Potential conflict of interest: Nothing to report.
Patients with cirrhosis require endoscopic screening for large esophageal varices. The aims of this study were to determine the cost-effectiveness and patient preferences of a strategy employing abdominal computerized tomography (CT) as the initial screening test for identifying large esophageal varices. In a prospective evaluation,102 patients underwent both CT and endoscopic screening for gastroesophageal varices. Two radiologists read each CT independently; standard upper gastrointestinal endoscopy was the reference standard. Agreement between radiologists, and between endoscopists regarding size of varices was determined using kappa statistic. Cost-effectiveness analysis was performed to determine the optimal screening strategy for varices. Patient preference was assessed by questionnaire. CT was found to have approximately 90% sensitivity in the identification of esophageal varices determined to be large on endoscopy, but only about 50% specificity. The sensitivity of CT in detecting gastric varices was 87%. In addition, a significant number of gastric varices, peri-esophageal varices, and extraluminal pathology were identified by CT that were not identified by endoscopy. Patients overwhelmingly preferred CT over endoscopy. Agreement between radiologists was good regarding the size of varices (Kappa = 0.56), and exceeded agreement between endoscopists (Kappa = 0.36). Use of CT as the initial screening modality for the detection of varices was significantly more cost-effective compared to endoscopy irrespective of the prevalence of large varices. Conclusion: Abdominal CT as the initial screening test for varices could be cost-effective. CT also permits evaluation of extra-luminal pathology that impacts management. (HEPATOLOGY 2008.)
Cirrhosis is often complicated by the development of portal hypertension. Depending on the severity of liver disease as determined by the Child-Turcotte-Pugh (CTP) classification, between 50%-80% of patients with cirrhosis will ultimately develop esophageal or gastric varices.1-3 Because of the significant morbidity and mortality associated with bleeding from varices, patients with cirrhosis undergo screening for esophageal varices using upper gastrointestinal endoscopy.4-6 Important predictors of bleeding risk include the presence of large esophageal varices, and CTP Class C.4, 7 Up to 30% of patients screened by endoscopy are found to have moderate-to-large varices (≥5 mm diameter),8 which are at high risk of hemorrhage.7 The presence of large varices is considered an indication for prophylaxis against variceal bleeding with either nonselective beta-blockers or endoscopic variceal ligation.4, 9 Patients without varices or with small varices (<5 mm diameter) are currently not candidates for prophylactic therapy but need to undergo endoscopic surveillance to monitor for the development of large varices.6, 10 Measurement of hepatic venous pressure gradient (HVPG) is not currently recommended to screen for varices even though it accurately measures portal pressure4 and is a predictor of hepatic decompensation.11
Compliance with endoscopic screening recommendations is limited, given its invasive and expensive nature, the need for sedation, and the sometimes poor tolerance of the procedure.12 A less invasive, better tolerated and less expensive test with high sensitivity and specificity for detection of large varices would allow for better selection of patients to undergo endoscopic screening for large esophageal varices. In an effort to limit the number of patients who should undergo endoscopic examinations, parameters such as platelet count and prothrombin time; and radiological criteria such as spleen size, have been studied, but found to be suboptimal predictors of high-risk varices.13-15 Ultrasound imaging also has limited specificity and cannot replace endoscopy as a screening tool for large esophageal varices.16, 17 The lack of benefit of beta-blockers in unselected patients with cirrhosis and portal hypertension, and to patients with cirrhosis but without varices18 lends importance to the identification of patients with large varices.4
Another group of patients with cirrhosis who undergo frequent endoscopies are those undergoing variceal ligation. It is also unclear based only on endoscopic findings which patients are at risk for redevelopment of varices following variceal obliteration. The presence of peri-esophageal varices on endo-sonography is correlated with recurrence of luminal esophageal varices after endoscopic variceal eradication has been performed.19 Peri-esophageal varices which cannot be detected by endoscopy, therefore, may be an important bed of collateral vessels to evaluate in patients with portal hypertension.
