- Email: [email protected]
Non-invasive diagnosis of esophageal varices in chronic liver diseases
Journal of Hepatology 1999; 31: M-873 Printed in Denmark All rights reserved Munksgaard Copenhagen
Copyright 0 European Association for the Study of the Liver 1999
Journalof Hepatology ISSN 0168-82 78
Non-invasive diagnosis of esophageal varices in chronic liver diseases Christophe
Pilette’, Frederic
Oberti’, Christophe AubC2, Marie Christine Rousselet3, Yves Gallois5, Herve Rifflet’ and Paul Cal&s’
Pierre Bedossa4,
‘Service d’Hkpato-Gastroentt!rologie, CHU, Angers, 2Service de Radiologie, CHU, Angers, 3Laboratoire d’Anatomie-Pathologique, CHV, Angers, 4Service d’dnatomie et Cytologic Pathologiques, CHU, Kremlin-Bict%re, ‘Laboratoire de Biochimie, CHU, Angers, France
Background/Aims: The primary prevention of bleeding from esophageal varices is a major therapeutic issue requiring early screening of esophageal varices. Our aim was to study the diagnostic accuracy of non-endoscopic means for the diagnosis of esophageal varices. Methods: Sixty-three clinical, biochemical, endoscopic and Doppler ultrasound variables were prospectively recorded in 207 consecutive patients with chronic liver disease. Diagnostic accuracy was evaluated by discriminant analysis, first globally using all variables with diagnostic accuracyr65% in univariate analysis, then by stepwise regression. Results: A) whole group (n=207), 1) diagnosis of esophageal varices: diagnostic accuracy was globally 81%, and 81% with 1 variable: irregular liver surface at ultrasound, 2) Diagnosis of large esophageal varices (grades 2+3): diagnostic accuracy was globally 80%, and 79% with 2 variables: prothrombin index, y-globulins. B) patients with cirrhosis (n=116), 1) diagnosis of esophageal varices: diagnostic accuracy
L
IVER FIBROSIS is
a common development in chronic liver disease. The most evolved stage of liver fibrosis is cirrhosis, which places the patient at risk of clinical complications such as portal hypertension (PHT). Variceal rupture is the second cause of death in cirrhosis (1). Early diagnosis is essential since cirrhosis is often revealed by complications - l/3 of gastrointestinal bleedings reveal cirrhosis - and the risk of variceal rupture is greatest in the 2 years following diagnosis (2). However, as there is clear evidence that the primary prevention of variceal rupture is effective in reducing Received I6 March; revised I June: accepted 8 June 1999 Correspondence: Paul Cal&s, Service d’HCpato-Gastroenterologie, CHU, 49033 Angers Cedex 01, France. Tel: 33 2 41 35 34 10. Fax: 33 2 41 35 41 19.
was globally 71%, and 72% with 2 variables: platelet count, prothrombin index, 2) diagnosis of large esophageal varices (grades 2+3): diagnostic accuracy was globally 71%, and 72% with 3 variables: platelet count, prothrombin index, spider naevi. The ROC curve showed that the best threshold for the diagnostic accuracy of platelet count was 160 G/l providing a sensitivity of 80% and a specificity of 58%. Platelet count 2260 G/l has a negative predictive value ~91%. Conclusions: Using a few non-endoscopic criteria, esophageal varices can be correctly diagnosed in 81% of patients with chronic liver disease and in 71% of patients with cirrhosis. These results show that the non-invasive screening of patients who are candidates for the primary prevention of variceal bleeding is possible, but should be improved before being used in a clinical setting.
Key words: Chronic liver disease; Cirrhosis; Diagnosis; Esophageal varices.
the death rate (3), the screening of esophageal varices (EV) is recommended (4). Screening takes place under two circumstances: at the initial diagnosis of cirrhosis, since EV are an independent predictive factor (5) and an early complication (2) of cirrhosis, and during the follow-up of patients with cirrhosis without EV at risk of bleeding at first examination. In this latter population, the occurrence of large EV can be predicted as a function of time to determine the frequency of endoscopic follow-up based on the initial grade of EV (4). On the other hand, it is possible to identify the population with a high probability of EV that requires confirmation by endoscopy (6). However, the clinical use of endoscopy is limited due to cost and discomfort, resulting in poor compliance (4). Recently, numerous techniques have been developed for the non-invasive diagnosis of cirrhosis (7) or portal 867
C. Pilette
et al.
hypertension (8). The aim of the present study was to evaluate the diagnostic accuracy of non-endoscopic means to diagnose EV in patients with chronic liver disease under conditions close to those of clinical prac-
Patients and Methods Patients
The 243 consecutive patients considered for this study were admitted to the hepatogastroenterology unit of the University Hospital in Angers, France, in 1994-5 for alcoholism or alcoholic liver disease, or for chronic hepatitis B or C infection. Patients were included who had drunk at least 50 g of alcohol per day for the past 5 years, who were positive for hepatitis B surface antigen or C serum markers, and who had had persistently elevated serum aminotransferases >1.5 N for at least 6 months or who had had previous liver complications (ascites, edema). Patients with a past history of gastrointestinal bleeding were not included. None of the patients had clinical, biological, ultrasonographic or histological evidence of other causes of chronic liver disease (Wilson’s disease, hemochromatosis, al-antitrypsin deficiency, biliary disease, auto-immune hepatitis, hepatocellular carcinoma). Except for these non-inclusion criteria, this study was performed under conditions close to clinical practice on a intentionto-diagnose principle. Thus, patients with incomplete ultrasonography (e.g. gas preventing a complete examination) were not excluded. Blood samples were taken at entry and a percutaneous (n= 210) or transjugular liver biopsy (n=33) was performed within 1 week. In the latter patients, the liver specimen was fragmented in 11 cases, making image analysis impossible. Among the 243 patients, 83 had viral liver disease and 160 alcoholic liver disease, and 116 patients had cirrhosis. Since no additional examinations were performed compared to clinical practice, informed consent was not obtained. Clinical evaluation
A full clinical examination was performed by a senior physician. The recorded variables were: age, sex, size, body weight (before paracentesis), mean alcohol consumption (g/d) before any eventual withdrawal, duration of alcohol abuse, alcohol withdrawal, duration of alcohol withdrawal, known duration of liver disease (time since the tirst clinical or biochemical abnormality suggesting chronic liver disease), palmar erythema, Dupuytren’s contracture, finger clubbing, splenomegaly (any palpable spleen), firm liver with a thin lower edge, abdominal venous collateral circulation, ascites, spider naevi (number), number connection test weighted as a function of age, and hepatic encephalopathy. Blood tests
Analyses of blood samples provided the following information: hemoglobin, mean corpuscular volume, lymphocyte count, platelet count, cholesterol, urea, creatinine, sodium, bilirubin, GGT, ASAT, ALAT, ASAT/ALAT, albumin, y-globulins, B-globulins, pr block, prothrombin index, apo Al, PGA score, AaM, PGAA score, PIIINP hyaluronate, TGF 81, laminin, and Child-Pugh score. Sera were kept at - 80°C for a maximum of 48 months for radioimmunoassays. Hyaluronate serum concentrations were determined with a radioimmunoassay technique (Kabi-Pharmacia RIA Diagnostics, Uppsala, Sweden). Laminin was measured using the RIA-gnost laminin Pl kit (CiS Diagnostic K.K.). Serum PIIINP concentrations were determined with a commercial RIA kit (RIA-gnost PIIIP kit, Hoechst, Tokyo, Japan). The TGF-/31 determination was performed with a Predicta TGF-/31 kit (Genzyme Diagnostics) by the ELISA method. Endoscopic evaluation
Endoscopic examinations were performed by two experienced endoscopists (either PC or FO); however, this examination was not available in 36 patients without cirrhosis (due to refusal, or a time interval 22 months between endoscopy and inclusion, or endoscopy performed in another center). These 36 patients were not included in the
868
statistical analysis. The following signs were recorded: erosions, portal hypertensive gastropathy (PHG), gastric varices and EV The classification and definition of endoscopic signs were based on the definition of Papazian et al. (9) for mosaic pattern (moderate PHG), the definition of McCormack et al. (10) for red spots (severe PHG), the definition of Sacchetti et al. (11) for erosions and a previously described classification for gastric varices (12). The grade of esophageal varices (grade 0 to 3) was evaluated according to a previously described classification (12). The interobserver agreement between the two observers was evaluated on a previous group of 168 patients with chronic liver disease (CLD) (5). Agreement was excellent for the presence of EV with a kappa index=O.88. Echo-Doppler
study
