Contrast-Enhanced CT May Be a Diagnostic Alternative for Gastroesophageal Varices in Cirrhosis with and without Previous Endoscopic Variceal Therapy

Background and Aims Liver fibrosis blood tests, platelet count/spleen diameter ratio (PSR), and contrast-enhanced CT are diagnostic alternatives for gastroesophageal varices, but they have heterogeneous diagnostic performance among different study populations. Our study is aimed at evaluating their diagnostic accuracy for esophageal varices (EVs) and gastric varices (GVs) in cirrhotic patients with and without previous endoscopic variceal therapy. Methods Patients with liver cirrhosis who underwent blood tests and contrast-enhanced CT scans as well as endoscopic surveillance should be potentially eligible. EVs needing treatment (EVNTs) and GVs needing treatment (GVNTs) were recorded according to the endoscopic results. Area under the curves (AUCs) were calculated. Results Overall, 279 patients were included. In 175 patients without previous endoscopic variceal therapy, including primary prophylaxis population (n = 70), acute bleeding population (n = 38), and previous bleeding population (n = 67), the diagnostic accuracy of contrast-enhanced CT for EVNTs was higher (AUCs = 0.816‐0.876) as compared to blood tests and PSR; by comparison, the diagnostic accuracy of contrast-enhanced CT for GVNTs was statistically significant among primary prophylaxis population (AUC = 0.731, P = 0.0316), but not acute or previous bleeding population. In 104 patients with previous endoscopic variceal therapy (i.e., secondary prophylaxis population), contrast-enhanced CT was the only statistically significant alternative for diagnosing EVNTs and GVNTs but with modest accuracy (AUCs = 0.673 and 0.661, respectively). Conclusions Contrast-enhanced CT might be a diagnostic alternative for EVNTs in cirrhotic patients, but its diagnostic performance was slightly weakened in secondary prophylaxis population. Additionally, contrast-enhanced CT may be considered for diagnosis of GVNTs in primary prophylaxis population without previous endoscopic variceal therapy and secondary prophylaxis population.


Introduction
Cirrhosis is the end stage of chronic liver disease, which is histologically characterized by fibrosis, scar, and regenerative nodules leading to structural deformation [1]. A major consequence of advanced cirrhosis is portal hypertension, which leads to the development of gastroesophageal varices (GEVs) [2]. Endoscopy should be performed at the time of first diagnosis of liver cirrhosis [3]. GEVs are observed in about 50% of patients with cirrhosis, and 8% of patients without GEVs develop them each year. Patients with no or small varices and without prior history of variceal bleeding should undergo endoscopic surveillance every 1-2 years. Bleeding from GEVs results in a mortality of 5-20% at 6 weeks. Endoscopic treatment, such as endoscopic variceal ligation (EVL) or tissue adhesive injection, is recommended for the management of high-risk varices and acute variceal bleeding [3][4][5]. However, patients undergoing endoscopic treatment for variceal bleeding have a high variceal recurrence rate of 8-48% [6,7], a rebleeding rate of 20-43%, and a bleeding related mortality of 19-34% [8]. Therefore, after endoscopic treatment, repeated EVL should be performed every 1-2 weeks until variceal obliteration. The first endoscopic surveillance for variceal recurrence should be performed within 1-3 months after variceal obliteration, and then endoscopic surveillance should be repeated every 6-12 months [5].
Despite endoscopy is the golden approach for diagnosis and surveillance of GEVs according to the current practice guideline and consensus, it is often limited by increased invasiveness, patients' discomfort and poor adherence, and high cost [9][10][11]. Recently, noninvasive blood tests have been used to diagnose GEVs [12,13], such as aspartate aminotransferase (AST) to platelet (PLT) ratio index (APRI), AST to alanine aminotransferase (ALT) ratio (AAR), fibrosis 4 index (FIB-4), Lok score, and King score. Contrast-enhanced computed tomography (CT), a conventional diagnostic imaging tool in patients with liver diseases, has also been explored for the assessment of GEVs [14][15][16][17]. Additionally, a combination of blood tests with imaging examination for screening GEVs, such as PLT count to spleen diameter ratio (PSR), has been frequently explored [18].
Notably, the performance of these diagnostic alternatives may be heterogeneous among different study populations. However, until now, no study has evaluated their diagnostic accuracy according to the patient characteristics [11]. For this reason, we conducted a retrospective observational study to evaluate the accuracy of blood tests, PSR, and contrastenhanced CT for diagnosing esophageal varices (EVs) and gastric varices (GVs) in cirrhotic patients with and without variceal bleeding or previous endoscopic variceal therapy.