With advancement in multi-detector CT imaging, spontaneous portosystemic shunts, esophageal and gastric varices, and peri-luminal varices are increasingly recognized in patients with cirrhosis. Since CT imaging is noninvasive, does not require sedation, and allows review and accurate measurement of variceal size, it is reasonable to believe that CT would be better tolerated than endoscopy by most patients. Furthermore, if the accuracy of CT in detecting esophageal varices is significant, a strategy that employs initial CT for surveillance for large varices could be cost-effective. Two recent studies suggest that multi-detector CT is comparable to upper endoscopy in detecting small and large varices.20, 21 Only one of these studies was carried out prospectively, but no cost analysis was carried out. We, therefore, carried out a prospective comparison of CT imaging against upper gastrointestinal endoscopy for the detection of large esophageal varices. Patient preferences for the procedure and cost-effectiveness were also addressed.
We screened a population of 581 consecutive patients with cirrhosis who were scheduled to undergo upper gastrointestinal endoscopy. Exclusion criteria included inability to provide consent, patients who had previously undergone liver transplantation, previous portosystemic shunt procedure, or had a recent history (<7 days) of upper gastrointestinal bleeding. In addition, patients with renal insufficiency defined as a serum creatinine of >1.7 mg/dL in nondiabetics or >1.5 mg/dL in diabetics were excluded given concerns regarding the requirement of intravenous contrast during CT. Approximately 300 patients met inclusion criteria for screening for esophageal varices. Patients who had endoscopic variceal therapy (n = 19) were also screened and 10 patients were included in the study to determine the role of CT imaging in determining the presence of varices in this group of patients. The last endoscopic therapy session in this group was greater than 4 weeks prior to the CT scan in order to eliminate any potential radiological artifact from the endoscopic treatment. Patients who met inclusion criteria but declined entry into the study cited either scheduling conflicts (n = 142),or fear of radiation (n = 24). The diagnosis of cirrhosis in the 134 patients who consented to the study was based on histology (N = 76); or the presence of ascites, thrombocytopenia or splenomegaly combined with low serum albumin (<3.4 g/dL) and prolongation of the prothrombin time (international normalized ratio ≥ 1.3) with compatible abdominal imaging (N = 58). No endoscopies were performed purely for research purposes. Written informed consent was obtained from all patients who were enrolled into the study. Of the 134 patients who consented, 102 completed both endoscopic and CT examinations.
To reflect practice in the real world, the endoscopic procedures were deliberately chosen to be carried out by regularly scheduled endoscopists rather than a small selected group of very experienced endoscopists. Varices were regarded as present or absent, with size recorded as either large (≥5 mm diameter) or small (<5 mm diameter) based on subjective assessment of diameter. Titrated sedation was undertaken in all patients with midazolam and fentanyl. The size of the esophageal varices was measured in the distal 5 cm of the esophagus during withdrawal of the instrument.
Two gastrointestinal radiologists (R1and R2), each with over 10 years of experience and blinded to each other's evaluation, as well as to the results of the endoscopy, read each CT study. Axial images were evaluated to determine the presence and size of esophageal and gastric varices.
Multidetector CT scans (4 detectors or higher)were performed using 0.5 second rotation time, scanner settings of 250 mAs and 120 kVp, slice thickness 3 mm, and 3 mm reconstruction interval. Intravenous low osmolar iodinated contrast (Omnipaque 300; GE Healthcare) was administrated at a rate of 4mL/second, with late arterial phase scanning initiated 35 seconds after contrast injection, and carried out from the liver dome through the liver and pancreas. Portal phase imaging was initiated 70 seconds after contrast injection, and was carried out from above the diaphragm to the iliac crest. Partial phase images were also reconstructed to a nominal slice thickness of 0.75-1.5 mm and a 20-cm field of view to maximize spatial resolution.. The total effective radiation dose for this protocol was 15 mSv. There was a possible clinical indication to perform CT of the abdomen in 44 of the patients; CT was performed solely for research purposes in 58 patients. Large esophageal varices on CT scan were defined as those that were measured as greater than or equal to 5 mm in diameter, with small varices being those that measured less than 5 mm in diameter.
