 with live-donor liver (waiting time 1 and 4 months).
A total of 17 patients developed bleeding from EV, in other words 20% (17/83), at a median age of 9.5 months (6–50 months). All bleeding was treated endoscopically using variceal ligation in 5 patients and sclerotherapy in 12 patients younger than 7 months. In total, 50 procedures were carried out (median number of procedures per patient: 3; range 1–6). Bleeding recurrence was observed in 6 patients. One case of uncontrollable bleeding resulted in death. During endoscopic treatment, 1 patient developed an esophagopleural fistula, 1 an esophageal perforation, and 1 heart rhythm disorders that were probably caused by the systemic distribution of ethoxysclerol.
Three patients from this series underwent preventive ligations. After a mean follow-up of 40 months, they exhibited no bleeding and still had their native liver.
Risk Factors For GI Bleeding
Table 3 presents a comparison of the clinical, laboratory, radiological, and endoscopic data from 17 patients with GI bleeding and 66 without GI bleeding, 83 patients in total. Significant differences were obtained for the following variables, even after multiple testing corrections: presence of EV (P < 0.001) and red color signs on first endoscopic examination (P < 0.001); low fibrinogen levels (P < 0.05) at the timing of first endoscopy. A comparison of the 3 EV severity grades revealed significant differences, with grade III patients with EV displaying a higher rate of bleeding as compared with the 2 other grades, both on first endoscopic examination and at the end of follow-up (P value after Pearson χ2 test <0.001 for both time points assessed). Waiting time for liver transplant and type of donor were not statistically significant features to predict the bleeding. The 3 additional calculated features, time between birth and first endoscopy, time between birth and orthotopic liver transplantation, and time between first endoscopy and orthotopic liver transplantation (approximately the waiting time before the graft), were not statistically different in the 2 groups (with GI bleeding vs without GI bleeding): 356.5 ± 284.6 vs 335.6 ± 281.2 days, 476.6 ± 284.6 vs 483.2 ± 281.2 days, and 168.1 ± 275.0 vs 181.7 ± 175.2, respectively.
Multivariate Analysis and Predictive Model for Risk of GI Bleeding
The multivariate analysis approach described above was applied to the remaining 75 patients after data cleansing. Each of these patients was described using 23 variables (Table 1). Among those 75 patients, 16 had GI bleeding as target value, and 59 did not. While building the 10,000 models on the randomly divided patient population (80% for model construction and the remaining 20% for model assessment), we obtained the average performances: mean odds ratio of the prognosis models 21.84, log odds ratio 1.34, accuracy 83% ± 8.3, balanced classification rate 82.4% ± 10.7, correct predicted rate of GI bleeding 81.3% ± 21, and correct predicted rate of non-GI bleeding 83.4% ± 10.1. Table 4 shows the top 10 variables in the average variable ranking by order of importance for all 10,000 predictive models. The signs “+” and “−” indicate whether the factor plays a positive or negative role in the decision-making process. Interestingly, the top 3 variables (blood fibrinogen, EV, and red color signs, all of them at first endoscopy) are also those that appear significantly different in the univariate analysis. These 3 variables are presented as survival curves in Figure 2. Given that fibrinogen is a continuous variable, an arbitrary limit was set at 150 mg/dL so as to yield 2 curves. Figure 3 illustrates the cumulative contribution of the 10 top-ranked features in the decision/prediction process resulting from the multivariate analysis.
The present study followed the evolution of EV in extremely young patients experiencing BA. After Kasai surgery (for most patients) and before liver transplantation (where applicable), EV were found on endoscopic examination in 80% of patients. Of all patients, 20% developed GI bleeding. In >12% of patients, initial endoscopy was performed in response to GI bleeding. Endoscopic treatment of varices was not carried out for prophylactic purposes except for partially successful Kasai in 3 patients. In both univariate and multivariate analyses, we identified the presence of red color signs, higher grade EV (on endoscopy), and low levels of fibrinogen, at first endoscopy, as risk factors for bleeding. Our findings as regards red color signs and grade III EV on endoscopy confirm the result published by Duché et al (3). In addition, however, we have revealed the role of low fibrinogen as a risk factor for bleeding, and this when rates are <150 mg/dL, a value still relatively high in cirrhotic patients. In our univariate analysis, only this coagulation factor was disrupted, thus precluding major defects in hepatic synthesis of coagulation factors. In our multivariate analyses, some other coagulation parameters were not available. Future studies should thus include a more detailed analysis of coagulation, notably with measurements of the levels of individual coagulation factors produced by the liver and whole blood coagulation analyses (eg, via thromboelastography). Indeed, new significant data on assessment of hemostasis in patients with chronic liver disease challenge the dogma that the major coagulopathy in these patients leads consistently to bleeding. Other changes that accompany chronic liver disease may restore the balance of anticoagulant and procoagulant effects (13).
