fax: (33) 1-48-02-62-02
Liver Failure and Liver Disease
Accuracy of liver stiffness measurement for the diagnosis of cirrhosis in patients with chronic liver diseases†
Article first published online: 28 NOV 2006
Copyright © 2006 American Association for the Study of Liver Diseases
Volume 44, Issue 6, pages 1511–1517, December 2006
How to Cite
Ganne-Carrié, N., Ziol, M., de Ledinghen, V., Douvin, C., Marcellin, P., Castera, L., Dhumeaux, D., Trinchet, J.-C. and Beaugrand, M. (2006), Accuracy of liver stiffness measurement for the diagnosis of cirrhosis in patients with chronic liver diseases. Hepatology, 44: 1511–1517. doi: 10.1002/hep.21420
Potential conflict of interest: Dr. Beaugrand received travel grants from Echosens. Dr. Ziol received grants from Echosens.
- Issue published online: 28 NOV 2006
- Article first published online: 28 NOV 2006
- Manuscript Accepted: 20 AUG 2006
- Manuscript Received: 16 DEC 2005
A proper diagnosis of cirrhosis is essential for the management of patients with chronic liver diseases. We assessed the accuracy of liver stiffness measurement by Fibroscan for the diagnosis of cirrhosis in 1,257 patients with chronic liver diseases of various causes enrolled in a prospective multicenter study as well as clarified causes of discrepancies between liver histology and Fibroscan. One hundred thirty-two patients had unsuitable biopsy specimens, and 118 had unreliable liver stiffness measurements. Because 232 patients overlapped with a previous study, analysis was performed in the 775 new patients then derived in the whole population (1,007; 165 cirrhosis). Diagnostic accuracy was assessed by receiver operator curve (ROC) analysis. Liver samples were re-analyzed in case of discrepancies. The area under the ROC (AUROC) was 0.95 (95% CI, 0.93-0.96) for the diagnosis of cirrhosis in either 775 or 1,007 patients. The cutoff value with optimal diagnosis accuracy was 14.6 kPa in 1,007 patients (positive and negative predictive values, 74% and 96%) with discrepancies among the etiological groups. Eighty patients were misclassified: (1) among 45 patients without cirrhosis with liver stiffness 14.6 kPa or greater, 27 (60%) had extensive fibrosis and 10 (22%) significant perisinusoidal fibrosis; and (2) among 35 patients with cirrhosis and liver stiffness less than 14.6 kPa, 10 (29%) had a macronodular pattern and 25 (71%) either none or mild activity. In conclusion, Fibroscan is a reliable method for the diagnosis of cirrhosis in patients with chronic liver diseases, better at excluding than at predicting cirrhosis using a threshold of 14.6 kPa. False-negatives are mainly attributable to inactive or macronodular cirrhosis. (HEPATOLOGY 2006;44:1511–1517.)
The prognosis and clinical management of chronic liver diseases (CLD) strongly depend on the extent of liver fibrosis, because life-threatening complications mainly occur in patients with cirrhosis. In patients with compensated cirrhosis, the annual incidence of hepatocellular carcinoma, decompensation, and death reach approximately 3%, 4%, and 3%, respectively.1–5 This emphasized the need for early identification of cirrhosis to screen for or prevent complications.