To determine degree of interobserver variability regarding characterization of variceal size between endoscopists, photographs of esophageal varices were taken during endoscopies and randomly selected images were circulated among 5 endoscopists, 2 with <5 years' experience in practice and 3 with <15 years' experience. The endoscopists were blinded to the results of the other's interpretation and were asked to characterize the endoscopic images as either small, large, or absent varices.
Patient satisfaction with endoscopy and CT was determined by administering a questionnaire for each patient to complete 24 hours after the procedure. The questionnaire determined the patient's opinion regarding elements of comfort and convenience during each test. In addition, patient preference for either study and the reason for preference were elicited.
Sensitivity, specificity, positive and negative predictive values of CT in determining characteristics of varices were determined for both radiologists with endoscopy regarded as the reference standard.
The sensitivity of CT for detecting large varices was determined by identifying which of those patients with large varices at endoscopy were identified by CT as having esophageal varices. The specificity of CT for identifying esophageal varices was defined by correctly identifying the absence of varices in those patients in whom no varices were found endoscopically. Dilated and tortuous veins that protruded into the lumen of the stomach on CT were termed gastric varices. Variceal channels that coursed along the adventitia of the esophagus, but did not protrude into the esophageal lumen were termed peri-esophageal varices. Kappa statistic was used to determine agreement between observers in grading the size of the varices, both on endoscopy and on CT.
We assumed a 40% prevalence rate of esophageal varices, and a 20 % rate of large varices.2, 8 To detect a 15% difference in detection rate of esophageal varices between CT and endoscopy, 95 patients would require to be studied for an alpha of 0.05 (one-tailed test), or alpha of 0.10 (two-tailed test), and beta of 0.20.
We compared the cost-effectiveness of 3 strategies for the detection of large varices in patients with cirrhosis using a decision tree approach. The comparison strategies were (1) Endoscopy, (2) CT and (3) CT + Endoscopy only for patients with small varices on CT. The model was contructed using Treeage Pro Suite 2007 (Treeage Software, Williamstown, MA). Patients considered in this analysis are assumed to have compensated cirrhosis in whom the presence or absence of esophageal varices is not known. The decision tree assumed a 2-year time horizon. The efficacy and costs of the 3 approaches were compared using the incremental cost-effectiveness ratios (ICERs). The decision tree is presented in the Supplementary Fig. 1.
We made a number of assumptions for this analysis (Supplementary Table 1). We assumed the sensitivity of testing strategies and subsequent complication and bleeding rates from various sources. The sensitivity of detection of large varices with CT was derived from the results presented in this paper. The analysis was conducted from the perspective of a third party payer, considering only direct health care costs. Costs were not discounted due to the relatively short time horizon of the analysis (2 years). The main outcome of the analysis was the cost per variceal bleed prevented. The ICERs were calculated compared to the “Do Nothing” strategy.
The study protocol was approved by the Institutional Review Board.
Of 102 consecutively enrolled patients with cirrhosis, 101 had a contrast enhanced CT performed according to the study protocol (one patient who had previously undergone endoscopic variceal ligation was found to have CT performed without optimal intravenous contrast and was, therefore, excluded). The median interval between performance of endoscopy and of the CT on the patients was 2 days (0 to 5 days interquartile range). Patient characteristics are detailed in Table 1. Nine patients had previously received band ligation in an attempt to eradicate esophageal varices
Table 1. Patient Characteristics
Large Varices N = 41
Small Varices N = 38
No Varices N = 22
Abbreviation: NASH, Nonalcoholic steatohepatitis.
Other: Includes autoimmune, cryptogenic, and hemochromatosis.
