Transient elastography accurately predicts presence of significant portal hypertension in patients with chronic liver disease


Dr C. Bureau, Service d’Hépato-Gastro-Entérologie, Fédération Digestive, Purpan, CHU, Toulouse, 31059 Toulouse cedex, France.


Background  Hepatic venous pressure gradient (HVPG) is a prognostic marker in patients with cirrhosis. Transient elastography measures liver stiffness (LS).

Aim  To assess the correlation between LS and HVPG and to investigate the performance of transient elastography for the diagnosis of significant portal hypertension (PHT).

Methods  Liver stiffness was measured by Fibroscan in 150 consecutive patients who underwent a liver biopsy with haemodynamic measurements. Usual clinical and biological data were collected. Significant PHT was defined as a HVPG ≥10 mmHg.

Results  Hepatic venous pressure gradient was found to be ≥10 mmHg in 76 patients. Cirrhosis was diagnosed in 89 patients. HVPG was found to be correlated with: LS (ρ = 0.858; < 0.001) and inversely correlated with prothrombin index (ρ = −0.718; < 0.001). Regarding significant PHT, AUROC for LS and prothrombin index were 0.945 [0.904–0.987] and 0.892 [0.837–0.947] respectively. The cut-off value of 21 kPa accurately predicted significant PHT in 92% of the 144 patients for whom LS was successful.

Conclusion  Liver stiffness measurement is correlated with HVPG and transient elastography identifies patients with significant PHT.


The prognosis and management of chronic liver diseases strongly depend on the severity of portal hypertension (PHT). Hepatic venous pressure gradient (HVPG) has been shown to be a prognostic marker in different clinical settings: prediction of digestive bleeding,1, 2 uncontrolled digestive bleeding and rebleeding,3–5 risk of liver decompensation,6–8 liver failure after surgery9 and more recently in identifying patients at risk for disease progression after liver transplantation.10 However, HVPG measurement by the catheterism of a hepatic vein is considered an invasive procedure, not cost effective and requires care and training.11 Until now, non-invasive methods to evaluate HVPG, mainly Doppler US, have remained inaccurate. In clinical practice, the degree of PHT is partially evaluated by upper gastrointestinal endoscopy using the staging of varices.12

Non-invasive methods that measure the degree of liver fibrosis have recently been developed, including transient elastography [FibroScan (FS), Echosens, Paris, France].13 Several reports have shown that liver stiffness (LS), measured by transient elastography, accurately predicts liver fibrosis in patients with chronic hepatitis C and chronic cholestatic liver diseases.14–16 A recent study in cirrhotic patients suggested that the higher the value of LS, the higher the risk of liver decompensation.17

The resistance within the liver, which is an important determinant of PHT, may be assessed through the measurement of liver stiffness. In keeping with this, Kazemi et al. reported that LS accurately predicts the presence of large oesophageal varices.18 However, the risk of bleeding does not depend only on variceal size19 and other complications of PHT cannot be predicted by the size of varices.20 Recently, two studies aiming to assess non-invasive methods for the diagnosis of PHT showed that LS21 and Fibrotest22 were correlated with HVPG. The aim of this study was to assess the relationship between liver stiffness and hepatic venous pressure gradient and to investigate the performance of transient elastography for the diagnosis of PHT.

Patients and methods


Between 15 November 2005 and 15 October 2006, LS was prospectively measured by FS in all patients who underwent a trans-jugular liver biopsy with haemodynamic measurements in our unit. All patients with liver biopsy because of chronic liver abnormalities were considered for inclusion. Only patients with portal vein thrombosis or Budd–Chiari syndrome were excluded from this study. Usual clinical and biological data were prospectively collected. Following the recommendations in patients with cirrhosis, an upper gastrointestinal endoscopy was performed by experienced endoscopists. Varices were classified as absent, small or large in accordance with the description of de Franchis et al.23 Patients were enrolled after they had given their written informed consent. At inclusion, no patients had antiviral therapy or pharmacological treatment known to modify portal pressure.