Liver biopsy (LB) is the gold standard for the assessment of cirrhosis. Although LB is generally safe, it is a costly procedure that carries a small risk of severe complications and is difficult to accept for patients.6, 7 In addition, sampling error is common because only 1/50,000 of the organ is analyzed, resulting in up to 30% of false-negative results.8 These limitations may lead to an underestimation of cirrhosis, especially when LB specimens are small or fragmented. Therefore, the assessment of liver fibrosis by noninvasive means is a major challenge that has stimulated the search for new approaches. Ultrasonography had a reported accuracy of 82%-88%9 to 100% in assessing a diagnosis of cirrhosis.10 Unfortunately, its value is tempered by significant interobserver variability and an inability to gather all the required measurements, because of technical limitations. Several serum markers, and their combination in scores, have been reported as predicting the presence of cirrhosis with a high degree of accuracy.11–25 However, most of them have been evaluated only in patients infected with hepatitis C virus (HCV), and few have been validated. In addition, these markers could be influenced by extrahepatic conditions, and their overall diagnostic value for cirrhosis is less than optimal. Transient elastography is a new, noninvasive, and reproducible technique that measures tissue stiffness.26 Liver stiffness measurement (LSM) has been demonstrated to be a reliable tool for assessing hepatic fibrosis and cirrhosis mainly in patients with chronic hepatitis C.27–30 Additional studies evaluated the accuracy of LSM for the diagnosis of liver fibrosis in patients with CLD of various causes31 and in patients with cholestatic liver diseases.32
The purpose of this prospective multicenter study was to assess the accuracy of LSM for the diagnosis of cirrhosis in a large cohort of patients with CLD of various causes, to determine the cutoff values of LSM for optimal diagnosis in the different subgroups, and to describe histological liver lesions found in misclassified patients.
Patients and Methods
Between November 2002 and September 2004, 1,257 patients were consecutively enrolled in a prospective study if they had compensated CLD and underwent LB and LSM within less than 6 months, in the hepatogastroenterology departments of four French university hospitals (1: Jean Verdier, n = 648; 2: Haut-Lévêque, n = 282; 3: Henri Mondor, n = 188; 4: Beaujon, n = 139). The study protocol was in accordance with the ethical guidelines of the 1975 Declaration of Helsinki and was approved by an independent ethics committee (CCPPRB, Aulnay-sous-bois, France). Patients were enrolled after providing their written and informed consent.
Besides the epidemiological data, the following biological parameters were determined for all patients at the time of LSM: serum aminotransferases, gamma-glutamyl-transpeptidase, bilirubin, platelet count, prothrombin time, and albumin. The cause of CLD was determined using standard diagnostic criteria. HCV or hepatitis B virus (HBV) infection was diagnosed by serological detection of HCV antibodies and HBV surface antigen, respectively. Alcoholic liver disease was diagnosed in patients with consumption of at least 40 g alcohol daily for 5 years or more. Other diseases were diagnosed according to the current criteria.
Liver Stiffness Measurement.
LSM was performed using Fibroscan (Echosens, Paris, France), a new medical device based on elastometry. Details of the technical description and examination procedure have been reported previously.26, 33 Briefly, an ultrasound transducer probe is mounted on the axis of a vibrator; vibrations of mild amplitude and low frequency are transmitted from the vibrator toward the tissues by the transducer itself, inducing an elastic shear wave that propagates through the tissue. In the meantime, pulse-echo ultrasound acquisitions are performed to follow the propagation of the shear wave and to measure its velocity, which is directly related to tissue stiffness. The results are expressed in kilopascal (kPa). The harder the tissue, the faster the shear wave propagates. Measurements were performed in the right lobe of the liver through intercostal spaces on patients lying in dorsal decubitus with the right arm in maximal abduction. The tip of the probe transducer was covered with coupling gel and placed on the skin, between the rib bones at the level of the right lobe of the liver. The operator, assisted by ultrasound time-motion images, located a liver portion at least 6 cm thick and free of large vascular structures. Once the measurement area had been located, the operator pressed the probe button to begin an acquisition. The measurement depth was between 25 and 65 mm below the skin surface. Acquisitions that did not have a correct vibration shape or a correct follow-up of the vibration propagation were automatically rejected by the software. The success rate was calculated as the ratio of the number of successful measurements over the total number of acquisitions. The median value of the successful acquisitions was kept as representative of the liver stiffness and was called LSM. Only LSM obtained with at least eight successful acquisitions and a success rate of at least 50% were considered reliable. Operators were masked to the results of LB but could have access to the medical file and biological data.
Liver Histology and Quantification of Liver Fibrosis.