Performance characteristics for CT in identifying esophageal varices (Figs. 1, 2) as compared with esophageal varices reported by the endoscopist are shown in Table 2. Both R1 and R2 identified 38/41 (93%) of patients with large varices as having esophageal varices. However, when restricting the criteria to identify only those patients with large varices who were correctly identified by CT as having varices >5 mm in size, the sensitivity decreased to 56% (23/41) for R1 and 66% (27/41) for R2.
Table 2. Endoscopic Grading of Varices
Variceal size by Computerized Tomography (CT)
Large Varices N = 41
Small Varices N = 38
No Varices N = 22
R1= Radiology 1
R2= Radiology 2
CT (≥5 mm)
CT (<5 mm)
R1 correctly identified the absence of varices in 12/22 (specificity 55%) patients and R2 correctly identified the absence of varices in 10/22 (specificity 45%) patients. Furthermore, R1 correctly classified 55/60 (92%) of patients who did not have large varices, while R2 correctly classified 52/60 (87%) patients who did not have large varices. R1 did not identify any patients who were negative for the presence of endoscopically identified varices as having large varices, and R2 labeled only one patient as having large varices who was reported not to have varices by endoscopy.
The agreement between radiologists regarding the size of esophageal varices was within 1 mm on 50% of scans (50/101) (K statistic 0.4). The agreement between the radiologists was 75% (76/101) when the tolerance was changed to within 2 mm of each other (K statistic 0.56), indicating “good” agreement.
Because of the poor agreement between all endoscopists regarding grading of varices, that is, because of a sub-optimal endoscopic reference standard, we re-analyzed the data using a “consensus” grading only among the senior endoscopists as the “gold standard”. The sensitivity in detecting large esophageal varices by R1increased to 75%, and by R2to 85%. The specificity in excluding large varices based on the consensus reference endoscopic grade was 62% for R1and 75% for R2.
Radiological Detection of Gastric Varices.
Table 3 shows the performance of CT in detecting gastric varices (Fig. 3) identified on endoscopy. Both radiologists demonstrated identical sensitivity in detecting gastric varices that were seen on EGD (87% - 13/15) (Kappa statistic 0.7) and also diagnosed a significant number of gastric varices that were not seen by endoscopy (22 by R1, and 9 by R2).
Table 3. Performance of Computerized Tomography in Detecting Gastric Varices
EGD (+) Gastric Varices N = 15
EGD (−) Gastric Varices N = 86
CT gastric varices (+)
CT gastric varices (−)
Because these variceal channels are invisible to endoscopy, but detected as an extra-luminal finding by CT (Fig. 4), no comparison could be made with endoscopy.
R1 identified 8 patients with only peri-esophageal varices, in the absence of either esophageal or gastric varices; 4 of these patients had had esophageal varices obliterated by ligation. R2 identified 6 patients with peri-esophageal varices in the absence of luminal varices; 3 of these patients had had esophageal varices obliterated by ligation. Agreement between the radiologists regarding peri-esophageal varices was poor.
Endoscopic Assessment of Variceal Size: Inter-Observer Agreement.
Of the 3 endoscopists with greater than 15 years of experience, there was complete agreement that the varices pictured were small in 23 of the 35 cases. Of the other 12 cases, at least one of the endoscopists disagreed with the others.. Nine of these cases were determined to be large varices by at least one of the experienced endoscopists. The other 3 cases were called “indeterminate” (unsure whether large or small) by at least one endoscopist and small by the others. Kappa statistic among experienced endoscopists was 0.46 suggesting “fair” agreement.
Of the two endoscopists with less than 5 years of endoscopic experience, there was agreement on 17 of the photographs that the varices pictured were small. Of the other 18 cases, the two less-experienced endoscopists agreed that only 8 of the images represented large varices. Of the other 10 cases, 7 were determined to represent large varices by one of the endoscopists and not to be large by the other. Three cases were graded as “indeterminate” by one of the endoscopists and small by the other (Kappa statistic between the less experienced endoscopists was 0.4).
Overall agreement between the various endoscopists regarding the size of the varices fell somewhat short of “fair agreement” with a Kappa statistic of 0.36.