Patients.
This was a single-center retrospective observational study on the basis of our prospective database regarding cirrhotic patients undergoing both contrast-enhanced CT and upper gastrointestinal endoscopy. This study was approved by the medical ethical committee of our hospital and the approval number was [k (2018) 08]. The patients' informed consents were waived. All patients consecutively admitted to our department from December 2014 to October 2018 were potentially eligible.
The inclusion criteria were as follows: (1) patients had a diagnosis of liver cirrhosis according to the medical history, clinical features, imaging, and/or histological results and (2) both contrast-enhanced CT and endoscopic examinations were performed at their admissions, and the time interval between the two examinations was within one month. Repeated admission was not excluded.
The exclusion criteria were as follows: (1) patients had a definite diagnosis of malignant tumors, (2) contrastenhanced CT was performed after endoscopic treatment at their admissions, and (3) contrast-enhanced CT images were not well preserved.

Groups.
According to the previous history of endoscopic treatment for variceal bleeding, history of gastrointestinal bleeding (GIB), and presence of acute upper gastrointestinal bleeding (AUGIB), the patients were divided into four groups: (1) Primary prophylaxis population (no history of endoscopic treatment, no history of GIB, and absence of AUGIB) (2) Acute bleeding population (no history of endoscopic treatment, but with presence of AUGIB, regardless of history of GIB) (3) Previous bleeding population (no history of endoscopic treatment, absence of AUGIB, but with a history of GIB) (4) Secondary prophylaxis population (a history of endoscopic treatment for variceal bleeding, but absence of GIB) As for the secondary prophylaxis population, the patients would be further excluded, if the time interval between prior endoscopic treatment and present admission was less than one month [19]. This is primarily because the esophagus and stomach lumen mucosa may not be fully recovered during a short postoperative period, which will cause a potential radiological artifact on CT images and influence its diagnostic performance.

Data Collection.
The data were collected as follows: age, sex, etiology of liver diseases, ascites, interval between prior endoscopic treatment and present admission, red blood cell (RBC), hemoglobin (Hb), white blood cell (WBC), PLT, total bilirubin (TBIL), direct bilirubin (DBIL), albumin (ALB), ALT, AST, alkaline phosphatase (AKP), γ-glutamine transferase (GGT), blood urea nitrogen (BUN), serum creatinine (SCr), prothrombin time (PT), activated partial thromboplastin time (APTT), and international normalized ratio (INR). The maximum diameter of the spleen was measured on axial contrast-enhanced CT images. The Child-Pugh [20] model for end-stage of liver disease (MELD) [21], APRI [22], AAR [23], FIB-4 [24], Lok [25], King [26], and PSR [27] scores were calculated as follows: Child − Pugh score = ALB score + TBIL score + INR score + ascites score + hepatic encephalopathy score, MELD score = 9:57 × ln Cr μmol/L ð Þ× 0:011 ½ Notably, they were blinded to the laboratory and endoscopic findings when the CT images were retrospectively analyzed. They independently evaluated the presence of GEVs. EVs or GVs were defined as enhancing lesions abutted the luminal surface of the esophageal or gastric wall or protruded into esophageal or gastric luminal space at the portal vein phases of contrast-enhanced CT images [28,29]. They also independently selected the CT layer with the maximum diameter of varices. In cases of any inconsistency in measuring the maximum diameter of varices between the two observers, a discussion with another investigator (XQ) was made until a consensus was achieved. Additionally, they evaluated the spleen and measured the maximum diameter of the spleen on contrast-enhanced CT images.  Table 2). The presence of EVs and diameter of EVs could be evaluated on CT in all of the 70 patients. The diameter of EVs measured on contrast-enhanced CT < 0:50 cm should be considered as the optimal cutoff value for ruling out the EVNTs. By using this cutoff value, 47.8% (32/67) of endoscopies were spared, and no (0/32) EVNTs was missed ( Figure 2(a)).
After a discussion among investigators, the presence of GVs could not be evaluated on CT in one patient and the diameter of GVs could not be measured on CT in 3 patients. The diameter of GVs measured on contrast-enhanced CT < 1:09 cm should be considered as the optimal cutoff value for ruling out the GVNTs. By using this cutoff value, 76.6% (49/64) of endoscopies were spared, but 4.1% (2/49) of GVNTs were missed (Figure 2(a)).
The presence of EVs and diameter of EVs could be evaluated on CT in all of the 38 patients. The diameter of EVs measured on contrast-enhanced CT < 0:38 cm should be considered as the optimal cutoff value for ruling out the EVNTs. By using this cutoff value, 10.5% (4/38) of endoscopies were spared, and no (0/4) EVNTs was missed ( Figure 2 After a discussion among investigators, the diameter of GVs could not be measured on CT in 3 patients. The diameter of GVs measured on contrast-enhanced CT < 1:01 cm should be considered as the optimal cutoff value for ruling out the GVNTs. By using this cutoff value, 45.7% (16/35) of endoscopies were spared, but 25% (4/16) of GVNTs were missed ( Figure 2(b)).
The presence of EVs and diameter of EVs could be evaluated on CT in all of the 67 patients. The diameter of EVs measured on contrast-enhanced CT < 0:46 cm should be considered as the optimal cutoff value for ruling out the EVNTs. By using this cutoff value, 12.1% (8/66) of endos-copies were spared, and no (0/8) EVNTs was missed ( Figure 2(c)).
After a discussion among investigators, the diameter of GVs could not be measured on CT in 3 patients (3/67). The diameter of GVs measured on contrast-enhanced CT < 0:95 cm should be considered as the optimal cutoff value for ruling out the GVNTs. By using this cutoff value, 21.9% (14/64) of endoscopies were spared, but 45.5% (5/14) of GVNTs were missed ( Figure 2(c)). As for EVs, only contrast-enhanced CT had statistically significant diagnostic performance; as for EVNTs, only contrast-enhanced CT and AAR score had statistically significant diagnostic performance; as for GVs, only contrast-enhanced CT had statistically significant diagnostic performance; as for GVNTs, only contrast-enhanced CT and FIB-4 score had statistically significant diagnostic performance (Table 2).