Examinations were performed in all patients by one of four investigators (one radiologist specialized in hepatology and three general radiologists) using an Acuson 128Xp/lOM apparatus according to a previously described technique (13) within 1 day before liver biopsy. The radiologists were blinded for the clinical diagnosis of liver disease. The probe used was a 3.5-2.5 MHz transducer and a 7.5 MHz linear transducer. The following 11 variables were selected based on the DA reported in the literature (14): hepatic length at the level of the right kidney and the middle of the clavicle in a sagittal plane; nodular aspect of the liver surface based on ultrasound of the left lobe with a linear high frequency transducer (7.5 Mhz); liver homogeneity; ratio of the caudate lobe/left lobe length on a sagittal scan which was considered abnormal if >0.33 (15); collateral circulation defined as any patent umbilical vein (i.e. diameter ~3 mm ) or a left gastric vein >5 mm (16) or any abnormal vein such as a spleno-renal shunt; respiratory variation of the splenic vein; spleen length; ascites; portal vein diameter; and maximum velocity of portal vein (17). These signs were evaluated in all patients. Histological assessment of the liver Semi-quantitative scores. The biopsy specimens were fixed in a form-
alin-alcohol-acetic acid solution and embedded in paraffin; 3 ethick sections were stained with hematoxylin-eosin-saffron, Masson’s trichrome and 0.1% picrosirius red solution. Fibrosis was graded by two independent pathologists according to a semi-quantitative score derived from the Metavir score (18). The score for fibrosis of the portal tract (PT) was as follows: 0: no fibrosis, 1: enlarged PT without septa, 2: enlarged PT with rare septa, 3: numerous septa without cirrhosis. 4: potential cirrhosis (for small specimens obtained by transjugular biopsy), 5: definite cirrhosis; this score was called the fibrosis score. The only difference from the Metavir score was the addition of a grade 4 for small specimens obtained by transjugular biopsy. The interobserver agreement was excellent (intra-class correlation coefficient: r=0.78). Other lesions (alcoholic hepatitis, steatosis, centrolobular vein fibrosis) were graded into 4 grades (none, discrete, moderate, severe). When the two pathologists did not agree, the specimens were reexamined with a binocular microscope to analyze discrepancies and reach a consensus. Quantitative score. Image analysis was performed on the same sections as those stained with picrosirius red solution for the semi-quantitative study with a Leica Quantimet 4570 image processor. The 2D patterns were measured by direct pixel counting on the binary images. In a given microscopic field, the surface of the section occupied by the hepatic parenchyma was used as the measuring frame and this surface area was considered referent area A. The fractional surface occupied by fibrosis was measured within the above defined frame and AA represented the percentage of fibrosis in the field. When using the motorized x-y stage facilities, the biopsy could be fully explored without any field overlapping. For a given patient, the percentage of fibrosis or the area of fibrosis could be obtained on n microscopic fields as (n 2 A.&/n. Statistical
analysis
Quantitative variables were expressed as mean&SD and were compared using Student’s t-test unless otherwise specified. Qualitative variables were expressed as percentages and were compared using the
Esophageal
x2 test. DA (percentage of correctly classified patients) were calculated by discriminant analysis. To assess the independent effect of qualitative or quantitative predictive variables on the DA of fibrosis (explained variable), we used discriminant analysis with forward stepwise addition of variables. Due to the high number of variables, only those with a DA ~70% or 65% in univariate analysis were selected in the multivariate analysis. To assess the independent effect of quantitative predictive variables on the prediction of EV, we used multiple linear regression analysis. As DA is dependent on the prevalence of the variable studied (fibrosis) in the population, several curves as a function of the best explanatory variables were constructed - ROC and DA - to select the appropriate variable threshold with respect to these informative indexes. An cz risk <5% for a two-sided test was considered statistically significant.
varice diagnosis
Child-Pugh classes in cirrhotic patients was: A: 50%, B: 24%, C: 26%. Cirrhosis was complicated (ascites and/or encephalopathy) in 45% of patients. Diagnosis of EV Univariclte analysis. As the screening of EV is mainly
relevant in patients with cirrhosis and for their large grade, results are presented as a function of these two characteristics. Large EV includes grades 2+3. The DA of variables with a DA 170% in at least one group for EV diagnosis is presented in Table 1. Multivariate analysis. The results according to each group are presented in Table 2. In the entire population with CLD, using the 11 significant variables provided by univariate analysis, the DA for EV or large EV obtained with multivariate analysis was about 80%. However, several variables selected by stepwise analysis suggested cirrhosis: liver surface at ultrasonography (US) and prothrombin index or y-globulins. Therefore, in an explanatory phase, we performed univariate analysis of histological variables (Table 3) predictive of EV and included histological variables in the same multivariate model to check whether the signifi-
Results Characteristics of population
Among the 207 patients included, 66% were male and the mean age was 50212 years. The prevalence of cirrhosis was significantly higher in patients with alcoholic CLD than in patients with viral CLD (60.5 vs 42%, respectively, ~~0.05). The distribution of EV was as follows: in the 91 patients with CLD, EV were present in lo%, all grade 1; in the 116 patients with cirrhosis EV were present in 72% with grade 1 in 28%, grade 2 in 42% and grade 3 in 2%. The distribution of
TABLE 1 Diagnostic accuracies (%) for esophageal varices (EV) in patients with chronic liver disease or cirrhosis according to univariate analysis. Diagnostic accuracies ~70% are depicted in bold characters Chronic liver disease All EV
Cirrhosis Large EV
All EV
Large EV
%
P
%
P
%
P
%
P
55
Clinical data Liver disease duration Dupuytren contraction Spider naevi Firm liver Splenomegaly Collateral circulation Ascites Encephalopathy
NS <10-s <1O-3
73 71 76.5 62 75 68 73 77
<10-s <0.05 <10-s <10-s co.05 <10-s <10-s <10-s
40 44 52 67 67 62 47 33
NS
64
61.5 69 13.5 58 70 66 57
NS NS NS NS NS NS NS
54 65 56 56 56 46 61
<10-l NS co.05 NS NS NS NS co.05
Echographic data Liver surface (irregular) Liver heterogeneity Spleen length Ascites Collateral circulation
65 64 63 67 63
<1O-3 <10-s NS <10-s <10-s
78 73 74 75 78
<10-s <1o-3 NS <10-s <10-s
48 53 50 50 43
NS NS NS NS co.05
63 63 51 60 61.5
CO.05 co.05 NS NS NS
Blood data Bilirubin Albumin y-globulins B-r block Prothrombin index Child-Pugh score PGA score PGAA score Hyaluronate Laminin
65 73 70 73 76 68 72 70 73 74
<10-s <10-s <1O-3 <1O-3 <1O-3 <1O-3 <10-s <10-s <1O-3 <1O-3
75 71 69 72 76 76 71 73 76 76
<10-z <1o-3 <10-s <10-s <10-s <10-s <10-s <10-s <10-s <10-s
60 56.5 51 63.5 61 47 62 64 52 58
NS NS NS NS co.05 NS NS NS NS NS
58 55 54 57 64 62 57 62 61 55
NS NS NS NS <10-s <10-a co.05 <0.05 <0.05 NS
869
C. Pilette
et al.
TABLE 2 Diagnostic accuracies and predictive values (%) of esophageal varices (EV) according to multivariate analysis in the different patient groups Chronic liver disease
Cirrhosis
EV
Large EV
EV
Large EV
Global analysis
DA: 81 (73-87) PPV: 53 (38-68) NPV: 96 (89-99) (n=ll)
DA: 80 (72.5-86) PPV: 72 (59-82) NPV: 87 (77-93) (n=ll)
DA: 71 (61-79) PPV: 84 (73-91) NPV: 48 (3264) (n=7)
DA: 71 (62-79) PPV: 65 (51-78) NPV 76 (63-86) (n=S)
Stepwise analysis
1st, liver surface: DA: 81 (73-87) PPV: 56 (36-75) NPV: 88 (79-93.5)
lst, prothrombin: DA: 76 (68-82.5) 2nd: y globulins: DA: 79 (72-85) PPV: 73 (60-83), NPV: 84 (74-91)
lst, platelet: DA: 67 (58-76) 2nd, prothrombin: DA: 72 (63~80), PPV: 85 (74.5-92) NPV: 50 (34-55)
lst, platelet count: DA: 67.5 (58-76) 2nd, prothrombin: DA: 69, 3rd: spider naevi: DA: 70 (61-78), PPV 64 (5&77), NPV: 75 (62285.5)
n indicates the number of variables, with diagnostic accuracy ~70% in univariate analysis (65% in the group of cirrhosis), included in the model. DA: diagnostic accuracy, PPV positive predictive value, NPV: negative predictive value. Figures into brackets indicate 95% confidence interval. Main data are presented in bold.
cant variables related to cirrhosis in fact predicted cirrhosis or fibrosis. For the presence of EV, the fibrosis score was isolated at the first step, whereas the fibrosis area was isolated at the first step for large EV (data not shown). This indicates that markers of EV in patients with CLD are markers of liver fibrosis. Therefore, we based our analysis mainly on the group of patients with cirrhosis (see Discussion). In the group of patients with cirrhosis, global DA was 71%, and platelet count was isolated at the first step and prothrombin index at the 2nd step either for all EV or large EV Thus, platelet count appeared to be the best single marker of EV or large EV, since other markers added little information (Table 2). The ROC curve of the platelet count for large EV is presented in Fig. 1. The best diagnostic threshold was set at 160 G/ 1 providing a sensitivity of 80% and a specificity of 58%. Fig. 2 is another ROC curve in which DA, negative predictive value (NPV) and positive predictive value (PPV) are expressed as a function of platelet count. The DA curve confirms that the highest DA was observed at 160 G/l. The PPV curve showed a clinically irrelevant shape. However, the NPV curve showed that
the higher the platelet count, the lower the probability of large EV, e.g. NPV was 91% for 260 G/l. In an explanatory phase using quantitative variables with a high DA from Table 1 as well as histological variables from Table 3 as explanatory variables and the grade of EV as an explained variable in multiple regression, global R* was 0.58 with 4 independent variables by stepwise analysis: 1st step: spider naevi (R*= 0.21) 2nd step: platelet count (R* =0.37), 3rd step: fibrosis area (R*=O.47), 4th step: liver length (US at right kidney, R*=0.51).