LB specimens were fixed in formalin and paraffin embedded. Four-micrometer-thick sections were stained with hematoxylin-eosin-safran and picrosirius red. All LB specimens were analyzed by an experienced liver pathologist blinded to the results of LSM and clinical data. Five hundred ninety-one (47%) of them were re-analyzed by two independent pathologists for the purpose of previous studies. LB specimens were measured in millimeters. Excluded from the histological analysis were non re-analyzed LB specimens with a length of less than 10 mm, and re-analyzed LB specimens considered by the pathologists as non-suitable for fibrosis assessment. Liver fibrosis was staged on a 0-4 scale according to the METAVIR scoring system (F0, no fibrosis; F1, portal fibrosis without septa; F2, portal fibrosis and few septa; F3, numerous septa without cirrhosis; and F4, cirrhosis) in patients with chronic hepatitis whatever the cause,34 and according to Brunt et al. (stage 1, zone 3 perisinusoidal fibrosis; stage 2, as in stage 1 with portal fibrosis; stage 3, as in stages 1 and 2 with bridging fibrosis; and stage 4, cirrhosis) in patients with alcoholic liver disease and nonalcoholic steatohepatitis.35
In cases of discrepancies between LSM and the liver fibrosis stage, LB specimens were re-analyzed by an experienced pathologist (M.Z.) to clarify the reasons for misclassification.
The diagnostic performance of LSM was assessed by using receiver operating characteristics (ROC) curves. Connected with any cutoff value is the probability of a true positive (sensitivity, Se) and the probability of a true negative (specificity, Sp). The ROC curve is a plot of Se versus 1- Sp for all possible cutoff values. The most commonly used accuracy index is the area under the ROC curve, when values close to 1.0 indicate high diagnostic accuracy. ROC curves were thus built for the detection of cirrhosis. Optimal cutoff values for LSM were chosen either to obtain a 95% Se, to maximize the sum of Se and Sp, to optimize the diagnostic performance (sum of true positives and true negatives over the total number of patients), or to obtain a 95% Sp according to the diagnostic question. Comparison of patient distribution according to cause between patients properly or improperly diagnosed by LSM was performed using chi-squared analyses. Other parameters of comparison (such as age, success rate, liver sample length, etc.) between patients properly or improperly diagnosed by LSM were performed using Kolmogorov-Smirnov tests. Statistical analyses were performed with NCSS 2004 software (Statistical Systems, Kayville, UT).
A posteriori exclusion criteria were the following: (1) LB unsuitable for fibrosis staging (10.5%) either non re-analyzed specimens with a length less than 10 mm (n = 68) or non-interpretable re-analyzed specimens (n = 64), and (2) unreliable LSM with fewer than eight successful acquisitions (n = 60), or a success rate of less than 50% (n = 58). The proportion of excluded patients per center was 1: 24.2%; 2: 25.2%; 3: 15.6%; 4: 7.5%. The significant lower proportion of excluded patients in center 4 was not related to the rate of LSM failure, which was not different among centres. Conversely, it could be related to the rate of suitable LB. Indeed, a significant greater mean length of LB was observed in center 4, probably due at least in part to a significant lower proportion of patients with cirrhosis (7% vs. 18% in other centers). As 232 patients overlapped with a previous study,30 statistical analysis was first performed in the 775 new patients including 120 with cirrhosis (Fig. 1). Table 1 summarizes the main characteristics of the 775 patients included. There were 487 males, mean age 48 ± 13 years. Causes of CLD were HCV (n = 298) or HBV (n = 122), alcohol (n = 75), non-alcoholic steatohepatitis (n = 47), hemochromatosis (n = 12), cholestatic liver disease (n = 16), mixed (n = 49, including HCV and alcohol in 41), and miscellaneous (n = 164). Most (80.4%) of the patients had LB and LSM within the same day (mean delay, 4.9 days). The median LB length was 17 mm (range, 5-40).