Eighty-four patients (83%) returned the completed questionnaires after both studies were performed. Of the responders, 74 indicated a preference for CT (88%), 5 (6%) for endoscopy, and 5(6%) had no preference. Among the reasons given for preference of CT over endoscopy were the absence of throat discomfort after the procedure, ability to function immediately after the CT and not to have to wait till the next day before driving, as well as a sense that the CT was an inherently safer exam to have performed. Of the 5 patients who stated a preference for esophagogastroduodenoscopy (EGD) over CT, no reason for this preference was stated on the questionnaire.
Additional findings on CT scans that yielded clinically important information not detected on ultrasound scanning are presented in Table 4.
The results of our base case analysis are presented in Table 5. This table presents the total 2-year costs and effectiveness (percentage of patients in whom a variceal hemorrhage is prevented over 2 years). The “do nothing” strategy cost $1,826 and “prevented” a variceal bleed in 82.6 percent of the patients. This strategy was the least costly but least effective strategy. The ICER for the CT screening and endoscopy screening strategies compared to the “do nothing” strategy were $232/bleed prevented and $35,960/bleed prevented, respectively. The combined CT + endoscopy for patients with small varices on CT strategy was dominated by the other strategies (CT and endoscopy) because it was more expensive and less effective.
Table 5. Costs, Effectiveness and ICERs for Alternative Strategies for Detecting Esophageal Varices
We conducted sensitivity analyses to evaluate the robustness of our findings to changes in base case probability estimates. First, we evaluated the impact of changing the sensitivity of detecting varices using endoscopy. If we decreased this sensitivity from 100% (in the base case) to 90% detection of large varices by endoscopy, the CT screening dominates both the endoscopy and combined CT and endoscopy strategies (that is, CT is the most cost-effective). Second, we evaluated the results by varying the prevalence of large varices. When we decreased the prevalence of large varices to 20% and 30% the ICERs for both the CT and endoscopy screening strategies increased; however, the ICER for CT continued to be lower than that for endoscopy. Decreasing the prevalence of large varices to 20% increased the ICER for CT screening to $7,800/bleed prevented and increased the cost for endoscopy screening to $75,400/ bleed prevented. CT was more cost –effective than endoscopy irrespective of the prevalence of large varices. However, if the prevalence of large varices fell to below 5%, then the ICER for CT screening was greater than $ 50,000, the threshold value accepted for cost-effectiveness.
Decisions regarding which strategy to use to screen patients with cirrhosis for large varices depends both on the cost-effectiveness of the strategy as well as patient preferences. Based on the results of this study, not only is the CT based strategy the most cost-effective, it is likely to be the strategy most preferred by patients. Moreover, the size of the varices can be measured accurately by the hepatologist since the CT images are a permanent record unlike endoscopy where the procedure is not usually recorded. Using the CT based strategy, patients identified by CT as having esophageal varices ≥5 mm in diameter can confidently be placed on primary prophylactic therapy for variceal bleeding with beta-blockers without further evaluation by endoscopy. Endoscopy would be required in these patients only if they have contraindications to beta-blockers and require endoscopic variceal ligation. If no varices are identified by CT scan, then endoscopy would not be required in these patients. Screening for varices could then be repeated in 2-3 years according to current recommendations. Another strategy would be to carry out endoscopy only in those patients identified by CT as having varices <5 mm in size. In our study, these patients were variably described as having either small or large varices on CT; the decision regarding whether to administer prophylactic therapy could then be made on endoscopy. However, this strategy would be less cost-effective. We did not compare the costs of CT imaging with capsule endoscopy since there are currently no published data on the efficacy of capsule endoscopy as a screening modality for esophageal varices.