Discussion The prevalence of cirrhosis in developed countries ranges between 0.4% in the USA or France (19,20) and 1.1% in Italy (21). PHT and liver dysfunction are the most frequent complications of cirrhosis. The main
loo-
80 -
60 -
TABLE 3 Diagnostic accuracies (%) of histological data for esophageal varices (EV) in patients with chronic liver disease according to univariate analysis. Diagnostic accuracies ~70% are depicted in bold characters
CLV centrolobular vein.
870
.
W)
40-
Large EV
All EV
Fibrosis score Cirrhosis CLV fibrosis Steatosis Alcoholic hepatitis Fibrosis area
se
%
P
%
P
80 go 83 61 64 80
<10-s <10-s <10-s
69 69 72 57 60 80
<10-s <10-s <10-s <1o-3 co.05 <10-s
“,
.
0
I
20
.
.
40 1 - spe
I
60 (%)
,
80
100 -
Fig. I. ROC curve of platelet count (Gil) for the diagnosis of large esophageal varices. Se: sensitivity, Spe: specSficity.
Esophageal varice diagnosis Indexer
(%) lOO-
Cm-
60 -
v-u
.=
/v-d
I
I’ :
70 -
60 -
60I
40 : 0
100
200
300
Platelet count (G/l)
Fig. 2. Curves of diagnostic accuracy (DA), positive predictive value (PPV) and negative predictive value (NPV) as a function of platelet count (Gil).
consequence of PHT is the development of collateral circulation, including EV which are the most frequent and most dangerous complication. EV and variceal bleeding are an early complication of cirrhosis justifying early screening for EV (4). The usual means of di-
agnosing EV is upper gastro-intestinal endoscopy. However, endoscopy can be considered invasive due to the technique and level of discomfort. Non-invasive methods of diagnosis of considerable interest have been developed in the past few years. In CLD, the main objective is the non-invasive diagnosis of fibrosis or cirrhosis (7,22), which can be obtained by a combination of clinical signs, biochemical data such as serum markers of fibrosis (7), radiology such as Doppler US (14) and finally by upper gastro-intestinal endoscopy since EV and PHT gastropathy have a high (88%) DA for the diagnosis of cirrhosis (5). Is the non-endoscopic diagnosis of EV in CLD relevant? Our results indicate that markers of EV in patients with CLD are markers of liver fibrosis. Moreover, the role of EV as a marker of cirrhosis in CLD was confirmed with an 81% DA in the present CLD population (7). Thus, the non-endoscopic diagnosis of EV does not seem to be relevant in patients with CLD since: a) non-endoscopic markers of EV are the noninvasive markers of cirrhosis in that population, and b) the endoscopic diagnosis of EV is relevant for the non-invasive diagnosis of cirrhosis which is a more important clinical issue. On the other hand, although there is a strong relationship between EV and cirrhosis, EV mainly need to be diagnosed in patients with cirrhosis. First, EV are present in 2/3 to 3/4 of patients
TABLE 4 Diagnostic signs of esophageal varices (EV) in the literature 1st author (ref)
Year
Liver disease
Variables ?I
Variable type
Univariate analysis
Multivariate analysis
EV (“XI)
Lavergne (27)
1997
Cirrhosis?
52
12
Ascites
_
79
CLD
32
_
28
Cirrhosis HCV
50
5
_
ss
Chalasani (30)
1998
Cirrhosis pre-liver transplantation
346
_
Clinics Blood tests
Cholinesterase Spleen length Hepatic vein waveform, portal vein diameter, congestion index _
_
1997
Clinics Blood tests US Blood tests US Doppler US Doppler
Zeijen (28)
1994
Gorka (29)
Garcia-Tsao (31)
1997
Compensated cirrhosis
180
15
Clinics Blood tests
Khishimoto (32)
1998
47
3
US Doppler
Present study
1999
Cirrhosis, idiopathic PHT 1. CLD
Age, sex, Child, spider naevi, bili, alb, PT, Hb, WBC, platelet Esophageal wall
65
Clinics Blood tests US Doppler
1. n=49 See text 2. n=7
2. Cirrhosis
Patients n
1. 207 2. 116
Splenomegaly, platelet count NASH Spider naevi, alb, platelet
_
16
64 1. Liver surface
1. 45
2. Platelet count, prothrombin index, spider naevi
2. 72
CLD: chronic liver disease, US: ultrasonography, -: not available, PHT: portal hypertension.
871
C. Pilette et al.
with cirrhosis (2) and, second, only large EV require primary prevention. Therefore, the non-invasive diagnosis of EV, especially large-sized EV, is clinically relevant. In the literature, the non-invasive diagnosis of EV has already been studied. A clear distinction must be made between studies evaluating the development of EV and those evaluating the diagnosis of EV The former studies the risk factors for the occurrence or the evolution of EV in longitudinal studies (4,6,23,24); the latter studies the markers of the presence of EV in a horizontal study. In this last category, one can distinguish direct markers of EV such as radiology (25), e.g. barium swallow (26) and indirect markers that are listed in Table 4 (27-32). Most of these papers were preliminary reports as abstracts. In the present study, we found an acceptable DA for the diagnosis of EV in patients with CLD. To be of clinical interest, the DA must be >75 % since this is the limit between (50%) and the perfect DA (100%). However, this was obtained with Doppler US (liver surface) which has some limitations, mainly anatomical, preventing examination. The lower DA for EV than that observed for cirrhosis (7) can be explained by the presence of EV in some patients without cirrhosis and the absence of EV in some patients with cirrhosis. The former is attributable to false-negative liver biopsies or to observer variability. Indeed, all EV were grade 1 in non-cirrhotic patients and this grade is responsible for the highest observer variability (12). For these reasons, as well as the clinical importance of the diagnosis of cirrhosis mentioned above, the noninvasive diagnosis of EV is more relevant in patients with cirrhosis. Moreover, at present, only the screening of large EV is clinically relevant. For this condition, three independent markers were disclosed by multivariate analysis, in decreasing order of importance: platelet count, prothrombin index and spider naevi. Platelet count has also been shown to be an independent marker in two other studies with multivariate analysis (30,31). It can be considered a marker of PHT in relation to the hypersplenism secondary to PHT The ROC curve showed that a platelet count fixed at 160 G/l was the best cut-off for the DA of cirrhosis. The ROC curve also showed that the presence of large EV is improbable if cirrhotic patients have a platelet count ~260 G/l (negative predictive value ~91%). The prothrombin index expresses prothrombin time as a percentage, and is more reproducible than time (33). In our experience, the prothrombin index is an excellent indirect marker of liver fibrosis or cirrhosis (7,34,35). Spider naevi were also shown to be an independent diagnostic factor for large EV in the present study. This 872
factor has already been shown to be related to EV, large EV (36,37) or variceal bleeding (37). The present results differ from the recommendations of a recent expert conference (38) which stated that EV should be screened: a) in all cirrhotic patients who are Child’s class B or C, and b) in Child’s A patients where there is clinical evidence of PHT, e.g. a low platelet count (cl40 G/l), an enlarged portal vein diameter (~13 mm), or evidence of collateral circulation at ultrasound. However, this statement is derived from a preliminary report where the methodology was not stated (39). Moreover, the sensitivity of a platelet count 1140 G/l was only 41% in Child’s A patients (39). In rats with PHT, multivariate analysis has shown that the area of liver fibrosis is the only predictive factor of the degree of porto-systemic shunts (40). Table 2 shows that the highest DA for the diagnosis of EV were related to histological variables of fibrosis. However, other serum markers of liver fibrosis than prothrombin index had a fairly low DA for EV in patients with cirrhosis. It should be noted that these markers also have a low diagnostic value for the level of portal pressure (8). Finally, the DA for EV or large EV in patients with cirrhosis was 72 and 70%, respectively. These figures are not clinically relevant. Other variables should be evaluated to improve this DA. Recently, a sensitivity of 92% and a specificity of 100% were observed for the detection of large EV in patients with cirrhosis due to hepatitis C virus by measuring US waveform in the hepatic veins (29). These promising data must be confirmed. In conclusion, by using a few non-endoscopic criteria, EV can be correctly diagnosed in 81% of patients with CLD and in 71% of patients with cirrhosis. These results show that the non-invasive screening of patients who are candidates for the primary prevention of variceal bleeding is possible, but should be improved before a clinical application can be recommended.
Acknowledgements We thank Pr. Daniel Chappard for his contribution to image analysis and MS Dale Roche for her contribution.
References 1. Schlichting P Christensen E, Fauerholdt L, Poulsen H, Juhl E, Tygstrup N. Main causes of death in cirrhosis. Stand J Gastroenter01 1983; 18: 881-8. 2. Calts P Pascal JP Histoire naturelle des varices cesophagiennes au tours de la cirrhose (de la naissance a la rupture). Gastroenter01 Clin Biol 1988; 12: 245-54. 3. Poynard T, Cal&s P Pasta L, Ideo G, Pascal JP, Pagliaro L, et al. and France-Italian Multicenter Study Group. Beta-adrenergic-
Esophageal varice diagnosis antagonist drugs in the prevention of gastrointestinal bleeding in patients with cirrhosis and esophageal varices. N Engl J Med 1991; 324: 1532-8. 4. Cal& P Desmorat H, Vine1 JP Caucanas JR Ravaud A, Germ P et al. Incidence of large esophageal varices in patients with cirrhosis. Application to prophylaxis of first bleeding. Gut 1990; 31: 1298-302. 5. Oberti F, Burtin P MaIga MY, Valsesia E, Pilette C, Cal&s I? Gastroesophageal endoscopic signs of cirrhosis: independent diagnostic accuracy, interassociation and relationship to etiology and hepatic dysfunction. Gastrointest Endosc 1998; 48: 1485?. 6. Angulo P, Malinchoc M, Therneau T, Jorgensen R, Dickson ER, Lindor KD. Predicting esophageal varices in patients with primary biliary cirrhosis (PBC) [abstract]. Gastroenterology 1997; 112: A1213. 7. Oberti F, Valsesia E, Pilette C, Rousselet MC, Bedossa P, Aube C, et al. Non Invasive diagnosis of hepatic fibrosis or cirrhosis. Gastroenterology 1997; 113: 1609-16. 8. Bahr MJ, Baker KHW, Horn W, Gtinzler V Manns MP Serum laminin Pl levels do not reflect critically elevated portal pressure in patients with liver cirrhosis. Hepata-Gastroenterology 1997; 44: 1200-5. 9. Papazian A, Braillon A, Dupas JL, Sevenet E Capron JI? Portal hypertensive gastric mucosa: an endoscopic study. Gut 1986; 27: 1199-203. 10. McCormack TT, Sims J, Eyre-Brook J, Kennedy H, Goepel J, Johnson AG, et al. Gastric lesions in portal hypertension; inflammatory gastritis or congestive gastropathy. Gut 1985; 26: 1226-32. 11. Sacchetti C, Capelo M, Rebecchi P, Roncucci L, Zanghieri G, Tripodi, et al. Frequency of upper gastroihtestinal lesions in patients with liver cirrhosis. Dig Dis Sci 1988; 33: 1218-22. 12. Cal&s P, Zabotto B, Meskens C, Caucanas JP Vine1 JP, Desmorat H, et al. Gastroesophageal endoscopic features in cirrhosis: observer variability, interassociations and relationship to hepatic dysfunction. Gastroenterology 1990; 98: 15662. 13. Tissot 0, AubC C, Namour A, Blanc F, Dauver A, Cal&s I? Semiologie echographique de la cirrhose: concordance inter-observatcur. Gastroenterol Clin Biol 1995; 19: 291-6. 14. AubC C, Oberti F, Korali N, Namour MA, Loisel D, Tanguy JY et al. Echographic diagnosis of hepatic fibrosis or cirrhosis. J Hepatol 1999; Xl: 472-8. 15. Seitz JF, Boustitre C, Maurin P, Aimino R, Durbec JP, Botta D, et al. Evaluation de l’ultrasonographie dans le diagnostic des cirrhoses. Etude retrospective de 100 examens consecutifs. Gastroenterol Clin Biol 1983; 7: 734-9. 16. Subramanyam BR, Balthazar EJ, Madamba MR, Raghavendra BN, Horii SC, Lefleur RS. Sonography of portosystemic venous collaterals in portal hypertension. Radiology 1983; 146: 1616. 17. Cioni G, D’Alimonte P, Cristani A, Ventura P, Abbati G, Tincani E, et al. Duplex-doppler assessment of cirrhosis in patients with chronic compensated liver disease. J Gastroenterol Hepatol 1992; 7: 3824. 18. Bedossa P, Bioulac-Sage P Callard P, Chevallier M, Degott C, Deugnier Y, et al. Intraobserver and interobserver variations in liver biopsy interpretation in patients with chronic hepatitis C. Hepatology 1994; 20: 15-20. 19. Quinn PG, Johnston DE. Detection of chronic liver disease: costs and benefits. Gastroenterologist 1997; 5: 58-77. 20. Cal&s I! Epidemiologic et pronostic de la cirrhose. Contours Medical 1995; 117: 2707-l 1. 21. Bellentani S, Tiribelli C, Saccoccio G, Sodde M, Fratti N, De Martin C, et al. Prevalence of chronic liver disease in the general population of Northern Italy: the Dionysos study. Hepatology 1994; 20: 1442-9. 22. Guyader D, Jacquelinet C, Moirand R, Turlin B, Mendler MH,
23.
24.
25. 26.
27.
28.
29.
30.
31.
32.
33.
34.
35.
36. 37.
38.
39.
40.
Chaperon J, et al. Noninvasive prediction of fibrosis in C282Y homozyous hemochromatosis. Gastroenterology 1998; 115: 92936. Zoli M, Merckel C, Magalotti D, Gueli D, Grimaldi M, Gatta A, et al. Natural history of cirrhotic patients with small esophageal varices: a prospective study [abstract]. Hepatology 1998; 28: 409A. Gores GJ, Wiesner RH, Dickson ER, Zinsmeister AR, Jorgensen RA, Langworthy A. Prospective evaluation of esophageal varices in primary biliary cirrhosis: development, natural history, and influence on survival. Gastroenterology 1989; 96: 1552-9. Brown JJ, Naylor MJ, Yagan N. Imaging of hepatic cirrhosis. Radiology 1997; 202: 1-16. Ginai AZ, Van Buuren HR, Hop WCP Schalm SW. Oesophageal varices: how reliable is a barium swallow ? Br J Radio1 1993; 66: 322-6. Lavergne J, Molina E, Reddy KR, Jeffers L, Leon R, Nader AK, et al. Ascites predicts the presence of high grade varices by screening gastroscopy [abstract]. Gastrointest Endosc 1997; 45: AB187. Zeijen RNM, Caenepeel E Stockbrttgger RW, Arends JW, Oei TK. Prediction of esophageal varices in liver disease; preliminary results [abstract]. Gastroenterology 1994; 106: A1013. Gorka W, Al Mulla A, Al Sebayel M, Altraif I, Gorka TS. Qualitative hepatic venous Doppler sonography versus portal flowmetry in predicting the severity of esophageal varices in hepatitis C cirrhosis. Am J Roentgen01 1997; 169: 511-5. Chalasani N, Imperiale T, Ismail A, Sood G, Wilcox CM, Kwo P et al. A predictive index for determining the risk for presence of large esophageal variccs in patients with cirrhosis [abstract]. Gastroenterology 1998; 114: A1222. Garcia-Tsao G, Escorsell A, Zakko M, Patch D, Matloff D, Grace N. Predicting the presence of significant portal hypertension and varices in compensated cirrhotic patients [abstract]. Hepatology 1997; 26: 360A. Khishimoto R, Chen MH, Ogawa H, Wakabayashi MN, Kogutt MS. Esophageal varices: evaluation with transabdominal US. Radiology 1998; 206: 647-50. Robert A, Chazouilleres 0. Prothrombin time in liver failure: time, ratio, activity percentage, or international normalized ratio? Hepatology 1996; 24: 1392-4. Pilette C, Rousselet MC, Bedossa E Chappard D, Oberti F, Rifllet H, et al. Histopathological evaluation of liver fibrosis: quantitative image analysis vs semi-quantitative scores. Comparison with serum markers. J Hepatol 1998; 28: 439-46. Pilette C, Croquet V, Oberti F, AubC C, Rousselet MC, Bedossa R et al. Prothrombin index is an accurate indirect marker of liver fibrosis or cirrhosis [abstract]. Hepatology 1998; 28(4Pt2): 385A. Weissberg JI, Stein DET, Fogel M, et al. Physical examination predicts the presence of varices. Surv Dig Dis 1984; 2: 141-55. Foutch PG, Sullivan JA, Gaines JA, Sanowski RA. Cutaneous vascular spiders in cirrhotic patients: correlation with hemorrhage from esophageal varices. Am J Gastroenterol 1988; 83: 7236. Grace ND, Groszmann RJ, Garcia-Tsao G, Burroughs AK, Pagliaro L, Makuch RW, et al. Portal hypertension and variceal bleeding: an AASLD single topic symposium. Hepatology 1998; 28: 868-80. Pagliaro L, D’Amico G, Pasta L, Politi E Vizzini G, Traina M, et al. Portal hypertension in cirrhosis: natural history. In: Bosch J, Groszmann RJ, editors. Portal Hypertension: Pathophysiology and Treatment. Oxford: Blackwell Scientific Publications; 1994. p. 72-92. Oberti F, Pilette C, Rifflet H, Maiga M, Moreau A, Gallois Y, et al. Effects of simvastatin, pentoxyfylline and spironolactone on hepatic fibrosis and portal hypertension in cirrhotic rats. J Hepato1 1997; 26: 1363-71.
873
Copyright 0 European Association for the Study of the Liver 1999
Journalof Hepatology ISSN 0168-82 78
Non-invasive diagnosis of esophageal varices in chronic liver diseases Christophe
Pilette’, Frederic
Oberti’, Christophe AubC2, Marie Christine Rousselet3, Yves Gallois5, Herve Rifflet’ and Paul Cal&s’
Pierre Bedossa4,
‘Service d’Hkpato-Gastroentt!rologie, CHU, Angers, 2Service de Radiologie, CHU, Angers, 3Laboratoire d’Anatomie-Pathologique, CHV, Angers, 4Service d’dnatomie et Cytologic Pathologiques, CHU, Kremlin-Bict%re, ‘Laboratoire de Biochimie, CHU, Angers, France
Background/Aims: The primary prevention of bleeding from esophageal varices is a major therapeutic issue requiring early screening of esophageal varices. Our aim was to study the diagnostic accuracy of non-endoscopic means for the diagnosis of esophageal varices. Methods: Sixty-three clinical, biochemical, endoscopic and Doppler ultrasound variables were prospectively recorded in 207 consecutive patients with chronic liver disease. Diagnostic accuracy was evaluated by discriminant analysis, first globally using all variables with diagnostic accuracyr65% in univariate analysis, then by stepwise regression. Results: A) whole group (n=207), 1) diagnosis of esophageal varices: diagnostic accuracy was globally 81%, and 81% with 1 variable: irregular liver surface at ultrasound, 2) Diagnosis of large esophageal varices (grades 2+3): diagnostic accuracy was globally 80%, and 79% with 2 variables: prothrombin index, y-globulins. B) patients with cirrhosis (n=116), 1) diagnosis of esophageal varices: diagnostic accuracy
L
IVER FIBROSIS is
a common development in chronic liver disease. The most evolved stage of liver fibrosis is cirrhosis, which places the patient at risk of clinical complications such as portal hypertension (PHT). Variceal rupture is the second cause of death in cirrhosis (1). Early diagnosis is essential since cirrhosis is often revealed by complications - l/3 of gastrointestinal bleedings reveal cirrhosis - and the risk of variceal rupture is greatest in the 2 years following diagnosis (2). However, as there is clear evidence that the primary prevention of variceal rupture is effective in reducing Received I6 March; revised I June: accepted 8 June 1999 Correspondence: Paul Cal&s, Service d’HCpato-Gastroenterologie, CHU, 49033 Angers Cedex 01, France. Tel: 33 2 41 35 34 10. Fax: 33 2 41 35 41 19.
was globally 71%, and 72% with 2 variables: platelet count, prothrombin index, 2) diagnosis of large esophageal varices (grades 2+3): diagnostic accuracy was globally 71%, and 72% with 3 variables: platelet count, prothrombin index, spider naevi. The ROC curve showed that the best threshold for the diagnostic accuracy of platelet count was 160 G/l providing a sensitivity of 80% and a specificity of 58%. Platelet count 2260 G/l has a negative predictive value ~91%. Conclusions: Using a few non-endoscopic criteria, esophageal varices can be correctly diagnosed in 81% of patients with chronic liver disease and in 71% of patients with cirrhosis. These results show that the non-invasive screening of patients who are candidates for the primary prevention of variceal bleeding is possible, but should be improved before being used in a clinical setting.
Key words: Chronic liver disease; Cirrhosis; Diagnosis; Esophageal varices.
the death rate (3), the screening of esophageal varices (EV) is recommended (4). Screening takes place under two circumstances: at the initial diagnosis of cirrhosis, since EV are an independent predictive factor (5) and an early complication (2) of cirrhosis, and during the follow-up of patients with cirrhosis without EV at risk of bleeding at first examination. In this latter population, the occurrence of large EV can be predicted as a function of time to determine the frequency of endoscopic follow-up based on the initial grade of EV (4). On the other hand, it is possible to identify the population with a high probability of EV that requires confirmation by endoscopy (6). However, the clinical use of endoscopy is limited due to cost and discomfort, resulting in poor compliance (4). Recently, numerous techniques have been developed for the non-invasive diagnosis of cirrhosis (7) or portal 867
C. Pilette
et al.
hypertension (8). The aim of the present study was to evaluate the diagnostic accuracy of non-endoscopic means to diagnose EV in patients with chronic liver disease under conditions close to those of clinical prac-
Patients and Methods Patients
The 243 consecutive patients considered for this study were admitted to the hepatogastroenterology unit of the University Hospital in Angers, France, in 1994-5 for alcoholism or alcoholic liver disease, or for chronic hepatitis B or C infection. Patients were included who had drunk at least 50 g of alcohol per day for the past 5 years, who were positive for hepatitis B surface antigen or C serum markers, and who had had persistently elevated serum aminotransferases >1.5 N for at least 6 months or who had had previous liver complications (ascites, edema). Patients with a past history of gastrointestinal bleeding were not included. None of the patients had clinical, biological, ultrasonographic or histological evidence of other causes of chronic liver disease (Wilson’s disease, hemochromatosis, al-antitrypsin deficiency, biliary disease, auto-immune hepatitis, hepatocellular carcinoma). Except for these non-inclusion criteria, this study was performed under conditions close to clinical practice on a intentionto-diagnose principle. Thus, patients with incomplete ultrasonography (e.g. gas preventing a complete examination) were not excluded. Blood samples were taken at entry and a percutaneous (n= 210) or transjugular liver biopsy (n=33) was performed within 1 week. In the latter patients, the liver specimen was fragmented in 11 cases, making image analysis impossible. Among the 243 patients, 83 had viral liver disease and 160 alcoholic liver disease, and 116 patients had cirrhosis. Since no additional examinations were performed compared to clinical practice, informed consent was not obtained. Clinical evaluation
A full clinical examination was performed by a senior physician. The recorded variables were: age, sex, size, body weight (before paracentesis), mean alcohol consumption (g/d) before any eventual withdrawal, duration of alcohol abuse, alcohol withdrawal, duration of alcohol withdrawal, known duration of liver disease (time since the tirst clinical or biochemical abnormality suggesting chronic liver disease), palmar erythema, Dupuytren’s contracture, finger clubbing, splenomegaly (any palpable spleen), firm liver with a thin lower edge, abdominal venous collateral circulation, ascites, spider naevi (number), number connection test weighted as a function of age, and hepatic encephalopathy. Blood tests
Analyses of blood samples provided the following information: hemoglobin, mean corpuscular volume, lymphocyte count, platelet count, cholesterol, urea, creatinine, sodium, bilirubin, GGT, ASAT, ALAT, ASAT/ALAT, albumin, y-globulins, B-globulins, pr block, prothrombin index, apo Al, PGA score, AaM, PGAA score, PIIINP hyaluronate, TGF 81, laminin, and Child-Pugh score. Sera were kept at - 80°C for a maximum of 48 months for radioimmunoassays. Hyaluronate serum concentrations were determined with a radioimmunoassay technique (Kabi-Pharmacia RIA Diagnostics, Uppsala, Sweden). Laminin was measured using the RIA-gnost laminin Pl kit (CiS Diagnostic K.K.). Serum PIIINP concentrations were determined with a commercial RIA kit (RIA-gnost PIIIP kit, Hoechst, Tokyo, Japan). The TGF-/31 determination was performed with a Predicta TGF-/31 kit (Genzyme Diagnostics) by the ELISA method. Endoscopic evaluation
Endoscopic examinations were performed by two experienced endoscopists (either PC or FO); however, this examination was not available in 36 patients without cirrhosis (due to refusal, or a time interval 22 months between endoscopy and inclusion, or endoscopy performed in another center). These 36 patients were not included in the
868
statistical analysis. The following signs were recorded: erosions, portal hypertensive gastropathy (PHG), gastric varices and EV The classification and definition of endoscopic signs were based on the definition of Papazian et al. (9) for mosaic pattern (moderate PHG), the definition of McCormack et al. (10) for red spots (severe PHG), the definition of Sacchetti et al. (11) for erosions and a previously described classification for gastric varices (12). The grade of esophageal varices (grade 0 to 3) was evaluated according to a previously described classification (12). The interobserver agreement between the two observers was evaluated on a previous group of 168 patients with chronic liver disease (CLD) (5). Agreement was excellent for the presence of EV with a kappa index=O.88. Echo-Doppler
study
Examinations were performed in all patients by one of four investigators (one radiologist specialized in hepatology and three general radiologists) using an Acuson 128Xp/lOM apparatus according to a previously described technique (13) within 1 day before liver biopsy. The radiologists were blinded for the clinical diagnosis of liver disease. The probe used was a 3.5-2.5 MHz transducer and a 7.5 MHz linear transducer. The following 11 variables were selected based on the DA reported in the literature (14): hepatic length at the level of the right kidney and the middle of the clavicle in a sagittal plane; nodular aspect of the liver surface based on ultrasound of the left lobe with a linear high frequency transducer (7.5 Mhz); liver homogeneity; ratio of the caudate lobe/left lobe length on a sagittal scan which was considered abnormal if >0.33 (15); collateral circulation defined as any patent umbilical vein (i.e. diameter ~3 mm ) or a left gastric vein >5 mm (16) or any abnormal vein such as a spleno-renal shunt; respiratory variation of the splenic vein; spleen length; ascites; portal vein diameter; and maximum velocity of portal vein (17). These signs were evaluated in all patients. Histological assessment of the liver Semi-quantitative scores. The biopsy specimens were fixed in a form-
alin-alcohol-acetic acid solution and embedded in paraffin; 3 ethick sections were stained with hematoxylin-eosin-saffron, Masson’s trichrome and 0.1% picrosirius red solution. Fibrosis was graded by two independent pathologists according to a semi-quantitative score derived from the Metavir score (18). The score for fibrosis of the portal tract (PT) was as follows: 0: no fibrosis, 1: enlarged PT without septa, 2: enlarged PT with rare septa, 3: numerous septa without cirrhosis. 4: potential cirrhosis (for small specimens obtained by transjugular biopsy), 5: definite cirrhosis; this score was called the fibrosis score. The only difference from the Metavir score was the addition of a grade 4 for small specimens obtained by transjugular biopsy. The interobserver agreement was excellent (intra-class correlation coefficient: r=0.78). Other lesions (alcoholic hepatitis, steatosis, centrolobular vein fibrosis) were graded into 4 grades (none, discrete, moderate, severe). When the two pathologists did not agree, the specimens were reexamined with a binocular microscope to analyze discrepancies and reach a consensus. Quantitative score. Image analysis was performed on the same sections as those stained with picrosirius red solution for the semi-quantitative study with a Leica Quantimet 4570 image processor. The 2D patterns were measured by direct pixel counting on the binary images. In a given microscopic field, the surface of the section occupied by the hepatic parenchyma was used as the measuring frame and this surface area was considered referent area A. The fractional surface occupied by fibrosis was measured within the above defined frame and AA represented the percentage of fibrosis in the field. When using the motorized x-y stage facilities, the biopsy could be fully explored without any field overlapping. For a given patient, the percentage of fibrosis or the area of fibrosis could be obtained on n microscopic fields as (n 2 A.&/n. Statistical
analysis
Quantitative variables were expressed as mean&SD and were compared using Student’s t-test unless otherwise specified. Qualitative variables were expressed as percentages and were compared using the
Esophageal
x2 test. DA (percentage of correctly classified patients) were calculated by discriminant analysis. To assess the independent effect of qualitative or quantitative predictive variables on the DA of fibrosis (explained variable), we used discriminant analysis with forward stepwise addition of variables. Due to the high number of variables, only those with a DA ~70% or 65% in univariate analysis were selected in the multivariate analysis. To assess the independent effect of quantitative predictive variables on the prediction of EV, we used multiple linear regression analysis. As DA is dependent on the prevalence of the variable studied (fibrosis) in the population, several curves as a function of the best explanatory variables were constructed - ROC and DA - to select the appropriate variable threshold with respect to these informative indexes. An cz risk <5% for a two-sided test was considered statistically significant.
varice diagnosis
Child-Pugh classes in cirrhotic patients was: A: 50%, B: 24%, C: 26%. Cirrhosis was complicated (ascites and/or encephalopathy) in 45% of patients. Diagnosis of EV Univariclte analysis. As the screening of EV is mainly
relevant in patients with cirrhosis and for their large grade, results are presented as a function of these two characteristics. Large EV includes grades 2+3. The DA of variables with a DA 170% in at least one group for EV diagnosis is presented in Table 1. Multivariate analysis. The results according to each group are presented in Table 2. In the entire population with CLD, using the 11 significant variables provided by univariate analysis, the DA for EV or large EV obtained with multivariate analysis was about 80%. However, several variables selected by stepwise analysis suggested cirrhosis: liver surface at ultrasonography (US) and prothrombin index or y-globulins. Therefore, in an explanatory phase, we performed univariate analysis of histological variables (Table 3) predictive of EV and included histological variables in the same multivariate model to check whether the signifi-
Results Characteristics of population
Among the 207 patients included, 66% were male and the mean age was 50212 years. The prevalence of cirrhosis was significantly higher in patients with alcoholic CLD than in patients with viral CLD (60.5 vs 42%, respectively, ~~0.05). The distribution of EV was as follows: in the 91 patients with CLD, EV were present in lo%, all grade 1; in the 116 patients with cirrhosis EV were present in 72% with grade 1 in 28%, grade 2 in 42% and grade 3 in 2%. The distribution of
TABLE 1 Diagnostic accuracies (%) for esophageal varices (EV) in patients with chronic liver disease or cirrhosis according to univariate analysis. Diagnostic accuracies ~70% are depicted in bold characters Chronic liver disease All EV
Cirrhosis Large EV
All EV
Large EV
%
P
%
P
%
P
%
P
55
Clinical data Liver disease duration Dupuytren contraction Spider naevi Firm liver Splenomegaly Collateral circulation Ascites Encephalopathy
NS <10-s <1O-3
73 71 76.5 62 75 68 73 77
<10-s <0.05 <10-s <10-s co.05 <10-s <10-s <10-s
40 44 52 67 67 62 47 33
NS
64
61.5 69 13.5 58 70 66 57
NS NS NS NS NS NS NS
54 65 56 56 56 46 61
<10-l NS co.05 NS NS NS NS co.05
Echographic data Liver surface (irregular) Liver heterogeneity Spleen length Ascites Collateral circulation
65 64 63 67 63
<1O-3 <10-s NS <10-s <10-s
78 73 74 75 78
<10-s <1o-3 NS <10-s <10-s
48 53 50 50 43
NS NS NS NS co.05
63 63 51 60 61.5
CO.05 co.05 NS NS NS
Blood data Bilirubin Albumin y-globulins B-r block Prothrombin index Child-Pugh score PGA score PGAA score Hyaluronate Laminin
65 73 70 73 76 68 72 70 73 74
<10-s <10-s <1O-3 <1O-3 <1O-3 <1O-3 <10-s <10-s <1O-3 <1O-3
75 71 69 72 76 76 71 73 76 76
<10-z <1o-3 <10-s <10-s <10-s <10-s <10-s <10-s <10-s <10-s
60 56.5 51 63.5 61 47 62 64 52 58
NS NS NS NS co.05 NS NS NS NS NS
58 55 54 57 64 62 57 62 61 55
NS NS NS NS <10-s <10-a co.05 <0.05 <0.05 NS
869
C. Pilette
et al.
TABLE 2 Diagnostic accuracies and predictive values (%) of esophageal varices (EV) according to multivariate analysis in the different patient groups Chronic liver disease
Cirrhosis
EV
Large EV
EV
Large EV
Global analysis
DA: 81 (73-87) PPV: 53 (38-68) NPV: 96 (89-99) (n=ll)
DA: 80 (72.5-86) PPV: 72 (59-82) NPV: 87 (77-93) (n=ll)
DA: 71 (61-79) PPV: 84 (73-91) NPV: 48 (3264) (n=7)
DA: 71 (62-79) PPV: 65 (51-78) NPV 76 (63-86) (n=S)
Stepwise analysis
1st, liver surface: DA: 81 (73-87) PPV: 56 (36-75) NPV: 88 (79-93.5)
lst, prothrombin: DA: 76 (68-82.5) 2nd: y globulins: DA: 79 (72-85) PPV: 73 (60-83), NPV: 84 (74-91)
lst, platelet: DA: 67 (58-76) 2nd, prothrombin: DA: 72 (63~80), PPV: 85 (74.5-92) NPV: 50 (34-55)
lst, platelet count: DA: 67.5 (58-76) 2nd, prothrombin: DA: 69, 3rd: spider naevi: DA: 70 (61-78), PPV 64 (5&77), NPV: 75 (62285.5)
n indicates the number of variables, with diagnostic accuracy ~70% in univariate analysis (65% in the group of cirrhosis), included in the model. DA: diagnostic accuracy, PPV positive predictive value, NPV: negative predictive value. Figures into brackets indicate 95% confidence interval. Main data are presented in bold.
cant variables related to cirrhosis in fact predicted cirrhosis or fibrosis. For the presence of EV, the fibrosis score was isolated at the first step, whereas the fibrosis area was isolated at the first step for large EV (data not shown). This indicates that markers of EV in patients with CLD are markers of liver fibrosis. Therefore, we based our analysis mainly on the group of patients with cirrhosis (see Discussion). In the group of patients with cirrhosis, global DA was 71%, and platelet count was isolated at the first step and prothrombin index at the 2nd step either for all EV or large EV Thus, platelet count appeared to be the best single marker of EV or large EV, since other markers added little information (Table 2). The ROC curve of the platelet count for large EV is presented in Fig. 1. The best diagnostic threshold was set at 160 G/ 1 providing a sensitivity of 80% and a specificity of 58%. Fig. 2 is another ROC curve in which DA, negative predictive value (NPV) and positive predictive value (PPV) are expressed as a function of platelet count. The DA curve confirms that the highest DA was observed at 160 G/l. The PPV curve showed a clinically irrelevant shape. However, the NPV curve showed that
the higher the platelet count, the lower the probability of large EV, e.g. NPV was 91% for 260 G/l. In an explanatory phase using quantitative variables with a high DA from Table 1 as well as histological variables from Table 3 as explanatory variables and the grade of EV as an explained variable in multiple regression, global R* was 0.58 with 4 independent variables by stepwise analysis: 1st step: spider naevi (R*= 0.21) 2nd step: platelet count (R* =0.37), 3rd step: fibrosis area (R*=O.47), 4th step: liver length (US at right kidney, R*=0.51).
Discussion The prevalence of cirrhosis in developed countries ranges between 0.4% in the USA or France (19,20) and 1.1% in Italy (21). PHT and liver dysfunction are the most frequent complications of cirrhosis. The main
loo-
80 -
60 -
TABLE 3 Diagnostic accuracies (%) of histological data for esophageal varices (EV) in patients with chronic liver disease according to univariate analysis. Diagnostic accuracies ~70% are depicted in bold characters
CLV centrolobular vein.
870
.
W)
40-
Large EV
All EV
Fibrosis score Cirrhosis CLV fibrosis Steatosis Alcoholic hepatitis Fibrosis area
se
%
P
%
P
80 go 83 61 64 80
<10-s <10-s <10-s
69 69 72 57 60 80
<10-s <10-s <10-s <1o-3 co.05 <10-s
“,
.
0
I
20
.
.
40 1 - spe
I
60 (%)
,
80
100 -
Fig. I. ROC curve of platelet count (Gil) for the diagnosis of large esophageal varices. Se: sensitivity, Spe: specSficity.
Esophageal varice diagnosis Indexer
(%) lOO-
Cm-
60 -
v-u
.=
/v-d
I
I’ :
70 -
60 -
60I
40 : 0
100
200
300
Platelet count (G/l)
Fig. 2. Curves of diagnostic accuracy (DA), positive predictive value (PPV) and negative predictive value (NPV) as a function of platelet count (Gil).
consequence of PHT is the development of collateral circulation, including EV which are the most frequent and most dangerous complication. EV and variceal bleeding are an early complication of cirrhosis justifying early screening for EV (4). The usual means of di-
agnosing EV is upper gastro-intestinal endoscopy. However, endoscopy can be considered invasive due to the technique and level of discomfort. Non-invasive methods of diagnosis of considerable interest have been developed in the past few years. In CLD, the main objective is the non-invasive diagnosis of fibrosis or cirrhosis (7,22), which can be obtained by a combination of clinical signs, biochemical data such as serum markers of fibrosis (7), radiology such as Doppler US (14) and finally by upper gastro-intestinal endoscopy since EV and PHT gastropathy have a high (88%) DA for the diagnosis of cirrhosis (5). Is the non-endoscopic diagnosis of EV in CLD relevant? Our results indicate that markers of EV in patients with CLD are markers of liver fibrosis. Moreover, the role of EV as a marker of cirrhosis in CLD was confirmed with an 81% DA in the present CLD population (7). Thus, the non-endoscopic diagnosis of EV does not seem to be relevant in patients with CLD since: a) non-endoscopic markers of EV are the noninvasive markers of cirrhosis in that population, and b) the endoscopic diagnosis of EV is relevant for the non-invasive diagnosis of cirrhosis which is a more important clinical issue. On the other hand, although there is a strong relationship between EV and cirrhosis, EV mainly need to be diagnosed in patients with cirrhosis. First, EV are present in 2/3 to 3/4 of patients
TABLE 4 Diagnostic signs of esophageal varices (EV) in the literature 1st author (ref)
Year
Liver disease
Variables ?I
Variable type
Univariate analysis
Multivariate analysis
EV (“XI)
Lavergne (27)
1997
Cirrhosis?
52
12
Ascites
_
79
CLD
32
_
28
Cirrhosis HCV
50
5
_
ss
Chalasani (30)
1998
Cirrhosis pre-liver transplantation
346
_
Clinics Blood tests
Cholinesterase Spleen length Hepatic vein waveform, portal vein diameter, congestion index _
_
1997
Clinics Blood tests US Blood tests US Doppler US Doppler
Zeijen (28)
1994
Gorka (29)
Garcia-Tsao (31)
1997
Compensated cirrhosis
180
15
Clinics Blood tests
Khishimoto (32)
1998
47
3
US Doppler
Present study
1999
Cirrhosis, idiopathic PHT 1. CLD
Age, sex, Child, spider naevi, bili, alb, PT, Hb, WBC, platelet Esophageal wall
65
Clinics Blood tests US Doppler
1. n=49 See text 2. n=7
2. Cirrhosis
Patients n
1. 207 2. 116
Splenomegaly, platelet count NASH Spider naevi, alb, platelet
_
16
64 1. Liver surface
1. 45
2. Platelet count, prothrombin index, spider naevi
2. 72
CLD: chronic liver disease, US: ultrasonography, -: not available, PHT: portal hypertension.
871
C. Pilette et al.
with cirrhosis (2) and, second, only large EV require primary prevention. Therefore, the non-invasive diagnosis of EV, especially large-sized EV, is clinically relevant. In the literature, the non-invasive diagnosis of EV has already been studied. A clear distinction must be made between studies evaluating the development of EV and those evaluating the diagnosis of EV The former studies the risk factors for the occurrence or the evolution of EV in longitudinal studies (4,6,23,24); the latter studies the markers of the presence of EV in a horizontal study. In this last category, one can distinguish direct markers of EV such as radiology (25), e.g. barium swallow (26) and indirect markers that are listed in Table 4 (27-32). Most of these papers were preliminary reports as abstracts. In the present study, we found an acceptable DA for the diagnosis of EV in patients with CLD. To be of clinical interest, the DA must be >75 % since this is the limit between (50%) and the perfect DA (100%). However, this was obtained with Doppler US (liver surface) which has some limitations, mainly anatomical, preventing examination. The lower DA for EV than that observed for cirrhosis (7) can be explained by the presence of EV in some patients without cirrhosis and the absence of EV in some patients with cirrhosis. The former is attributable to false-negative liver biopsies or to observer variability. Indeed, all EV were grade 1 in non-cirrhotic patients and this grade is responsible for the highest observer variability (12). For these reasons, as well as the clinical importance of the diagnosis of cirrhosis mentioned above, the noninvasive diagnosis of EV is more relevant in patients with cirrhosis. Moreover, at present, only the screening of large EV is clinically relevant. For this condition, three independent markers were disclosed by multivariate analysis, in decreasing order of importance: platelet count, prothrombin index and spider naevi. Platelet count has also been shown to be an independent marker in two other studies with multivariate analysis (30,31). It can be considered a marker of PHT in relation to the hypersplenism secondary to PHT The ROC curve showed that a platelet count fixed at 160 G/l was the best cut-off for the DA of cirrhosis. The ROC curve also showed that the presence of large EV is improbable if cirrhotic patients have a platelet count ~260 G/l (negative predictive value ~91%). The prothrombin index expresses prothrombin time as a percentage, and is more reproducible than time (33). In our experience, the prothrombin index is an excellent indirect marker of liver fibrosis or cirrhosis (7,34,35). Spider naevi were also shown to be an independent diagnostic factor for large EV in the present study. This 872
factor has already been shown to be related to EV, large EV (36,37) or variceal bleeding (37). The present results differ from the recommendations of a recent expert conference (38) which stated that EV should be screened: a) in all cirrhotic patients who are Child’s class B or C, and b) in Child’s A patients where there is clinical evidence of PHT, e.g. a low platelet count (cl40 G/l), an enlarged portal vein diameter (~13 mm), or evidence of collateral circulation at ultrasound. However, this statement is derived from a preliminary report where the methodology was not stated (39). Moreover, the sensitivity of a platelet count 1140 G/l was only 41% in Child’s A patients (39). In rats with PHT, multivariate analysis has shown that the area of liver fibrosis is the only predictive factor of the degree of porto-systemic shunts (40). Table 2 shows that the highest DA for the diagnosis of EV were related to histological variables of fibrosis. However, other serum markers of liver fibrosis than prothrombin index had a fairly low DA for EV in patients with cirrhosis. It should be noted that these markers also have a low diagnostic value for the level of portal pressure (8). Finally, the DA for EV or large EV in patients with cirrhosis was 72 and 70%, respectively. These figures are not clinically relevant. Other variables should be evaluated to improve this DA. Recently, a sensitivity of 92% and a specificity of 100% were observed for the detection of large EV in patients with cirrhosis due to hepatitis C virus by measuring US waveform in the hepatic veins (29). These promising data must be confirmed. In conclusion, by using a few non-endoscopic criteria, EV can be correctly diagnosed in 81% of patients with CLD and in 71% of patients with cirrhosis. These results show that the non-invasive screening of patients who are candidates for the primary prevention of variceal bleeding is possible, but should be improved before a clinical application can be recommended.
Acknowledgements We thank Pr. Daniel Chappard for his contribution to image analysis and MS Dale Roche for her contribution.
References 1. Schlichting P Christensen E, Fauerholdt L, Poulsen H, Juhl E, Tygstrup N. Main causes of death in cirrhosis. Stand J Gastroenter01 1983; 18: 881-8. 2. Calts P Pascal JP Histoire naturelle des varices cesophagiennes au tours de la cirrhose (de la naissance a la rupture). Gastroenter01 Clin Biol 1988; 12: 245-54. 3. Poynard T, Cal&s P Pasta L, Ideo G, Pascal JP, Pagliaro L, et al. and France-Italian Multicenter Study Group. Beta-adrenergic-
Esophageal varice diagnosis antagonist drugs in the prevention of gastrointestinal bleeding in patients with cirrhosis and esophageal varices. N Engl J Med 1991; 324: 1532-8. 4. Cal& P Desmorat H, Vine1 JP Caucanas JR Ravaud A, Germ P et al. Incidence of large esophageal varices in patients with cirrhosis. Application to prophylaxis of first bleeding. Gut 1990; 31: 1298-302. 5. Oberti F, Burtin P MaIga MY, Valsesia E, Pilette C, Cal&s I? Gastroesophageal endoscopic signs of cirrhosis: independent diagnostic accuracy, interassociation and relationship to etiology and hepatic dysfunction. Gastrointest Endosc 1998; 48: 1485?. 6. Angulo P, Malinchoc M, Therneau T, Jorgensen R, Dickson ER, Lindor KD. Predicting esophageal varices in patients with primary biliary cirrhosis (PBC) [abstract]. Gastroenterology 1997; 112: A1213. 7. Oberti F, Valsesia E, Pilette C, Rousselet MC, Bedossa P, Aube C, et al. Non Invasive diagnosis of hepatic fibrosis or cirrhosis. Gastroenterology 1997; 113: 1609-16. 8. Bahr MJ, Baker KHW, Horn W, Gtinzler V Manns MP Serum laminin Pl levels do not reflect critically elevated portal pressure in patients with liver cirrhosis. Hepata-Gastroenterology 1997; 44: 1200-5. 9. Papazian A, Braillon A, Dupas JL, Sevenet E Capron JI? Portal hypertensive gastric mucosa: an endoscopic study. Gut 1986; 27: 1199-203. 10. McCormack TT, Sims J, Eyre-Brook J, Kennedy H, Goepel J, Johnson AG, et al. Gastric lesions in portal hypertension; inflammatory gastritis or congestive gastropathy. Gut 1985; 26: 1226-32. 11. Sacchetti C, Capelo M, Rebecchi P, Roncucci L, Zanghieri G, Tripodi, et al. Frequency of upper gastroihtestinal lesions in patients with liver cirrhosis. Dig Dis Sci 1988; 33: 1218-22. 12. Cal&s P, Zabotto B, Meskens C, Caucanas JP Vine1 JP, Desmorat H, et al. Gastroesophageal endoscopic features in cirrhosis: observer variability, interassociations and relationship to hepatic dysfunction. Gastroenterology 1990; 98: 15662. 13. Tissot 0, AubC C, Namour A, Blanc F, Dauver A, Cal&s I? Semiologie echographique de la cirrhose: concordance inter-observatcur. Gastroenterol Clin Biol 1995; 19: 291-6. 14. AubC C, Oberti F, Korali N, Namour MA, Loisel D, Tanguy JY et al. Echographic diagnosis of hepatic fibrosis or cirrhosis. J Hepatol 1999; Xl: 472-8. 15. Seitz JF, Boustitre C, Maurin P, Aimino R, Durbec JP, Botta D, et al. Evaluation de l’ultrasonographie dans le diagnostic des cirrhoses. Etude retrospective de 100 examens consecutifs. Gastroenterol Clin Biol 1983; 7: 734-9. 16. Subramanyam BR, Balthazar EJ, Madamba MR, Raghavendra BN, Horii SC, Lefleur RS. Sonography of portosystemic venous collaterals in portal hypertension. Radiology 1983; 146: 1616. 17. Cioni G, D’Alimonte P, Cristani A, Ventura P, Abbati G, Tincani E, et al. Duplex-doppler assessment of cirrhosis in patients with chronic compensated liver disease. J Gastroenterol Hepatol 1992; 7: 3824. 18. Bedossa P, Bioulac-Sage P Callard P, Chevallier M, Degott C, Deugnier Y, et al. Intraobserver and interobserver variations in liver biopsy interpretation in patients with chronic hepatitis C. Hepatology 1994; 20: 15-20. 19. Quinn PG, Johnston DE. Detection of chronic liver disease: costs and benefits. Gastroenterologist 1997; 5: 58-77. 20. Cal&s I! Epidemiologic et pronostic de la cirrhose. Contours Medical 1995; 117: 2707-l 1. 21. Bellentani S, Tiribelli C, Saccoccio G, Sodde M, Fratti N, De Martin C, et al. Prevalence of chronic liver disease in the general population of Northern Italy: the Dionysos study. Hepatology 1994; 20: 1442-9. 22. Guyader D, Jacquelinet C, Moirand R, Turlin B, Mendler MH,
23.
24.
25. 26.
27.
28.
29.
30.
31.
32.
33.
34.
35.
36. 37.
38.
39.
40.
Chaperon J, et al. Noninvasive prediction of fibrosis in C282Y homozyous hemochromatosis. Gastroenterology 1998; 115: 92936. Zoli M, Merckel C, Magalotti D, Gueli D, Grimaldi M, Gatta A, et al. Natural history of cirrhotic patients with small esophageal varices: a prospective study [abstract]. Hepatology 1998; 28: 409A. Gores GJ, Wiesner RH, Dickson ER, Zinsmeister AR, Jorgensen RA, Langworthy A. Prospective evaluation of esophageal varices in primary biliary cirrhosis: development, natural history, and influence on survival. Gastroenterology 1989; 96: 1552-9. Brown JJ, Naylor MJ, Yagan N. Imaging of hepatic cirrhosis. Radiology 1997; 202: 1-16. Ginai AZ, Van Buuren HR, Hop WCP Schalm SW. Oesophageal varices: how reliable is a barium swallow ? Br J Radio1 1993; 66: 322-6. Lavergne J, Molina E, Reddy KR, Jeffers L, Leon R, Nader AK, et al. Ascites predicts the presence of high grade varices by screening gastroscopy [abstract]. Gastrointest Endosc 1997; 45: AB187. Zeijen RNM, Caenepeel E Stockbrttgger RW, Arends JW, Oei TK. Prediction of esophageal varices in liver disease; preliminary results [abstract]. Gastroenterology 1994; 106: A1013. Gorka W, Al Mulla A, Al Sebayel M, Altraif I, Gorka TS. Qualitative hepatic venous Doppler sonography versus portal flowmetry in predicting the severity of esophageal varices in hepatitis C cirrhosis. Am J Roentgen01 1997; 169: 511-5. Chalasani N, Imperiale T, Ismail A, Sood G, Wilcox CM, Kwo P et al. A predictive index for determining the risk for presence of large esophageal variccs in patients with cirrhosis [abstract]. Gastroenterology 1998; 114: A1222. Garcia-Tsao G, Escorsell A, Zakko M, Patch D, Matloff D, Grace N. Predicting the presence of significant portal hypertension and varices in compensated cirrhotic patients [abstract]. Hepatology 1997; 26: 360A. Khishimoto R, Chen MH, Ogawa H, Wakabayashi MN, Kogutt MS. Esophageal varices: evaluation with transabdominal US. Radiology 1998; 206: 647-50. Robert A, Chazouilleres 0. Prothrombin time in liver failure: time, ratio, activity percentage, or international normalized ratio? Hepatology 1996; 24: 1392-4. Pilette C, Rousselet MC, Bedossa E Chappard D, Oberti F, Rifllet H, et al. Histopathological evaluation of liver fibrosis: quantitative image analysis vs semi-quantitative scores. Comparison with serum markers. J Hepatol 1998; 28: 439-46. Pilette C, Croquet V, Oberti F, AubC C, Rousselet MC, Bedossa R et al. Prothrombin index is an accurate indirect marker of liver fibrosis or cirrhosis [abstract]. Hepatology 1998; 28(4Pt2): 385A. Weissberg JI, Stein DET, Fogel M, et al. Physical examination predicts the presence of varices. Surv Dig Dis 1984; 2: 141-55. Foutch PG, Sullivan JA, Gaines JA, Sanowski RA. Cutaneous vascular spiders in cirrhotic patients: correlation with hemorrhage from esophageal varices. Am J Gastroenterol 1988; 83: 7236. Grace ND, Groszmann RJ, Garcia-Tsao G, Burroughs AK, Pagliaro L, Makuch RW, et al. Portal hypertension and variceal bleeding: an AASLD single topic symposium. Hepatology 1998; 28: 868-80. Pagliaro L, D’Amico G, Pasta L, Politi E Vizzini G, Traina M, et al. Portal hypertension in cirrhosis: natural history. In: Bosch J, Groszmann RJ, editors. Portal Hypertension: Pathophysiology and Treatment. Oxford: Blackwell Scientific Publications; 1994. p. 72-92. Oberti F, Pilette C, Rifflet H, Maiga M, Moreau A, Gallois Y, et al. Effects of simvastatin, pentoxyfylline and spironolactone on hepatic fibrosis and portal hypertension in cirrhotic rats. J Hepato1 1997; 26: 1363-71.
873