|Cirrhosis N = 120||No Cirrhosis N = 655|
|Male gender||87 (73%)||400 (61%)|
|Age (mean ± SD, years)||53 ± 13||47 ± 13|
|Causes of liver disease|
|Hepatitis C virus||30 (26%)||268 (41%)|
|Hepatitis B virus||16 (13%)||106 (16%)|
|Alcohol||34 (28%)||41 (6%)|
|Hepatitis C virus + alcohol||8 (7%)||33 (5%)|
|Nonalcoholic steatohepatitis||3 (2%)||44 (8%)|
|Hemochromatosis||5 (4%)||7 (1%)|
|Cholestatic liver disease||0||16 (2%)|
|Other||24 (21%)||140 (22%)|
|Platelet count (mean ± SD, × 10 3/mm3)||160 ± 75||229 ± 79|
|Prothrombin time (mean ± SD, %)||74 ± 18||93 ± 11|
|Albumin (mean ± SD, g/L)||39.5 ± 7.7||43.5 ± 5.8|
|Bilirubin (mean ± SD, μmol/L)||45 ± 82||13 ± 21|
|AST (mean ± SD, IU/L)||90 ± 84||65 ± 113|
|ALT (mean ± SD, IU/L)||98 ± 75||114 ± 209|
|Liver biopsy (mean, mm)||14||19|
|Liver stiffness measurement|
|Median (kPa, ranges)||27.4 (5.8–75)||6.3 (1.3–75)|
|Success rate (% ± SD)||88 ± 15||89 ± 15|
|Delay between LB and LSM (mean, days)||9.4||4.1|
Relationship Between LSM and Cirrhosis.
LSM values ranged from 1.3 to 75 kPa (median, 6.9 kPa; mean interquartile range, 2.57 kPa). The mean number of measurements per patient was 11.7 ± 2.4 (range, 9-20), and the mean success rate was 89% ± 14 % (range, 50%-100%). Median LSM were 27.4 kPa (range, 5.8-75) in patients with cirrhosis and 6.3 (range, 1.3-75) in those without. Figure 2 shows the diagnostic value (ROC curves) of LSM for the diagnosis of cirrhosis in 775 patients, in 298 with HCV, in 122 with HBV, and in 122 with either alcoholic or nonalcoholic steatohepatitis. The corresponding area under the ROC (95% CI) were 0.95 (0.93-0.97) in the whole population, 0.96 (0.91-0.98) in HCV, 0.90 (0.77-0.96) in HBV, and 0.96 (0.90-0.98) in patients with either alcoholic or nonalcoholic steatohepatitis. We then checked that the results did not differ, taking into account all patients, including the 232 enrolled in the previous study.30 Results were comparable (area under the ROC 0.95 [CI 95%: 0.93-0.96] in 1,007 patients and 0.96 [CI 95%: 0.93-0.97] in 493 HCV patients).
Based on the LSM distribution according to cirrhosis and the ROC curves, four optimal cutoff values were defined: cutoff values with 95% Se, cutoff values that maximize the sum of Se and Sp, cutoff values with 95% Sp, and cutoff values with optimal diagnosis accuracy. These optimal cutoff values were, respectively, 9.4, 11.7, 17.1, and 17.1 kPa in 775 patients, and 9.2 kPa, 11.7 kPa, 14.6 kPa, and 14.6 kPa in 1,007 patients (Table 2). For patients with chronic hepatitis B or C, optimal cutoff values were very close (Table 3). For patients with either alcoholic or nonalcoholic steatohepatitis, optimal cut-off values were higher (Table 3).
|Cutoff||Sensitivity||Specificity||Positive Predictive Value||Negative Predictive Value||Likelihood Ratio||Diagnostic Accuracy|
|Cutoff||Sensitivity||Specificity||Positive Predictive Value||Negative Predictive Value||Likelihood Ratio||Diagnostic Accuracy|
|Optimal Cutoff (kPa)|
|Hepatitis C N = 298||Hepatitis B N = 122||Alcohol or NASH N = 122|
|Maximum sum of sensitivity and specificity||10.4||10.3||21.5|
|Best diagnostic accuracy||20.2||16.9||21.5|
Because the cutoff value of LSM reported in our previous study30 was very close to the cutoff value defined for HCV patients in this study, we decided to mix the 232 HCV patients enrolled in the previous study and the 775 new patients to provide values in a population more representative of common practice in Western countries in which HCV represents approximately 50% of the indications of liver biopsy.7 Obviously, the number of misclassified patients would have been lower if specific cutoff values for each etiological group would have been used because 164 patients belonged to the miscellaneous etiological group where a specific cutoff was questionable. Thus, we considered the cutoff with optimal diagnostic accuracy in the whole population (14.6 kPa, positive and negative predictive values 74% and 96%, respectively) for the second part of the study.