LS measurement by FS

Liver stiffness, hemodynamic measurement and liver biopsy were performed on the same day. As previously described,14 LS was assessed on the right lobe of the liver, through inter-costal spaces with the patient in the supine position and the right arm in maximal abduction. Ten validated measures were performed in each patient. The success rate was calculated as the number of validated measures divided by the total number of measures. Results were expressed in kilopascals (kPa). The median value was considered representative of the elastic modulus of the liver. Only procedures with 10 validated measures and a success rate of at least 60% were considered reliable. The operator was blinded to clinical data and to the results of liver biopsy.

Haemodynamic measurements

The haemodynamic study was performed after a 12-h fast, in the supine position according to a procedure previously described. HVPG was calculated as wedged hepatic venous pressure minus free hepatic venous pressure and expressed in millimeters of mercury (mmHg). PHT was defined as a HVPG >5 mmHg and significant PHT as a HVPG ≥10 mmHg.24

Liver biopsy

Liver biopsy was performed by the transjugular route after pressure measurements. Specimens were fixed in formalin and embedded in paraffin. All biopsy specimens were analysed by one experienced pathologist blinded to the results of LS. The length of each liver biopsy specimen (in millimeters) was recorded after fixation. Liver fibrosis and necroinflammatory activity were evaluated semi quantitatively according to the METAVIR scoring system.25

Statistical analysis

Given the non-Gaussian distribution of clinical variables, non-parametric tests were used, whenever appropriate. Similarly, the overall correlation between quantitative variables was determined using the Spearman Rho’s pair-wise coefficient matrix analysis between the variables. To assess the performance of transient elastography in predicting significant PHT, we calculated the area under the ROC curves (AUROC) and sensitivity, specificity, positive and negative predictive values. HVPG values were dichotomized using a 10 mmHg threshold according to the definition of significant PHT proposed by the experts’ conference of Baveno.24 Other thresholds used in the analysis are indicated in the tables. Odds ratios and adjusted odds ratios were calculated both in univariate and in multivariate analyses. Confidence intervals are given at 95%. ROC curves were compared with Hanley’s test.


Patients’ characteristics

One hundred and fifty consecutive patients underwent a trans-jugular liver biopsy during the study period and were included. Their main clinical and biochemical characteristics are presented in Table 1. HVPG was found to be higher than 5 mmHg in 101 patients (67%) and higher than 10 mmHg in 76 patients (51%). The mean size of the liver biopsy was 15.5 ± 7.1 mm. Cirrhosis was diagnosed in 89 patients (59%) of whom 68 (77%) had compensated cirrhosis. LS could be measured in 144/150 (96%). Causes of failure were obesity in four patients and unrecognized in two patients. The mean success rate of LS acquisition was 85 ± 15% when excluding the six patients for whom LS was not available according to the criteria defined in Methods section.

Table 1.   Main characteristics of the patients
 All patients
  1. Results are expressed as raw numbers (and percentage) for qualitative variables, as median [and inter-quartile range] for quantitative variables.

Gender (male – female) %90 (60%) – 60 (40%)
Age (years)55 [45–65]
Cause of liver abnormalities
 Alcohol51 (34%)
 Viral hepatitis46 (31%)
 Mixed (alcohol + virus)7 (5%)
 NASH11 (7%)
 Autoimmune hepatitis10 (7%)
 Cholestatic disease6 (4%)
 Miscellaneous19 (12%)
BMI (kg/m2)24.4 [21.0–27.1]
HVPG (mmHg)10.0 [4–17]
Liver stiffness (kPa)17.1 [7.3–48.8]
Prothrombin index (%)79 [63–95]
Platelets (count/mm3)151 000 [93 000–232 000]
γGT (IU/L) 176 [59–316]
Serum albumin (g/L)35.0 [29.0–42.0]
Serum bilirubin (μmol/L)16.0 [10.6–45.0]
AST (IU/L)62 [36–96]
ALT (IU/L)48 [30–94]
HVPG ≥10 mmHg76 (51%)
Cirrhosis89 (59%)
  Absent25 (28%)
  Large43 (48%)
 Child Pugh score7.5 [5–10]
 Child Pugh Class A/B/C30/38/21

Correlation of HVPG with other parameters

The matrix of pair wise correlations using Spearman ρ-test is presented in Table 2. HVPG was found to be correlated with LS, age, serum bilirubin, concentration, γGT, AST and inversely correlated with prothrombin index, platelets count, serum albumine concentration and ALT. The highest correlation coefficients were found for LS and prothrombin index. Stratifying the correlation analysis according to the presence or not of cirrhosis, HVPG remained significantly correlated with LS, prothrombin index, γGT, serum albumin and serum bilirubin concentrations.

Table 2.   Correlations between HVPG and the different parameters
 HVPGLSProthrombin indexAgePlateletsγGTSerum albuminSerum bilirubinAST
  1. The overall correlation between quantitative variables was determined using the Spearman Rho’s pair-wise coefficient matrix analysis. The highest correlation coefficients were found between HVPG on one hand, and LS and prothrombin index, on the other (bold characters). All the P-values are given uncorrected.

< 0.001
Prothrombin index−0.718
< 0.001
< 0.001
< 0.001
= 0.001
= 0.010
< 0.001
< 0.001
< 0.001
= 0.762
< 0.01
< 0.001
= 0.006
= 0.593
= 0.241
Serum albumin−0.647
< 0.001
< 0.001
< 0.001
< 0.001
< 0.001
= 0.001
Serum bilirubin0.600
< 0.001
< 0.001
< 0.001
= 0.161
= 0.001
< 0.001
< 0.001
< 0.01
< 0.001
= 0.001
= 0.122
= 0.249
< 0.001
< 0.001
< 0.001
= 0.030
= 0.137
= 0.047
< 0.001
= 0.176
< 0.001
= 0.176
= 0.587
< 0.001

Parameters associated with significant PHT

In the whole population, HVPG ≥10 mmHg was found to be associated with higher age, presence of cirrhosis, higher LS value, higher bilirubin concentration, higher γGT and AST levels, a lower prothrombin index, a lower serum albumin concentration, a lower ALT level and a lower platelets count (Table 3).

Table 3.   Comparison of patients with HVPG ≥10 mmHg or HVPG <10 mmHg
 HVPG <10 mmHgHVPG ≥10 mmHgP
  1. Raw numbers (and percentage) are given for qualitative variables, median [and inter-quartile range] for quantitative variables. Non-parametric comparisons were used when the observed distribution was not in agreement with parametric test assumptions.

Gender (male – female) %45 (61%) – 29 (39%)45 (59%) – 31 (41%)N.S.
Age (years)50 [40–60]57 [47–68]<0.001
BMI (kg/m2)23.8 [20.4–27.0]25.1 [21.3–27.5]N.S.
Liver stiffness (kPa)7.9 [5.6–11.8]43.3 [33.3–75.0] <0.001
Prothrombin index (%)94 [86–100]65 [53–76]<0.001
Platelets (mm3/mL)198 000 [127 000–259 000]125 000 [80 500–167 500]<0.001
γGT (IU/L)103 [52–229]236 [112–385]<0.001
Serum albumin (g/L)42 [37–46]31 [25–34]<0.001
Serum bilirubin (μmol/L)12 [8–16]36 [17–83]<0.001
AST (IU/L)50 [32–80]67 [40–116]=0.03
ALT (IU/L)59 [33–125]40 [28–62]=0.01

The median values of liver stiffness in patients without or with significant PHT were 7.9 kPa [5.6–11.8] and 43.3 kPa [33.3–75.0] respectively (< 0.01). The median values of prothrombin index in patients without significant PHT and with significant PHT were 94% [86–100] and 65% [53–76] respectively (< 0.01). The highest Odds ratios for HVPG ≥ 0 mmHg were found with LS and prothrombin index (Table 4). In multivariate analysis, we suggest a model taking into account two parameters: LS and prothrombin index to explain the presence or the absence of significant PHT. According to this model, the adjusted Odds ratios for LS (cut-off value 21 kPa) and prothrombin index (cut-off value 82.5%) were 57.0 [16.1–201.7] and 8.4 [2.4–29.4] respectively.

Table 4.   Calculated odds ratio for the risk of significant PHT in univariate analysis
 HVPG ≥ 10 mmHgP
Odds ratio and 95% CI
  1. Odds ratio and confidence interval at 95% were calculated resulting in the dichotomization of the sample according to the threshold applied to the variables indicated in column 1. To assess the significance of the association with HVPG ≥10 mmHg, the P-values of chi-squared tests are reported in the last column.

Age >50 years3.1 [1.5–6.6]=0.0011
LS >21 kPa120.4 [32.5–485.4]<0.0001
Prothrombin index <82.5%28.9 [11.0–70.0]<0.0001
Platelets >150 000/mm30.23 [0.1–0.5]<0.0001
γGT >140 IU3.0 [1.5–6.3]=0.001
Serum albumin <35 g/L12.6 [5.2–30.9]<0.0001
Serum bilirubin >17 μmol/L11.4 [4.9–26.9]<0.0001
ALAT <55 IU2.6 [1.2–5.4]=0.005
ASAT >50 IU1.7 [0.2–2.6]=0.700

Performance of LS and prothrombin index for predicting significant PHT

AUROC for LS measurement was 0.945 [0.904–0.987] (Figure 1). The performance of LS for predicting significant portal hypertension according to several cut-off values is presented in Table 5. The optimal cut-off value was found to be 21 kPa with an accurate prediction of significant PHT in 92% of the 144 patients for whom LS was successful. AUROC for prothrombin index was 0.892 [0.837–0.947], not significantly different from AUROC observed with LS (Figure 1). Optimal cut-off value for prothrombin index was 82.5% with a sensitivity of 87%, a specificity of 81%, a positive predictive value of 83% and a negative predictive value of 85%. Accurate prediction of HVPG ≥10 mmHg was made in 84% of the 150 patients.

Figure 1.

 ROC curves of LS and prothrombin index for the prediction of significant PHT.

Table 5.   Performance of LS for predicting significant PHT according to several cut-off values
 Sensitivity (%)Specificity (%)PPV (%)NPV (%)OR (LR+/LR−)
  1. PPV, positive predictive value; NPV, negative predictive value; LR+, likelihood ratio positive; LR−, likelihood ratio negative.

11.7 kPa94.27477.493.146.2
13 kPa92.883.684.292.465.1
17 kPa92.887.787.792.891
21 kPa89.993.292.590.7120

We compared the performance of LS measurement in predicting significant PHT between the subgroup of patients with alcoholic liver disease (= 51) and the subgroup of patients with chronic viral hepatitis (= 46). AUROC were not significantly different: 0.906 [0.757–1.054] and 0.971 [0.923–1.018] for the former and the latter second group respectively.

Fifty eight patients presented with both a LS measure above 21 kPa and a prothrombin index below 82.5%. Among them, only two had a HVPG below 10 mmHg (8 mmHg for both patients). Conversely, LS below 21 kPa combined with prothrombin index above 82.5% identified 56 patients among whom only two had a HVPG > 10 mmHg (11 mmHg for both patients). Overall, using this combination, 96% of the 114 patients selected with those criteria were well classified. Accordingly, sensitivity, specificity, positive and negative predictive values were 99.6% in this subgroup of patients.

Value of LS and prothrombin index for the prediction of oesophageal varices in patients with cirrhosis

Liver stiffness was positively correlated with the presence of OV (Wilcoxon rank-sum, < 0.01). However, no association between LS and the size of OV was found (Kruskall-Wallis test). In multivariate analysis, parameters associated with the presence of OV and LOV were a lower prothrombin index, and a higher LS. AUROC for LS in predicting OV and LOV were: 0.851 [0.778–0.824] and 0.762 [0.672–0.852] respectively. Using a cut-off value of 21.1 kPa for the prediction of OV, sensitivity and specificity were respectively 84% and 71%. Using a cut-off value of 29.3 kPa for the prediction of LOV, sensitivity and specificity were respectively 81% and 61%.

AUROC for prothrombin index in predicting OV and LOV were not significantly different from LS: 0.843 [0.765–0.921] and 0.775 [0.688–0.863] respectively.


The main result of this study is that LS is highly correlated with HVPG and accurately predicts the presence of significant PHT in patients with chronic liver disease.

Transient elastography is a new non-invasive procedure that measures liver stiffness. LS was reported to be associated with liver fibrosis. It is a quantitative variable distributed over a wide range of values, which could provide more precise information than semi quantitative serological scores.

Liver stiffness has been prospectively assessed in three fully published studies in both chronic hepatitis C14, 15 and cholestatic liver diseases.16 In a retrospective study, Foucher et al. also found LS to be associated with the risk of decompensation in patients with cirrhosis.17 More recently, it has been suggested that LS could predict the presence and the size of oesophageal varices in a cohort of selected patients with cirrhosis.18 Finally, LS was assessed in patients after liver transplantation and more recently in 61 patients with chronic hepatitis C and was found to be correlated with HVPG.21, 26

In our whole sample of patients, we validated the accuracy of LS for the prediction of significant PHT. Indication for liver biopsy was the presence of liver function tests abnormalities. We chose to include patients with or without cirrhosis. By including only patients with cirrhosis, it would lead to select a population with a very high prevalence of significant PHT. Furthermore, non-invasive methods, particularly elastography and biochemical markers in chronic viral hepatitis, are presented as an alternative to liver biopsy. These non-invasive approaches could limit the use of pathology for the diagnosis of the cause of the liver disease whenever it is unclear on clinical and biological grounds. Therefore, an increasing number of patients will have a diagnosis of severe fibrosis or cirrhosis without histological proof. In patients with a well identified cause of liver disease, two questions remain to be answered: how severe is fibrosis? Is there a significant PHT? To answer these questions, both cirrhotic and non-cirrhotic patients have to be included.

In the present study, LS assessed by transient elastography was correlated with HVPG. This could result from a strong association between cirrhosis and PHT. Most patients with PHT in our study had cirrhosis as commonly experienced in Western countries and hence it is clear that prediction of PHT by LS could overlap with the prediction of cirrhosis. Actually, AUROC was found to be superior to 0.9 for predicting both cirrhosis and portal hypertension defined as HVPG >5 mmHg and optimal cut-off values were similar (data not shown). Nevertheless, LS was independently correlated with HVPG whether the patients had cirrhosis or not and using multivariate analysis, LS and prothrombin index were the two parameters highly associated with significant PHT. Seven patients without cirrhosis had significant PHT (of whom three had a LS measure >21 kPa) and 20 patients with cirrhosis had no significant PHT (of whom 15 had a LS measure <21 kPa). Therefore, the presence of significant PHT was not predicted by LS through the diagnosis of cirrhosis. Our results are in agreement with the previous study by Vizzutti et al.,21 which found a strong relationship between LS and HVPG. The main discrepant result is the LS cut-off value used for accurate prediction of significant PHT. This could be the result of the smaller sample of the previous study on one hand and on the other hand, a lower rate of decompensated cirrhoses. However, we cannot exclude that the different causes of liver abnormalities could have also influenced the cut-off value of LS. Nevertheless, if we applied our cut-off value (21 kPa) to the population of the previous report, we observed that no patient with HVPG < 10 mmHg had a LS > 21 kPa and less than 15% of patients with significant PHT had an LS < 21 kPa. Finally, the threshold chosen by Vizzutti et al. for significant PHT is similar to that retained for cirrhosis. Hence, this threshold provides no additional information and could not be used for selection of patients with cirrhosis.

In our study, liver diseases from different causes were included even though most patients had chronic alcoholic or chronic viral liver disease. The performance of LS for predicting cirrhosis was similar to that previously reported in chronic hepatitis C.14, 15 It is also noteworthy that the performance was similar in the two main subgroups of patients according to the cause of liver disease (alcoholic or viral hepatitis). No conclusion can be drawn in the other groups of patients. The good performance of LS in viral induced cirrhosis is important. In these patients, the diagnosis of cirrhosis is often made earlier in the course of the disease than in alcoholic patients because of the screening for chronic viral hepatitis and the strict monitoring of such patients. Therefore, the prevalence of PHT could be lower than that in patients with decompensated alcoholic liver disease and the usual endoscopic surveillance could be excessive in this subgroup of patients.

In the whole population of our patients, prothrombin index allowed also to predict significant PHT. When prothrombin index and LS were combined, significant PHT could be predicted in 1/3 of our patients (= 58) and significant PHT could be excluded in another 1/3 (= 56). Both procedures are non-invasive and easy to repeat. Previous studies have shown that the intra and inter-observer variability of LS measurement is very low.

Many parameters, clinical, biochemical, and physical are associated with PHT. As it was stated in Baveno’s conference by G Garcia Tsao et al., more accurate and non-invasive predictive models for PHT complications are needed.12 Most of them have been assessed for the prediction of varices and were found to have a poor prognostic value.27 In our study, platelet count could not accurately predict PHT. Doppler US is so far considered not accurate enough28, 29 to select patients at risk for bleeding or other complications of PHT.

In our study, the predictive performance of LS was better for significant PHT than for the presence of varices whatever their grade, and for the presence of large oesophageal varices. Vizzutti et al. observed the same result.21 This is in keeping with the fact that the development of porto-systemic collaterals is not strictly correlated with HVPG.1 One of the explanations is that after the development of PHT, other parameters than liver resistance are involved in the growing of esophageal varices. Initially, PHT is determined by an increase in resistance to blood flow through the liver. Thereafter, it is because of the increase in both resistance and blood flow within the portal vein. Portal blood flow itself is conditioned by splanchnic blood flow and the amount of collateral circulation. LS could be an indirect measure of resistance within the liver, but is not capable of quantifying blood flow and the extent of collateral circulation.

The prediction of significant PHT defined in our study by a HVPG ≥10 mmHg could be more relevant than the prediction of variceal size considering other complications of PHT such as ascites, hepatorenal syndrome or liver failure.

Interestingly, there was a trend towards a better performance of LS for assessing PHT compared with prothrombin index, but the difference was not significant. The performance of both parameters was similar for the prediction of oesophageal varices. In the recent study by Kazemi et al., prothrombin index was found to be significantly lower in patients with varices than in patients without varices.18 Unfortunately, the authors did not state or compare the performances of prothrombin index with LS for the prediction of varices.

We cannot provide a decision analysis taking into account the cost of the procedure. Cost analysis is a very important issue that clearly remains to be assessed in further studies. However, at least in France, such a study is not possible as there is no price approved by health authority. Furthermore, this non-invasive method could be useful in other indications than the prediction of PHT, such as non-invasive assessment of liver fibrosis. The carrying cost of this material should be calculated considering its different applications.

This study is a cross-sectional one and these patients have to be followed up to investigate the risk of liver disease related complications according to the value of liver stiffness.

In conclusion, LS is correlated with HVPG and, in patients with chronic liver abnormalities, the measurement of LS identifies patients with significant PHT. This method could therefore be assessed for its usefulness in a prognostic model and to select patients for upper gastrointestinal endoscopy.


Declaration of personal interests: None. Declaration of funding interests: This work was funded by CHU de Toulouse’, a grant from “Programme Activités Nouvelles.”