Secondary Prophylaxis
After a discussion among investigators, the diameter of EVs could not be measured on CT in one patient. The diameter of EVs measured on contrast-enhanced CT < 0:33 cm should be considered as the optimal cutoff value for ruling out the EVNTs. By using this cutoff value, 7.8% (8/103) of endoscopies were spared, and no (0/8) EVNTs was missed (Figure 2(d)).      After a discussion among investigators, the presence of GVs could not be evaluated on CT in 2 patients and the diameter of GVs could not be measured on CT in 2 patients. The diameter of GVs measured on contrast-enhanced CT < 1:11 cm should be considered as the optimal cutoff value for ruling out the GVNTs. By using this cutoff value, 56% (56/100) of endoscopies were spared, but 5.4% (3/56) of GVNTs were missed (Figure 2(d)).

Discussion
Currently, noninvasive diagnosis of GEVs is a hot topic. Severity of liver fibrosis is often in parallel with that of portal hypertension in compensated cirrhosis. Thus, the markers reflecting the severity of liver fibrosis are frequently used for noninvasive assessment of portal hypertension in such patients [10,31]. Considering that liver stiffness measured by transient elastography can stage liver fibrosis and PLT indicates portal hypertension, Baveno VI consensus has recommended that liver stiffness < 20 kPa combined with PLT > 150 × 10 9 /L should be a criterion for sparing endoscopy in compensated cirrhosis [4], and only a minority of patients within this Baveno VI criterion have a risk of variceal bleeding [32]. Researchers attempted to further improve its diagnostic accuracy by means of optimizing the thresholds of liver stiffness and PLT or establishing stepwise ruling-out and/or ruling-in strategies (Supplementary Table 2). Noninvasive approaches on the basis of Baveno VI criterion can accurately diagnose EVNTs with a missing rate of <5% [33][34][35][36][37][38][39][40][41][42][43]. Despite so, it should be noted that Baveno VI criterion should be appropriate for only patients with compensated cirrhosis without any history of gastrointestinal bleeding or endoscopic treatment. By comparison, few well-established tools have been employed for patients with advanced and decompensated cirrhosis, in whom extrahepatic factors, such as development of extrahepatic collaterals and splanchnic vasodilation, became more important for the progression of portal hypertension than intrahepatic resistance caused by liver fibrosis [44]. In this setting, we have for the first time evaluated the diagnostic accuracy of blood tests, PSR, and contrastenhanced CT for GEVs according to the severity of liver cirrhosis and portal hypertension, including patients without variceal bleeding (primary prophylaxis population), with variceal bleeding (acute bleeding population and previous bleeding population), and with history of endoscopic treatment for variceal bleeding (secondary prophylaxis population). Our previous meta-analysis demonstrated that APRI, AAR, FIB-4, and Lok scores had low to moderate diagnostic accuracy in predicting the presence of EVs and EVNTs in liver cirrhosis, and their AUCs were 0.6774-0.7885 and 0.7095-0.7448, respectively [12]. Notably, among the studies included in the meta-analysis, most of patients had wellpreserved liver function. By comparison, our previous observational study where a majority of patients were decompensated demonstrated that APRI, AAR, FIB-4, and Lok scores had low accuracy for EVs and EVNTs with AUCs of 0.539-0.567 and 0.506-0.544, respectively [13]. Similarly, our present observational study also confirmed that these blood tests were insufficient to replace endoscopy in diagnosing EVs, EVNTs, GVs, and GVNTs in advanced decompensated patients.
PSR had relatively high diagnostic accuracy in predicting the presence of EVs in compensated cirrhotic patients and its AUC was 0.85 [18]. The advantages of PSR as a potential diagnostic alternative for EVs can be explained by the fact that splenomegaly and hypersplenism are common clinical manifestations of portal hypertension, and the PSR model associates decreased PLT with splenomegaly [27,45]. By contrast, our present study suggested that PSR was unsatisfactory for prediction of GEVs. This might be related to the characteristics of our patients that a majority of patients in primary prophylaxis population group had Child-Pugh class B or C and all patients in 3 other groups (i.e., secondary prophylaxis population, acute bleeding population, and previous bleeding population) were decompensated with recent or previous bleeding. This was in consistency with the results of a previous study which also included patients receiving secondary prophylaxis and achieved only an AUC of 0.715 [46].