Few patients exhibited gastroesophageal varices or isolated gastric varices and, unlike other authors, we found no relation between bleeding risk and the presence of gastric varices in the cardia (3).
We are nonetheless inclined to recall that ours was a single-center, retrospective study with a limited number of patients. As such, the possibility of optimistic bias cannot be excluded. We have therefore taken various precautions during the development of statistical analysis protocol to protect it as possible to such optimism. Regarding the univariate component, a correction for multiple testing was performed. Regarding the multivariate aspect, the identification of important variables was performed on many different models. Each of these models is also regularized, to overcome the small number of patients available. Finally, the advanced performances are systematically calculated on patients who have served neither to the identification of important variables, nor to estimate the predictive model, by using a patients’ resampling mechanism. It should be noted that in our multivariate analysis, the top 10 variables (of 23 in total) are of greatest interest, as ranked according to weight. We consider the first 3 variables in these analyses to be particularly important, not only because of their multivariate ranking but also because they were already present in the univariate analysis. This was not the case for the remaining 7 variables found by the multivariate analysis. Furthermore, these variables are also of obvious clinical interest. This could provide more accurate assistance in the process of deciding whether to administer prophylactic treatment of EV in individual pediatric patients (14).
Since its initiation at our center in 1993, the above-mentioned related live-donor liver transplant program has grown considerably, accounting in recent years for up to 84% of total annual pediatric liver transplants (unpublished data from our center, (15–17)). This has resulted in significantly shorter transplant waiting times. This leads to the question of whether or not prophylactic treatment of upper GI varices confers any medical benefit. It should also be noted that most patients in our cohort were eventually placed on the waiting list for liver transplantation (80%). As compared to a relatively similar series of 139 patients with BA published by the Bicêtre hospital team (Paris, France) (3), liver transplants were actually performed early at our center (mean age 14 vs 28 months at the Bicêtre hospital). GI bleeding rates were nonetheless found to be relatively similar (20% vs 15% at the Bicêtre hospital), although such events occurred at a younger median age in our center (9.5 vs 17 months). This difference can probably be explained by our center's greater number of patients coming from foreign countries with more advanced disease and no Kasai procedure (21% [18/83] vs only 6% (9/139) of patients treated in France).
Few studies on prophylactic treatment of EV have been conducted in children. Two small prospective pediatric investigations have been undertaken in patients older than 4 years and experiencing cirrhosis of various etiologies, one using ligation and the other sclerotherapy. These trials suggest that both procedures are relatively safe and reduce the risk of subsequent bleeding as compared with control groups not given preventive treatment (7,18). A more recent study by Duché et al reported on the efficacy of preventive sclerotherapy in 13 younger (mean age 13 months) patients with BA (6). One patient from this series died nonetheless from uncontrollable bleeding after 2 sclerotherapy sessions. Nor is the procedure free of complications, as the risk of esophageal stenosis, fistula, dysmotility, or gastroesophageal reflux has been well described (19–22). Compared to sclerotherapy, ligation appears to require a lesser number of procedures for the complete eradication of varices and is associated with a lower rate of bleeding recurrence and complications (23). The larger size of ligation material is nonetheless a limiting factor in young children, whence the use of sclerotherapy in cases where small-caliber endoscopy is required.
In conclusion, we have identified 3 clinically relevant predictive factors for the risk of bleeding from EV in 83 patients experiencing BA and treated at our center. These factors included red color signs and high-grade EV on first endoscopic examination, as well as low fibrinogen levels. These factors proved significant in both multivariate and univariate analyses, even after multiple corrections. Based on them, it is possible to build a predictive model of EV bleeding risk. Validation of this model in subsequent multicenter prospective studies remains mandatory. These prospective studies must also include analysis of fibrinogen supplementation effect on variceal bleeding risk.
1. Lykavieris P, Chardot C, Sokhn M, et al. Outcome in adulthood of biliary atresia: a study of 63 patients who survived for over 20 years with their native liver. Hepatology
2. de Francis R. Revising consensus in portal hypertension: report of the Baveno V consensus workshop on methodology of diagnosis and therapy in portal hypertension. J Hepatol
3. Duché M, Ducot B, Tournay E, et al. Prognostic value of endoscopy in children with biliary atresia at risk for early development of varices and bleeding. Gastroenterology
4. Garcia-Pagan JC, De Gottardi A, Bosch J. The modern management of portal hypertension—primary and secondary prophylaxis of variceal bleeding in cirrhotic patients. Aliment Pharmacol Ther
5. Tiani C, Abraldes JG, Bosch J. Portal hypertension: pre-primary and primary prophylaxis of variceal bleeding. Dig Liver Dis
6. Duché M, Habès D, Roulleau P, et al. Prophylactic endoscopic sclerotherapy of large esophagogastric varices in infants with biliary atresia. Gastrointest Endos
7. Celińska-Cedro D, Teisseyre M, Woynarowski M, et al. Endoscopic ligation of esophageal varices for prophylaxis of first bleeding in children and adolescents with portal hypertension: preliminary results of a prospective study. J Pediatr Surg
8. Beppu K, Inokuchi K, Koyanagi N, et al. Prediction of variceal hemorrhage by esophageal endoscopy. Gastrointest Endosc
9. Sarin SK, Lahoti D, Saxena SP, et al. Prevalence, classification and natural history of gastric varices: a long-term follow-up study in 568 portal hypertension patients. Hepatology
10. Benjamini Y, Hochberg Y. Controlling the false discovery rate: A practical and powerful approach to multiple testing. J R Stat Soc B
11. Abeel T, Helleputte T, Van de Peer Y, et al. Robust biomarker identification for cancer diagnosis with ensemble feature selection methods. Bioinformatics
12. Roth V. The generalized LASSO. IEEE Trans Neural Netw
13. Tripodi A, Mannucci PM. The coagulopathy of chronic liver disease. N Engl J Med
14. Ling SC, Walters T, McKiernan PJ, et al. Primary prophylaxis of variceal hemorrhage in children with portal hypertension: a framework for future research. J Pediatr Gastroenterol Nutr
15. Sokal EM, Cleghorn G, Goulet O, et al. Liver and intestinal transplantation in children: Working Group Report of the First World Congress of Pediatric Gastroenterology, Hepatology, and Nutrition. J Pediatr Gastroenterol Nutr
16. Bourdeaux C, Darwish A, Jamart J, et al. Living-related versus deceased donor pediatric liver transplantation: a multivariate analysis of technical and immunological complications in 235 recipients. Am J Transplant
17. Darwish AA, Bourdeaux C, Kader HA, et al. Pediatric liver transplantation using left hepatic segments from living related donors: surgical experience in 100 recipients at Saint-Luc University Clinics. Pediatr Transplant
18. Gonçalves ME, Cardoso SR, Maksoud JG. Prophylactic sclerotherapy in children with esophageal varices: long-term results of a controlled prospective randomized trial. J Pediatr Surg
19. Poddar U, Thapa BR, Singh K. Frequency of gastropathy and gastric varices in children with extrahepatic portal venous obstruction treated with sclerotherapy. J Gastroenterol Hepatol
20. Reinoso MA, Sharp HL, Rank J. Endoscopic variceal ligation in pediatric patients with portal hypertension secondary to liver cirrhosis. Gastrointest Endosc
21. Price MR, Sartorelli KH, Karrer FM, et al. Management of esophageal varices in children by endoscopic variceal ligation. J Pediatr Surg
22. Nijhawan S, Patni T, Sharma U, et al. Endoscopic variceal ligation in children. J Pediatr Surg
23. Zargar SA, Javid G, Khan BA, et al. Yattoo Endoscopic ligation compared with sclerotherapy for bleeding esophageal varices in children with extrahepatic portal venous obstruction. Hepatology