Patient preference for CT over EGD was demonstrated unequivocally in our study as in previous studies.21 In one of the previous studies, virtual esophagography could be carried out using CT scans, but this procedure requires time-consuming and invasive intubation of the esophagus with a catheter for air insufflation.21 Therefore, a scanning procedure that does not require esophageal intubation might be preferred by patients. As one of the major barriers to surveillance for esophageal varices is patient compliance with and tolerance of endoscopy, the use of a fast, non-invasive CT scan may increase compliance with surveillance recommendations. Moreover, CT has always been recognized as a preferred test to evaluate extra-luminal pathology in the abdomen. Patients with cirrhosis are at an increased risk of development of hepatocellular carcinoma (HCC) and recommendations for screening for HCC are centered on ultrasound imaging of the liver. In our study, not only was CT effective for the evaluation of liver masses, it was able to detect five primary liver tumors that were not detected on abdominal ultrasound. The higher percentage of patients with “large esophageal varices” and HCC as compared with other epidemiological studies probably reflects both the tertiary referral practice as well as over-estimation of variceal size on endoscopy.
One clear limitation of our study is the use of endoscopy as the reference standard. Previous studies have demonstrated that there is a lack of good correlation between endoscopists for the determination of variceal size, an observation confirmed by our study.22, 23 In our study, given the open access nature of our endoscopy unit, many different endoscopists performed endoscopy. This may have resulted in an inaccurate grading system of variceal size by endoscopy and therefore, a less than optimal reference standard. The lack of a confirmatory opinion regarding the size of varices reported endoscopically reflects the real-life experience in grading of varices. There was a high degree of agreement regarding only small varices that are not in need of prophylactic therapy. However, there was considerable disagreement among endoscopists in determining whether varices were large of not. Using video clips of the procedure rather than photographs may have increased agreement between observers. Unlike in the two previous studies comparing CT scans with endoscopy for the detection of esophageal varices20, 21 agreement between the radiologists in our study was higher than between endoscopists. More accurate grading of varices on endoscopy using a reference such as open biopsy forceps is probably necessary to accurately grade the size of the varices. Because of the lack of a true gold-standard for grading of varices, the performance of CT imaging in assessing variceal size may actually be higher than in our study.
The role of assessment of size of gastric varices by endoscopy is not well defined in that there are no clear recommendations regarding prophylactic therapy. CT demonstrated high sensitivity for the assessment of gastric varices and, in addition, detected gastric varices in many patients in whom gastric varices were not reported at endoscopy. This suggests that CT may either be more sensitive than endoscopy for the detection of gastric varices, or less specific.
One distinction that is clear in our study is that the distension of the esophagus is different during endoscopy and during CT which may account for some of the discrepancy between the two techniques in assessing variceal size. During CT, as with capsule endoscopy, the esophagus is in its normal, nondistended state which permits variceal channels to be unaffected. During endoscopy, the esophagus is insufflated which alters the trans-luminal pressure and, therefore, the hemodynamics of the varix. Because bleeding from varices occurs while the esophagus is in its normal nondistended state, the measurement of varices in a nondistended state may provide a more accurate measure of which varices are at risk of bleeding.
Whereas the acquisition parameters for the performance of CT scanning used in our study were routine, two sub-specialized radiologists interpreted all examinations. Further, the generation of portal phase, small field of view images with approximately 1-mm slice thickness is not routine in our clinical practice. Several advances in CT technology may improve the detection and grading of varices by radiologists. Sixty-four slice CT scanning systems routinely generate data sets with isotropic spacial resolution, such that sagittal and coronal reconstructions through the esophagus can be generated, potentially decreasing measurement variability as vessels are displayed within the imaging plane. The radiation dose with our protocol was slightly higher than that for an abdominal and pelvis CT. The use of 80 KV scanning will further improve the visualization of esophageal varices as the iodine signal is 1.5-2.0 times greater when scanning at this low energy, but may require new two-tube CT systems to reduce imaging noise in heavy patients.
In conclusion, our study has demonstrated contrast enhanced abdominal CT to be a promising technique for screening patients with cirrhosis for large esophageal varices. In addition, abdominal CT offers the potential for the detection of hepatocellular carcinoma as well as unsuspected extraluminal pathology that may affect the future care of the patient.