Histological Study of Misclassified Patients in 1,007 Patients.
Discrepancies between LSM and the histological assessment of cirrhosis were observed in 7.9% of cases (percentage of cases by center: 1: 8.8%; 2: 10.6%; 3: 7.6%; 4: 4.6%). Compared with others, misclassified patients were more significantly male (77%, P = .004) with a higher proportion of alcoholic CLD or nonalcoholic steatohepatitis (20 of 122, 16.4% vs. 5.9% for patients infected with HCV and 8.2% for patients infected with HBV). Their mean age (47 ± 12 years), the mean interval between LSM and LB (2.7 days), the mean number of measurements per patient (11.6), the mean success rate (89%), and the mean length of LB (17.8 mm) were not significantly different from the others.
LB specimens of 80 misclassified patients were reanalyzed to clarify the reasons for the misclassification. Among patients without cirrhosis with LSM of 14.6 kPa or greater (n = 45), the main liver histological lesions observed were extensive fibrosis with numerous septa (F3) in 27 cases (60%), mild to marked perisinusoidal fibrosis in 10 cases (22%), massive necrosis with collapse in three cases (including a case with marked perisinusoidal fibrosis), and trabecular architectural modifications in five cases either without significant fibrosis or associated with mild to marked perisinusoidal fibrosis. In the four remaining cases, cirrhosis was formally ruled out and no additional lesions explaining increased LSM were found.
Among the 35 patients with cirrhosis with LSM less than 14.6 kPa, the diagnosis of cirrhosis was confirmed by the pathologist in all cases. Twenty-five (71%) displayed absent or mild activity grade. Eleven of them (31%) had current or past etiological treatment of CLD, including five with HCV chronic hepatitis and sustained virological response more than 18 months after the end of treatment. Ten had macronodular cirrhosis (28.6%).
Beside potential bias including the lack of maskedness to clinical and laboratory data, and the larger percentage of patient exclusion in the centers with the greater patient enrollment, this prospective study shows that Fibroscan is an efficient technique for the diagnosis of cirrhosis whatever the cause of CLD. In the whole population (n = 1,007), with a cutoff value of 14.6 kPa, chosen to obtain a 95% Sp, positive and negative predictive values for the diagnosis of cirrhosis were 74% and 96%, respectively, and the performance accuracy rate was optimal (92%). In clinical practice, such results could be of major relevance, mainly for excluding cirrhosis in patients with CLD. In addition to previous studies mostly performed in HCV patients,27–30 our results suggest that cutoff values could be optimized if specifically defined for each cause, mainly chronic hepatitis versus alcoholic CLD or nonalcoholic fatty liver or steatohepatitis presumably attributable to the discrepancies in fibrosis amounts between these various conditions.
The second goal of this study was to review histological data in patients with conflicting results between LSM and LB. The reasons for the discrepancies were presumably not technical in most cases because, although not assessed in this study, reproducibility of LSM either intraobserver or interobserver has been previously considered satisfactory.26, 36
In 37 of 45 patients without cirrhosis with LSM ≥ 14.6 kPa (82%), increased LSM value could be related to extensive fibrosis. For 27 of them with F3 fibrosis, results of LB could be considered as mostly false negative. Even if needle LB has been used as the “gold standard” for the assessment of liver fibrosis, several authors have shown that a significant percentage of patients with cirrhosis can be misclassified as F3. When three percutaneous LB were performed in the same patients using different entry points, the overall concordance rate for cirrhosis, that is, a histopathological feature of cirrhosis in all three biopsy specimens, was only 50%.37 Similarly, Abdi et al.38 performed several post mortem biopsies and showed that the diagnosis of cirrhosis could be obtained from one specimen in only 80% of cases. According to Bedossa et al.,8 sampling variation of liver fibrosis is an important limitation in the assessment of fibrosis with LB and the median LB length in our study (17 mm), reflecting clinical practice, was far less than the optimal length defined by Bedossa et al.8 However, as suggested by Rousselet et al.,39 the pathologist's level of experience has more influence on agreement than the length of biopsy specimen. Ten additional patients had mild to moderate portal or centrolobular fibrosis associated with marked perisinusoidal fibrosis. This kind of misclassification is related to the principle of Fibroscan that seems to accurately reflect liver fibrosis whatever its location and influence on liver architecture. Increased LSM was also associated with extensive necrosis because of acute rebound of chronic hepatitis in three cases. Thus, even if histological activity was not related to LSM in HCV patients,30 likely severe necrosis and inflammation could influence LSM. Lastly, LSM results could be considered as false positives in six patients without significant fibrosis on the biopsy sample, including two patients with vascular liver disease and four in whom no explanation for increased LSM was found, suggesting technical errors in these later cases.