Our previous meta-analysis demonstrated that contrastenhanced CT had high diagnostic accuracy in predicting the presence of EVs, EVNTs, and GVs, and their AUCs were 0.8958, 0.9461, and 0.9127, respectively [14]. Similarly, another meta-analysis also confirmed that the AUCs were 0.86 and 0.95 in predicting the presence of EVs and GVs, respectively [15]. By comparison, our present study confirmed such high diagnostic accuracy of contrast-enhanced CT in predicting EVs and EVNTs and further suggested that no EVNTs would be missed according to the optimal cutoff value. However, the diagnostic performance of contrastenhanced CT was insufficient in secondary prophylaxis population.
Several pitfalls of contrast-enhanced CT scans for assessment of GEVs should be recognized. First, esophageal wall may form scars and stiffen after repeated endoscopic treatments, in which enhanced vascular shadows do not obviously protrude into esophageal lumen on contrast-enhanced CT images (Figure 3(a)). Second, during the endoscopic examinations, small EVs may be flattened after dilating esophageal lumen, thereby leading to a missed diagnosis (Figure 3(b)). Third, the images obtained at the portal vein phases of contrast-enhanced CT scans are inappropriately selected by radiological technicians, in which esophageal venous vessels cannot be obviously enhanced. Fourth, abdominal CT scans are selected for our present study, in which the lesions at middle and upper esophagus cannot be observed. Fifth, contrast-enhanced CT scans can detect GVs located deeply in gastric mucosa [29], which are hard to be distinguished from gastric mucosal folds by endoscopy. Sixth, when the gastric cavity is not fully expanded, small GVs do not protrude from the surface and cannot be differentiated from the gastric mucosa folds on CT images (Figure 3(c)). Seventh, some GVs appear as irregular vascular shadows on contrastenhanced CT images, thereby misjudging the maximum diameter of varices (Figure 3(d)). Several other advantages of contrast-enhanced CT scans should not be ignored, because it can simultaneously evaluate the severity of liver cirrhosis and its related complications, such as grade or quantification of ascites [47], thrombosis within portal vein system [48], portosystemic collaterals [49], and liver cancer [50], except for GEVs. On the other hand, the disadvantages of contrast-enhanced CT scans include the following. First, the risk of radiation will be increased. Second, contrast-enhanced CT is not applicable to patients with renal failure, hyperthyroidism, and hypersensitivity to contrast media. Third, RC sign is valuable for evaluating the severity of GEVs, but it cannot be observed on contrast-enhanced CT images.
Our study had several limitations. First, Western studies evaluated EVNTs by the size of EVs under endoscopy, and our study employed the Chinese guideline to identify EVNTs. Second, our patients were more severe and had a high prevalence of EVNTs. Because the prevalence of EVNTs should be inversely associated with the rate of spared endos-copy, the rate of sparing more endoscopy was relatively lower in our study ( Figure 4). Third, the present study was of the retrospective nature and performed at a single center. Fourth, the sample size was small in different study population, especially in acute bleeding population.
In conclusion, contrast-enhanced CT seemed to have higher diagnostic accuracy for EVs and EVNTs in cirrhotic patients as compared to APRI, AAR, FIB-4, FI, Lok, and King scores and PSR. Among the secondary prophylaxis population requiring repeated endoscopic surveillance, contrast-enhanced CT seemed to be the only useful diagnostic alternative for GEVs in cirrhotic patients. However, the potential pitfalls of contrast-enhanced CT, such as stiff and scarred esophagus, small or irregular vascular shadows, and technical errors, can decrease its diagnostic accuracy in secondary prophylaxis population.