Variceal bleeding is among the most serious consequences of chronic liver disease or portal vein obstruction in children and is associated with a high risk of death (1–8). Management of children at risk for variceal bleeding differs considerably among pediatric hepatologists, largely because of a lack of pediatric research studies that examine approaches to screening for varices and treatment to prevent hemorrhage (9).
To improve the management of children at risk for bleeding from esophageal varices, it is first necessary to identify affected children who can be included in pediatric studies of primary prophylaxis. Esophagogastroduodenoscopy (EGD) is the present reference standard diagnostic test, but it is invasive, time consuming, expensive, and associated with risk. There is therefore a pressing need for a noninvasive test that reliably enables targeting of EGD to those children with the highest risk of varices, and avoidance of EGD in children without varices.
The aim of the present study was to derive a noninvasive clinical prediction rule capable of identifying children with esophageal varices, in line with 1 of the research priorities of the American Association for the Study of Liver Disease and the American College of Gastroenterology (10).
Setting and Eligibility
In the present retrospective study, consecutive children younger than 18 years of age who underwent EGD between 2000 and 2007 at the Hospital for Sick Children, Toronto, were included if they had liver disease or portal vein thrombosis. The hospital specializes in the delivery of secondary and tertiary level care to the children of the greater Toronto area and provides the largest pediatric liver transplant program in Canada. Endoscopies were undertaken to screen for either esophageal varices or gastrointestinal symptoms. We excluded children with history of use of β-blockers, previous portal-systemic shunt surgery or transjugular-intrahepatic-portal-systemic shunt, endoscopic ligation, or sclerotherapy of varices or upper gastrointestinal bleeding before the EGD. Children with malignancy or who had undergone organ transplantation before the EGD were also excluded. The study was approved by the institutional research ethics board.
Data were collected by review of the health record and included demographic details, primary diagnoses, comorbidities, and medications. Clinical, imaging, and laboratory test parameters were recorded, including those that may provide evidence of the severity of portal hypertension. Bloodwork and abdominal ultrasound scan (USS) results were obtained from the test performed closest in time to the EGD and always within a maximum of 6 months. All of the endoscopies were performed by 1 of 3 experienced physicians who used the same Paquet classification system for gradation of esophageal variceal size. In this system, varix size is graded on a 4-point Likert scale: grade 1 varices are small and flattened by insufflation of air, grade 2 are slightly larger and do not flatten, grade 3 are larger but do not touch in the middle of the lumen, and grade 4 varices are large and touch each other in the middle of the lumen (11). Routine practice at our institution is to measure spleen length during abdominal USS. These measurements were expressed as standard deviation scores (z scores) relative to previously established reference standards of spleen length for different ages (12).
First, a descriptive univariate analysis was performed on variables that were determined subjectively by the investigators to be potentially associated with esophageal varices. Categorical variables (eg, existence of collaterals on Doppler ultrasound examination) are presented as proportions and compared using χ2 or Fisher exact tests. Continuous variables (eg, platelet count, albumin concentration, international normalized ratio [INR], spleen length) are presented as mean ± standard deviation (SD) or median interquartile range and compared using unpaired Student t test or Wilcoxon rank-sum test as appropriate for the normal distribution. Univariate logistic regression was used to obtain the corresponding odds ratio (OR) and 95% confidence interval (CI) for each predictive variable. A correction factor of 0.5 was added to cells that contained 0 within the data tables.
Multivariate logistic regression was then modeled to associate predictors with esophageal varices. Screening of variables for the multivariable predictive rules using univariate statistical significance levels involves multiple comparison problems and is known to produce unreliable models (13–17). Therefore, we followed the strong recommendation to set possible predictors a priori based on extensive literature review and expert opinion (16,18). Fine-tuning of the initial limited list of possible predictive variables may be aided by fitting a few models while trying to maximize the c statistic of the binary model. Calibration of the prediction rule was assessed using the Hosmer and Lemeshow goodness-of-fit test. The small sample size in the present study limited the number of independent variables to be included. We selected a priori 2 sets of variables that were consistently reproduced in the adult literature (19–27); albumin or INR to be the first variable (reflecting the severity of liver disease) and platelet count or platelet/spleen z score to be the second variable (reflecting severity of portal hypertension). To avoid negative z scores, a factor of 5 was added to all of the spleen length z scores. The choice of either variable in each set (albumin or INR) depended on optimizing the model by maximizing the c statistic and minimizing −2 log-likelihood after fitting all of the possible models. The β-estimates of the predictors retained in the best model were used to guide the weights of the variables in the clinical prediction rule, after multiplication by 10 and rounding to the nearest 0.5.
The best cutoff point of the resulting clinical prediction rule to differentiate children with and without varices was determined to be the point where the second diagonal crossed the receiver operating characteristic (ROC) curve. However, other cutoffs to maximize sensitivity or specificity were also explored. An area under the ROC curve (±95% CI) of more than 0.7 was considered a fair prediction rule, 0.8 good, and more than 0.9 excellent. Sensitivity, specificity, negative and positive predictive values, and likelihood ratios were calculated for different cutoff points. Ninety-five percent CI were calculated for all of the point estimates.
All of the comparisons were made using 2-sided significance levels of P < 0.05. Statistical analyses were performed using SAS version 9.1 (SAS, Cary, NC) and SPSS version 15.0 (SPSSInc, Chicago, IL).
Of the 87 children with liver disease or portal vein thrombosis who underwent EGD during the study period, 51 met the eligibility criteria and were included in the present study (mean age 11 years, range 2 months–17 years, 25 males) (Table 1). The common reason for exclusion was a previous episode of bleeding (n = 17). Ninety-two percent of all of the data fielded were complete for the 51 eligible children. Forty patients were taking medications, the most frequent was ursodeoxycholic acid (n = 25), but none were receiving β-blockers. Basic characteristics did not differ between the patients with or without esophageal varices, including age, sex, Child-Pugh Score, and Peiatric End-Stage Liver Disease (PELD) or Model for End-Stage Liver Disease (MELD) scores. There were no patients with evidence of encephalopathy.
EGD identified esophageal varices in 17 of the 51 children (33%), 9 of whom had large varices (grade 2 or more than 18% of the total, 53% of the varices group). Eight patients had portal hypertensive gastropathy and 3 patients with large esophageal varices also had gastric varices.
Variables found to differ significantly between children with and without varices included splenomegaly on physical examination, spleen length z score measured by USS, presence of collaterals on USS, platelet/spleen length z score ratio, platelet count, white blood cell count, aspartate aminotransferase/platelet ratio, INR, aspartate aminotransferase/alanine aminotransferase ratio, albumin, and creatinine (Table 2).
Derivation of the Noninvasive Prediction Rule
Multivariate logistic regression was modeled to associate predictors with esophageal varices. A model with platelets, spleen length z score, and albumin as the independent variables had the best fit among all of the models, with the smallest −2 log-likelihood and highest c statistic (Table 3).
After adjusting the β-coefficient as described, the resulting clinical prediction rule was:
where platelet count is measured in units × 109/L, SAZ is the spleen length z score, and albumin is measured in grams per liter. Smaller values are associated with a higher likelihood of varices, with the best cutoff value of 130 chosen to maximize sensitivity and negative predictive value as required for a screening test (Fig. 1, Table 4). The model was well calibrated as evident from the Hosmer and Lemeshow goodness-of-fit test (P = 0.81). The area under the ROC curve of this clinical rule to predict esophageal varices was 0.93 (0.85–0.99), implying excellent discriminant ability. Other cutoff values maximizing specificity are also presented (Table 4).
We report the derivation of a predictive model that has a high sensitivity and negative predictive value for the identification of children with esophageal varices. The rule could be used to identify children most likely to have varices, thereby avoiding EGD in children who do not have varices. We considered only simple, commonly available, reproducible variables because we believe that the other previously reported noninvasive predictors of esophageal varices were less reproducible in clinical practice (28) and were subject to interobserver variability (29). We measured spleen length using ultrasonography which is an easily obtainable, noninvasive, and reproducible test (30,31). The expression of SAZ enables appropriate comparison among children of different ages. The platelet count/spleen diameter ratio has been previously shown to accurately identify varices in adults with cirrhosis and has a logical pathophysiological basis in children with portal hypertension. The increase in spleen length in patients with chronic liver disease almost always reflects the increased portal pressure (32,33), and thrombocytopenia may be the result of splenic pooling of platelets because of portal hypertension, immune-mediated mechanisms, or lower thrombopoietin synthesis (34–36).
Endoscopic screening is recommended for adults with cirrhosis to identify those with varices who may benefit from prophylactic treatment, because there is evidence that prophylactic therapy with β-blockers or endoscopic variceal ligation is effective in reducing the incidence of variceal bleeding (10). However, EGD is invasive and unpleasant and will find no varices in up to 50% of adults with cirrhosis (10,37). There is, therefore, a strong interest in identifying a noninvasive test that can either replace screening EGD or select patients for EGD who have a greater likelihood of having varices.
Data to support the noninvasive identification of varices in children are sparse. In a recent study of children with portal hypertension, children having cirrhosis with splenomegaly were 14.6-fold more likely to have esophageal varices compared with children having cirrhosis without splenomegaly. Hypoalbuminemia increased the likelihood of varices (OR 4.17 [95%] CI 1.43–12.18), whereas the significance of thrombocytopenia in the univariate analysis did not hold in the multivariable modeling (38). Fifteen children with portal vein cavernoma and esophageal varices were evaluated in another pediatric study that analyzed ultrasound markers for esophageal varices. Abdominal USS revealed an increased lesser omentum/aorta diameter ratio in children with portal hypertension, compared with controls (P < 0.001) (39).
Several studies in adults with cirrhosis have demonstrated the potential for noninvasive tests to identify esophageal varices (20–24,26,27,40–50). The ratio between spleen diameter and platelet count has repeatedly shown high diagnostic accuracy in these adult studies, ranging from 0.86 to 1 (24,25,40–42). The optimal cutoff value for this ratio has yet to be determined and depends in part upon the severity of disease in the patient population under consideration. Interestingly, patients identified by the ratio as “false-positives” were more likely to have esophageal varices at follow-up compared with the patients with “true-negative” results (25).
The limitations of the present study include its retrospective design and small sample size. Screening EGD for esophageal varices in children was not routinely performed in our center and the study population is therefore also composed of children undergoing endoscopy for investigations of gastrointestinal symptoms. For this reason, younger children with biliary atresia are poorly represented, and there is an excess of older children with primary sclerosing cholangitis who underwent endoscopy as a part of their assessment for inflammatory bowel disease. Blinding of endoscopists to the presence and degree of splenomegaly and thrombocytopenia cannot be guaranteed in retrospect. The inclusion of children with portal vein thrombosis as well as intrahepatic disease may make the clinical prediction rule more widely applicable to different patient groups, although validation studies will be required in each group. The number of patients in the present study did not allow for reliable subgroup analysis to compare children with prehepatic and hepatic portal hypertension. Nonetheless, EGD was performed in children with various severities of liver disease in the same center and using a single classification system. We focused on the presence of esophageal varices of any size, rather than on the presence of large esophageal varices, because this is the first step in the diagnostic workup at which point decisions regarding follow-up and treatment are made.
Although our clinical prediction rule was the most accurate test of those we studied, other simple tests such as splenomegaly and platelet count also showed a high degree of diagnostic accuracy. A previous study suggested that splenomegaly detected on physical examination has a high sensitivity and specificity for the diagnosis of esophageal varices (38). These simple tests would be more practical to use on a daily basis. Nevertheless, the clinical prediction rule standardizes subjective assessments of physicians and we therefore speculate that the reliability of our clinical prediction rule will prove to be higher when measured in validation studies. The clinical prediction rule uses variables that are commonly measured in routine clinical care of children with liver disease or portal vein thrombosis, thus reducing the potential cost of the present approach to the identification of varices.
In summary, we found that variables linked to portal hypertension (platelet count, spleen length) as well as those linked to the severity of liver dysfunction (albumin, INR) are associated with the presence of esophageal varices. We have derived a clinical prediction rule that uses simple noninvasive tests and shows accuracy for the identification of children who require EGD to characterize their esophageal varices. We are now undertaking a prospective study to validate these encouraging results in an additional cohort of children.
1. Lykavieris P, Gauthier F, Hadchouel P, et al
. Risk of gastrointestinal bleeding during adolescence and early adulthood in children with portal vein obstruction. J Pediatr 2000; 136:805–808.
2. Goncalves ME, Cardoso SR, Maksoud JG. Prophylactic sclerotherapy in children with esophageal varices
: long-term results of a controlled prospective randomized trial. J Pediatr Surg 2000; 35:401–405.
3. Miga D, Sokol RJ, Mackenzie T, et al
. Survival after first esophageal variceal hemorrhage in patients with biliary atresia. J Pediatr 2001; 139:291–296.
4. van Heurn LW, Saing H, Tam PK. Portoenterostomy for biliary atresia: Long-term survival and prognosis after esophageal variceal bleeding. J Pediatr Surg 2004; 39:6–9.
5. Kobayashi A, Itabashi F, Ohbe Y. Long-term prognosis in biliary atresia after hepatic portoenterostomy: analysis of 35 patients who survived beyond 5 years of age. J Pediatr 1984; 105:243–246.
6. Toyosaka A, Okamoto E, Okasora T, et al
. Outcome of 21 patients with biliary atresia living more than 10 years. J Pediatr Surg 1993; 28:1498–1501.
7. Mitra SK, Kumar V, Datta DV, et al
. Extrahepatic portal hypertension
: A review of 70 cases. J Pediatr Surg 1978; 13:51–57.
8. Webb LJ, Sherlock S. The aetiology, presentation and natural history of extra-hepatic portal venous obstruction. Q J Med 1979; 48:627–639.
9. Shneider BL. Approaches to the management of pediatric portal hypertension
: results of an informal survey. In: Groszmann RJ, Bosch J, eds. Portal Hypertension in the 21st Century
. Dordrecht. the Netherlands: Kluwer Academic Publishers; 2004:167–172.
10. Garcia-Tsao G, Sanyal AJ, Grace ND, et al
. Prevention and management of gastroesophageal varices and variceal hemorrhage in cirrhosis. Hepatology 2007; 46:922–938.
11. Paquet KJ. Prophylactic endoscopic sclerosing treatment of the esophageal wall in varices – a prospective controlled randomized trial. Endoscopy 1982; 14:4–5.
12. Megremis SD, Vlachonikolis IG, Tsilimigaki AM. Spleen length in childhood with US: normal values based on age, sex, and somatometric parameters. Radiology 2004; 231:129–134.
13. Harrell FE Jr, Lee KL, Califf RM, et al
. Regression modelling strategies for improved prognostic prediction. Stat Med 1984; 3:143–152.
14. Harrell FE Jr, Lee KL, Matchar DB, et al
. Regression models for prognostic prediction: advantages, problems, and suggested solutions. Cancer Treat Rep 1985; 69:1071–1077.
15. Harrell F. Regression coefficients and scoring rules. J Clin Epidemiol 1996; 49:819.
16. Harrell FE Jr, Lee KL, Mark DB. Multivariable prognostic models: issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Stat Med 1996; 15:361–387.
17. Austin PC, Tu JV. Automated variable selection methods for logistic regression produced unstable models for predicting acute myocardial infarction mortality. J Clin Epidemiol 2004; 57:1138–1146.
18. Laupacis A, Sekar N, Stiell IG. Clinical prediction rules. A review and suggested modifications of methodological standards. JAMA 1997; 277:488–494.
19. Sanyal AJ, Fontana RJ, Di Bisceglie AM, et al
. The prevalence and risk factors associated with esophageal varices
in subjects with hepatitis C and advanced fibrosis. Gastrointest Endosc 2006; 64:855–864.
20. Zaman A, Hapke R, Flora K, et al
. Factors predicting the presence of esophageal or gastric varices in patients with advanced liver disease. Am J Gastroenterol 1999; 94:3292–3296.
21. Ng FH, Wong SY, Loo CK, et al
. Prediction of oesophagogastric varices in patients with liver cirrhosis. J Gastroenterol Hepatol 1999; 14:785–790.
22. Pilette C, Oberti F, Aube C, et al
. Non-invasive diagnosis of esophageal varices
in chronic liver diseases. J Hepatol 1999; 31:867–873.
23. Madhotra R, Mulcahy HE, Willner I, et al
. Prediction of esophageal varices
in patients with cirrhosis. J Clin Gastroenterol 2002; 34:81–85.
24. Giannini E, Botta F, Borro P, et al
. Platelet count/spleen diameter ratio: proposal and validation of a non-invasive parameter to predict the presence of oesophageal varices in patients with liver cirrhosis. Gut 2003; 52:1200–1205.
25. Giannini EG, Botta F, Borro P, et al
. Application of the platelet count/spleen diameter ratio to rule out the presence of oesophageal varices in patients with cirrhosis: a validation study based on follow-up. Dig Liver Dis 2005; 37:779–785.
26. Chalasani N, Imperiale TF, Ismail A, et al
. Predictors of large esophageal varices
in patients with cirrhosis. Am J Gastroenterol 1999; 94:3285–3291.
27. Zein CO, Lindor KD, Angulo P. Prevalence and predictors of esophageal varices
in patients with primary sclerosing cholangitis. Hepatology 2004; 39:204–210.
28. Valletta EA, Loreti S, Cipolli M, et al
. Portal hypertension
and esophageal varices
in cystic fibrosis. Unreliability of echo-Doppler flowmetry. Scand J Gastroenterol 1993; 28:1042–1046.
29. Tamayo SG, Rickman LS, Mathews WC, et al
. Examiner dependence on physical diagnostic tests for the detection of splenomegaly: a prospective study with multiple observers. J Gen Intern Med 1993; 8:69–75.
30. Winkfield B, Aube C, Burtin P, et al
. Inter-observer and intra-observer variability in hepatology. Eur J Gastroenterol Hepatol 2003; 15:959–966.
31. O'Donohue J, Ng C, Catnach S, et al
. Diagnostic value of Doppler assessment of the hepatic and portal vessels and ultrasound of the spleen in liver disease. Eur J Gastroenterol Hepatol 2004; 16:147–155.
32. Liangpunsakul S, Ulmer BJ, Chalasani N. Predictors and implications of severe hypersplenism in patients with cirrhosis. Am J Med Sci 2003; 326:111–116.
33. Pockros PJ, Duchini A, McMillan R, et al
. Immune thrombocytopenic purpura in patients with chronic hepatitis C virus infection. Am J Gastroenterol 2002; 97:2040–2045.
34. Peck-Radosavljevic M. Thrombocytopenia in liver disease. Can J Gastroenterol 2000; 14(Suppl D):60D–66D.
35. Peck-Radosavljevic M. Review article: coagulation disorders in chronic liver disease. Aliment Pharmacol Ther 2007; 26(Suppl 1):21–28.
36. Giannini E, Botta F, Borro P, et al
. Relationship between thrombopoietin serum levels and liver function in patients with chronic liver disease related to hepatitis C virus infection. Am J Gastroenterol 2003; 98:2516–2520.
37. de Franchis R, Primignani M. Natural history of portal hypertension
in patients with cirrhosis. Clin Liver Dis 2001; 5:645–663.
38. Fagundes ED, Ferreira AR, Roquete ML, et al
. Clinical and laboratory predictors of esophageal varices
in children and adolescents with portal hypertension
syndrome. J Pediatr Gastroenterol Nutr 2008; 46:178–183.
39. Moreno E, Torres P, Trejo C, et al
. Predictive value of ultrasonography in portal hypertension
. G E N 1991; 45:101–104.
40. Giannini EG, Zaman A, Kreil A, et al
. Platelet count/spleen diameter ratio for the noninvasive diagnosis of esophageal varices
: results of a multicenter, prospective, validation study. Am J Gastroenterol 2006; 101:2511–2519.
41. Sethar GH, Ahmed R, Rathi SK, et al
. Platelet count/splenic size ratio: a parameter to predict the presence of esophageal varices
in cirrhotics. J Coll Physicians Surg Pak 2006; 16:183–186.
42. Zimbwa TA, Blanshard C, Subramaniam A. Platelet count/spleen diameter ratio as a predictor of oesophageal varices in alcoholic cirrhosis. Gut 2004; 53:1055.
43. Schepis F, Camma C, Niceforo D, et al
. Which patients with cirrhosis should undergo endoscopic screening for esophageal varices
detection? Hepatology 2001; 33:333–338.
44. Thomopoulos KC, Labropoulou-Karatza C, Mimidis KP, et al
. Non-invasive predictors of the presence of large oesophageal varices in patients with cirrhosis. Dig Liver Dis 2003; 35:473–478.
45. Thabut D, Trabut JB, Massard J, et al
. Non-invasive diagnosis of large oesophageal varices with FibroTest in patients with cirrhosis: a preliminary retrospective study. Liver Int 2006; 26:271–278.
46. Kazemi F, Kettaneh A, N'kontchou G, et al
. Liver stiffness measurement selects patients with cirrhosis at risk of bearing large oesophageal varices. J Hepatol 2006; 45:230–235.
47. Cottone M, D'Amico G, Maringhini A, et al
. Predictive value of ultrasonography in the screening of non-ascitic cirrhotic patients with large varices. J Ultrasound Med 1986; 5:189–192.
48. Zaman A, Becker T, Lapidus J, et al
. Risk factors for the presence of varices in cirrhotic patients without a history of variceal hemorrhage. Arch Intern Med 2001; 161:2564–2570.
49. Vanbiervliet G, Barjoan-Marine E, Anty R, et al
. Serum fibrosis markers can detect large oesophageal varices with a high accuracy. Eur J Gastroenterol Hepatol 2005; 17:333–338.
50. Sethar GH, Ahmed R, Rathi SK, et al
. Platelet count/splenic size ratio: a parameter to predict the presence of esophageal varices
in cirrhotics. J Coll Physicians Surg Pak 2006; 16:183–186.