In all 35 patients with cirrhosis with LSM less than 14.6 kPa, LSM results could be considered as false negatives. The main histological findings in these patients were the macronodular pattern of cirrhosis and the absence of significant histological activity reflecting inactive or even cured diseases. These findings explain why the performance accuracy of LSM in the diagnosis of cirrhosis was slightly inferior in patients with HBV infection, the main cause of macronodular cirrhosis.
These results, combined with those of a study assessing the area of fibrosis measured by morphometry in patients who had LSM (M. Ziol, personal communication), suggest that elastography accurately reflects the amount of fibrosis. Portal fibrosis, which is the hallmark of chronic hepatitis, and perisinusoidal fibrosis, which is more common in alcoholic or non-alcoholic steatohepatitis, both contribute to increasing LSM. Perisinusoidal fibrosis modifies liver stiffness additionally to the increase attributable to portal fibrosis. Moreover, although the amount of fibrosis in micronodular cirrhosis is usually high and the diagnosis of cirrhosis correctly assessed in this case by Fibroscan, transient elastography is not such an accurate tool for assessing liver architectural abnormalities with a limited amount of fibrosis such as in macronodular cirrhosis, characterized by large nodules delimited by very thin septa. Similar discrepancies were observed in F3-F4 patients with chronic C hepatitis previously treated by interferon or with sustained virological response, a condition that could contribute to decreased liver fibrosis. In all cases, a limitation of LSM was its inability to diagnose the grade of necroinflammatory activity.
Despite these limitations, and the fact that we cannot rule out that the agreement between LB and Fibroscan may result from similar sampling errors, our study suggests that LSM is an accurate method for the diagnosis of cirrhosis whatever the cause of CLD. A cutoff of 9.2 kPa could help to exclude cirrhosis with a negative predictive value of 99%. A cutoff of 14.6 kPa would identify cirrhosis with a positive predictive value of 74%. These diagnostic performances could even be refined if specific cutoffs according to the cause of CLD were used. Taken together, false-positive LSM values are mainly observed in patients with extensive fibrosis and could correspond to an underdiagnosed cirrhosis by LB, or extensive fibrosis without nodular architecture. False negatives are mostly attributable to inactive or macronodular cirrhosis. These results obtained in a French multicentric study must be validated externally in other large cohorts of patients.
- 2Natural history of hepatitis B. In: EASL International Consensus Conference on Hepatitis B 2002; Genève; 2002: 47–62..
- 12AST/ALT ratio predicts cirrhosis in patients with chronic hepatitis C virus infection. Am J Gastroenterol 1998; 93: 44–48., , , .Direct Link:
- 14Serum aminotransferase levels and platelet counts as predictors of degree of fibrosis in chronic hepatitis C virus infection. Am J Gastroenterol 2001; 96: 3142–3146., , , , , .Direct Link:
- 16Need for validation of clinical decision aids: use of the AST/ALT ratio in predicting cirrhosis in chronic hepatitis C. Am J Gastroenterol 2000; 95: 2328–2332., , , .Direct Link:
- 35Nonalcoholic steatohepatitis: a proposal for grading and staging the histological lesions. Am J Gastroenterol 1999; 94: 2467–2474., , , , .Direct Link: