- Email: [email protected]
A Meta-analysis for the Diagnostic Performance of Transient Elastography for Clinically Significant Portal Hypertension
Ultrasound in Med. & Biol., Vol. -, No. -, pp. 1–10, 2016 Copyright Ó 2016 World Federation for Ultrasound in Medicine & Biology Printed in the USA. All rights reserved 0301-5629/$ - see front matter
http://dx.doi.org/10.1016/j.ultrasmedbio.2016.07.025
d
Review A META-ANALYSIS FOR THE DIAGNOSTIC PERFORMANCE OF TRANSIENT ELASTOGRAPHY FOR CLINICALLY SIGNIFICANT PORTAL HYPERTENSION MYUNG-WON YOU,* KYUNG WON KIM,*y JUNHEE PYO,z JIMI HUH,* HYOUNG JUNG KIM,*y SO JUNG LEE,*y and SEONG HO PARK* * Department of Radiology and Research Institute of Radiology, Asan Medical Center, University of Ulsan College of Medicine, Seoul, Korea; y Asan Liver Center, Asan Medical Center, University of Ulsan College of Medicine, Seoul, Korea; and z WHO Collaborating Center for Pharmaceutical Policy and Regulation, Department of Pharmaceutical Science, Utrecht University, Netherlands (Received 8 November 2015; revised 18 July 2016; in final form 31 July 2016)
Abstract—We aimed to evaluate the correlation between liver stiffness measurement using transient elastography (TE-LSM) and hepatic venous pressure gradient and the diagnostic performance of TE-LSM in assessing clinically significant portal hypertension through meta-analysis. Eleven studies were included from thorough literature research and selection processes. The summary correlation coefficient was 0.783 (95% confidence interval [CI], 0.737–0.823). Summary sensitivity, specificity and area under the hierarchical summary receiver operating characteristic curve (AUC) were 87.5% (95% CI, 75.8–93.9%), 85.3 % (95% CI, 76.9–90.9%) and 0.9, respectively. The subgroup with low cut-off values of 13.6–18 kPa had better summary estimates (sensitivity 91.2%, specificity 81.3% and partial AUC 0.921) than the subgroup with high cut-off values of 21–25 kPa (sensitivity 71.2%, specificity 90.9% and partial AUC 0.769). In summary, TE-LSM correlated well with hepatic venous pressure gradient and represented good diagnostic performance in diagnosing clinically significant portal hypertension. For use as a sensitive screening tool, we propose using low cut-off values of 13.6–18 kPa in TE-LSM. (E-mail: medimash@ gmail.com) Ó 2016 World Federation for Ultrasound in Medicine & Biology. Key Words: Transient elastography, Fibroscan, Portal hypertension, Liver stiffness measurement, Chronic liver disease.
gradient between the wedged (i.e., balloon-occluded) hepatic venous pressure and the free hepatic venous pressure and represents pressure from the hepatic sinusoids (Carrion et al. 2006; Colecchia et al. 2012; Hong et al. 2013). In patients with liver cirrhosis with sinusoidal portal hypertension, the inter-sinusodal communications are blocked due to fibrosis, and the hepatic sinusoidal pressure equilibrates with the portal perfusion pressure. A normal range of HVPG is 0–5 mmHg. In general, portal hypertension is subclinical when the HVPG is 6–9 mmHg and becomes clinically significant, resulting in complications, when the HVPG is $10 mmHg. Therefore, the term clinically significant portal hypertension (CSPH) is usually defined as a portal pressure gradient $10 mmHg, as estimated by HVPG in many reports (Berzigotti et al. 2013; Procopet et al. 2013). Measurement of HVPG requires a specialized angio-intervention unit. In addition, the invasive nature of the procedure and its high cost render HVPG measurement difficult to routinely perform in every medical
INTRODUCTION Portal hypertension is an increase in blood pressure in the portal venous system and is a primary consequence of liver fibrosis and cirrhosis. As the severity of portal hypertension increases, it can result in complications such as the development of esophageal varices, variceal bleeding, ascites, spontaneous bacterial peritonitis and hepatorenal syndrome (Bureau et al. 2008). Therefore, it is important to estimate the severity of portal hypertension to ensure appropriate patient management. Currently, the gold-standard method for estimating portal hypertension is measurement of the hepatic venous pressure gradient (HVPG) by catheterizing the hepatic vein using a balloon-tipped catheter. The HVPG is the
Address correspondence to: Kyung Won Kim, Department of Radiology and Research Institute of Radiology, University of Ulsan College of Medicine, Asan Medical Center, 88 Olympic-ro 43-gil, Songpagu, Seoul 138-736, Korea. E-mail: [email protected] 1
2
Ultrasound in Medicine and Biology
center for the purpose of making a diagnosis or providing longitudinal follow up. Therefore, there has been growing interest in developing novel, non-invasive approaches to assess the severity of portal hypertension. Transient elastography (TE) is an ultrasonographybased technology for liver stiffness measurement (LSM; hereafter referred as to TE-LSM) using the kilopascal scale and allows us to diagnose advanced liver fibrosis and cirrhosis (Augustin et al. 2014; Bureau et al. 2008). Recent studies have demonstrated that TE-LSM is well correlated with HVPG, can be used to estimate the severity of portal hypertension and can diagnose CSPH. Owing to the advantages of TE, including its noninvasiveness and ease of use, TE-LSM has begun to replace HVPG for diagnosing CSPH, at least at our medical institution. Although several studies have evaluated the correlation of TE-LSM and HVPG and the diagnostic performance of TE-LSM to detect CSPH, the majority were retrospective studies with low-level evidence. There is a previous meta-analysis to examine TE accuracy in the diagnosis of CSPH; however, only five studies were analyzed (Shi et al. 2013). Generating a more evidencebased systematic summary would be of great help for more evidence-based, standardized management of liver cirrhosis patients. For this purpose, we performed this systemic review and meta-analysis to determine the correlation between TE-LSM and HVPG and to evaluate the performance of TE-LSM for the screening and monitoring of CSPH.
MATERIALS AND METHODS Literature search strategy A comprehensive search of the PubMed MEDLINE and EMBASE databases was conducted to identify relevant, original publications regarding the correlation between TE-LSM and HVPG measurement and the diagnostic performance of TE-LSM to diagnose CSPH in patients with chronic liver disease. The following search terms were used: (liver cirrhosis OR portal hypertension) AND (elastography) AND (HVPG OR ‘‘hepatic vein pressure gradient’’). ‘‘Liver cirrhosis OR portal hypertension’’ was intended for the patient group and ‘‘elastography and hepatic vein pressure gradient’’ was intended for the imaging modality and its comparator. The search for bibliographies of included articles was also conducted. No beginning date limit was used. We searched the literature published before 30 August 2014. Our search was restricted to human patients and English-language studies. For management of the searched literature, Endnote version X7 (Thomson Reuters, New York, NY) was used.
Volume -, Number -, 2016
Inclusion criteria Studies investigating the correlation or diagnostic performance of TE for CSPH in patients with chronic liver disease were included. Studies that satisfied all of the following criteria were included: (i) population—patients who were evaluated for chronic liver disease and portal hypertension with both TE and HVPG; (ii) reference standard—CSPH should be defined as HVPG $10 mmHg; (iii) study designs—all observational studies (retrospective or prospective) and clinical trials that have received approval by an ethics committee or institutional review board; and (iv) outcomes—study results available to extract data such as correlation coefficient regarding the correlation between TE-LSM and HVPG and sensitivity, specificity regarding the diagnostic performance of TE-LSM to diagnose CSPH. Exclusion criteria The exclusion criteria were as follows: (i) case reports and case series with the sample size smaller than eight patients; (ii) review articles, editorials, letters, comments and conference proceedings; (iii) studies using imaging tools other than TE for liver stiffness measurement; (iv) studies that did not use HVPG as the reference standard; (v) studies that were not within the field of interest of this study; and (vi) studies with overlapping patients and data. Because CSPH is defined as HVPG $10 mmHg in this systematic review, we excluded studies which defined CSPH as HVPG $12 mmHg or other criteria. We included studies that used the transient TE using Fibroscan (EchoSens, Paris, France) and excluded other kinds of elastography, such as real time elastography, shear wave elastography, acoustic radiation force impulse elastography or magnetic resonance (MR) elastography, as TE has been widely used for the evaluation of liver fibrosis and was used in the majority of the included studies. We also excluded the study population with alcoholic hepatitis in the Lemoine et al. (2008) study (n 5 48) because most of the included study population had the virus-related chronic liver disease. Titles and abstracts identified by the search strategy were independently reviewed by two reviewers (K.W.K. and M.W.Y.). For all abstracts that met the inclusion/ exclusion criteria or were potentially eligible, full articles were retrieved, independently reviewed and a consensus was reached in cases of discrepancy by the same reviewers. Data extraction From the selected studies, we extracted the following data to standardized data forms: (i) study characteristics such as authors, year of publication, hospital or medical school and study design; (ii) demographic and clinical characteristics of the patients (i.e., age, sex, cause
Transient elastography for CSPH d M.-W. YOU et al.
of chronic liver disease, population size); (iii) imaging characteristics including definition of CSPH as the reference standard and the cut-off value of TE for diagnosing CSPH defined in each study; and (iv) outcomes, that is, the statistical method and results for investigating the correlation between TE-LSM and HVPG, a 2 3 2 table composed of the number of true positives, false positives, false negatives and true negatives of patients for diagnosing CSPH. Two reviewers (K.W.K. and M.W.Y.) independently extracted data from the studies, and all discrepancies were resolved by consensus. Quality assessment The methodological quality of the included studies was assessed by the Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) criteria in consensus between the two reviewers (K.W.K. and M.W.Y.) (Whiting et al. 2003). QUADAS-2 consists of four key domains covering patient selection, index test, reference standard and ‘‘flow and timing’’. ‘‘Flow and timing’’ includes flow of patients through the study and timing of the index test and reference standard. Each domain is assessed in terms of the ‘‘risk of bias’’ and except for ‘‘flow and timing.’’ The rest of the three domains are also assessed in terms of ‘‘concerns regarding applicability.’’ The studies are judged either ‘‘low,’’ ‘‘high’’ or ‘‘unclear’’ in each domain. ‘‘Low’’ indicates ‘‘low risk of bias’’ or ‘‘low concern regarding applicability,’’ and ‘‘high’’ or ‘‘unclear’’ indicate ‘‘at risk of bias’’ or ‘‘concerns regarding applicability.’’ Data synthesis for assessing the correlation between TE-LSM and HVPG To calculate the summary correlation coefficient (r) between TE-LSM and HVPG, we used the Hedges-Olkin method using a Fisher Z transformation of the correlation coefficients (Hedges and Olkin, 1985). With this method, the Pearson or Spearman correlation coefficients were converted to Z transforms after which a mean transformed correlation weighted by the sample size was calculated. Once a 95% confidence interval (CI) was obtained for the pooled Z score, it was transformed back to a 95% CI for the pooled r with both fixed and random effects models. Heterogeneity was assessed using the Cochran Q test based on the chi-square statistic. The pvalue ,0.10 of the Q test was considered to indicate substantial heterogeneity (Higgins et al. 2003). If heterogeneity existed, a random-effect model was used to obtain the summary correlation coefficient and its 95% CI. For the statistical analysis, R version 3.0.1 (R Foundation, Vienna, Austria) with the ‘‘metacor’’ package (Laliberte, 2011) and MedCalc version 14.8.1 (MedCalc Software, Mariakerke, Belgium) were used.
3
Data synthesis for calculating the diagnostic accuracy of TE-LSM We calculated descriptive statistics for the diagnostic accuracy of TE-LSM in order to diagnose CSPH, including the sensitivity, specificity, positive likelihood ratio (LR), negative LR and diagnostic odds ratio (DOR) from 2 3 2 tables of true positives, false positives, false negatives and true negatives. If 0 counts occurred in any cell in the 2 3 2 table, 0.5 was added to all the cell values for a continuity correction. The 95% CIs of these descriptive statistics were calculated based on the generic inverse variance method for calculating weights (Deeks, 2001). Heterogeneity of the sensitivity and specificity variables was assessed using the Cochran Q test, and a p-value ,0.10 was considered to indicate substantial heterogeneity (Higgins et al. 2003). As the threshold effect is an important cause of heterogeneity in meta-analysis of studies of diagnostic test accuracy, we explored whether the threshold effect existed by creating coupled forest plots of sensitivity and specificity. If there is a threshold effect, as the cut-off value of the diagnostic test is varied in order to increase the sensitivity, the specificity generally decreases (i.e., inverse correlation) (Reitsma et al. 2005). Therefore, we calculated the Spearman correlation coefficient between the sensitivity and specificity. Summary sensitivity and specificity were calculated using hierarchical logistic regression models including a bivariate model and a hierarchical summary receiver operating characteristic (HSROC) model. As the sensitivity and specificity are often interrelated according to the cut-off values, separate pooling of the sensitivity and specificity without consideration of correlation between them can be misleading. The traditional MosesLittenberg summary receiver operating characteristic (SROC) curve uses the DOR as the main outcome measures without incorporating the effect of a possible threshold, which also can be misleading. Therefore, to estimate the summary sensitivity and specificity, we used the bivariate model which preserves the 2-D nature of data in that pairs of sensitivity and specificity are jointly analyzed and incorporating any correlation between these two measures (Reitsma et al. 2005). The HSROC curve with a 95% confidence region and prediction region was also plotted to graphically present the results. This was particularly important for our meta-analysis as individual literature included in this study used different cutoff values of TE for the diagnosis of CSPH and the threshold effect was thereby expected. To explore the effect of cut-off values that differ in each study, a bivariate meta-regression with a covariate of the cut-off value of TE-LSM was performed. We then performed subgroup analysis based on similar cutoff values and set two groups as the lower cut-off value
4
Ultrasound in Medicine and Biology
group (13.6–18 kPa) versus the higher cut-off value group (21–25 kPa) (de Franchis and Baveno, 2015). Summary sensitivity, specificity and the bivariate SROC curve of each subgroup were calculated and compared based on the bivariate random effects model. Publication bias, which indicates that published studies might be systematically different from those of unpublished studies (mainly due to statistically nonsignificant results not being published), was visually assessed using the funnel plot that plots a DOR (i.e., diagnostic accuracy) against the standard error of DOR (i.e., measure of the study sample size). When there is no publication bias, the funnel plot shows symmetric shape; in the presence of publication bias, the funnel plot shows asymmetric shape. The statistical significance of the publication bias was tested using the Egger test (Egger et al. 1997). For the statistical analysis, we used ‘‘midas’’ and ‘‘metandi’’ modules in Stata 10.0 (StataCorp LP, College Station, TX, USA) and R version 3.0.1 with the ‘‘mada’’ package (Doebler and Holling, 2012).
RESULTS Literature search Our literature search process is illustrated in Figure 1. In the literature search using Ovid-MEDLINE and EMBASE databases, after removing duplicates, 103 articles were screened for eligibility. Of these, 76 articles were
Volume -, Number -, 2016
excluded after review of the titles and abstracts: 20 review articles; one case report; 52 letters, editorials or conference abstracts; and three original articles whose topics were outside the correlation between TE-LSM and HVPG. Full texts of the remaining 27 articles were retrieved. Search of bibliographies of these articles did not find additional eligible studies. The following 15 articles were further excluded after reviewing the full texts: seven articles whose topics were outside the field of interest of our study including studies evaluating for hepatic encephalopathy, decompensated liver cirrhosis, esophageal varices, histologic collagen proportionate area, two different pattern of fibrosis progression, effect of meal ingestion and outcome of antiviral therapy; four articles with insufficient data to generate a 2 3 2 table for the reference standard (i.e., CSPH) in this study; two articles evaluating the liver stiffness measurement using an imaging tool such as MR elastography or real-time elastography other than TE; and one article that contained a 2 3 2 table for portal hypertension-related complications other than CSPH. The other article that presented data using a study population overlapped with other studies (Berzigottis et al. 2013). The remaining 12 studies were used for qualitative and quantitative review (Augustin et al. 2014; Bureau et al. 2008; Carrion et al. 2006; Colecchia et al. 2012; Hong et al. 2013; Lemoine et al. 2008; Llop et al. 2012; Reiberger et al. 2012; Salzl et al. 2014; Sanchez-Conde et al. 2011; Schwabl et al., 2015; Vizzutti et al. 2008).
Fig. 1. Flow diagram of the study selection process.
188 61
Prospective Prospective Prospective
Retrospective Retrospective
Austria (NA) Austria (Feb. 2009 to Apr. 2010) Spain (Jan. 2007 to Jan. 2009)
Austria (NA) Italy (Mar. 2005 to Jun. 2006)
Reiberger et al. (2012) Salzl et al. (2014) Sanchez-Conde et al. (2011)
Schwabl et al. (2015) Vizzutti et al. (2008)
CLD HCV-related CLD
Prospective Retrospective Retrospective Prospective Italy (Sept. 2009 to Feb. 2011) Korea (Feb. 2009 to Feb. 2010) France (Jan. 2004 to Sept. 2006) Spain (Jun. 2007 to Apr. 2011) Colecchia et al. (2012) Hong et al. (2013) Lemoine et al. (2008) Llop et al. (2012)
CLD 5 chronic liver disease; HCV 5 hepatitis C virus; HIV 5 human immunodeficient virus; HVPG 5 hepatic venous pressure gradient; LC 5 liver cirrhosis; TE-LSM 5 liver stiffness measurement by transient elastography; NA 5 not available.
94.8%/86.9% 97%/92% 16.1 13.6
18 16.8 14 502 59 38
100 59 92 79
Prospective Prospective Prospective Spain (Jan. 2010 to Apr. 2014) France (Nov. 2005 to Oct 2006) Spain (Jul. 2004 to Oct. 2005) Augustin et al. (2014) Bureau et al. (2008) Carrion et al. (2006)
Asymptomatic CLD LC with esophageal varix HCV-infected liver transplant recipients HCV-induced LC LC Compensated LC Child-Pugh A with resectable liver tumors CLD LC HIV/HCV-coinfected patients
40 144 129
NA
21
82.2%/83.4% 89.7%/75% 92.8%/50%
Pearson, r 5 0.836 Spearman, r 5 0.704 Spearman, r 5 0.728 Linear regression analysis, r 5 0.552 Spearman, r 5 0.799 Spearman, r 5 0.765 Linear regression analysis, r 5 0.678 Spearman, r 5 0.846 Spearman, r 5 0.781 52.3%/97.1% 82.5%/73.7% NA 53%/91% 24.2 21.95
NA Spearman, r 5 0.858 Pearson, r 5 0.84 65%/93% 89.9%/93.2% 100%/60.8% 25 21 8.74
Correlation between LSM and HVPG Study design Author (y)
Location (Study period)
Eligible population
Sample size
Table 1. Characteristics of the included studies
Cut-off of TE-LSM (kPa)
Sensitivity/specificity for diagnosis of CSPH
Transient elastography for CSPH d M.-W. YOU et al.
5
Among the 12 articles, the Lemoine et al. (2008) study was used only for evaluating the correlation between TE-LSM and HVPG, and not for evaluating the diagnostic performance for CSPH as this study provided a 2 3 2 table extracted from a portion of the entire population. Augustin et al.’s (2014) study was used only for evaluating the diagnostic performance for CSPH as no relevant information regarding the correlation was provided in this study. Study characteristics and quality The basic characteristics of the final 12 studies included are summarized in Table 1. Almost all studies were conducted in Europe (four studies in Spain, two studies in France, two studies in Italy and three studies in Austria) except for one study conducted in Korea. Eight studies were prospective studies, and the remaining four studies were retrospective observational studies. The included patient populations were heterogeneous among the 12 studies. Half of the studies included patients with liver cirrhosis, and the other half of the studies included patients with chronic liver disease (CLD), according to the diagnostic criteria defined in each corresponding study. Among six studies with liver cirrhosis populations, one study included patients with liver cirrhosis with esophageal varix and another study included patients with Child-Pugh A liver cirrhosis with resectable hepatic tumor. Among the six studies with a CLD population, one study included patients with asymptomatic CLD, one study included HIV/HCV-coinfected patients and one study included HCV-infected patients after liver transplantation. The cut-off values of TE-LSM to diagnose CSPH varied among the studies and ranging from 8.74–25 kPa. The quality of the included studies, as assessed using the QUADAS-2 tool, was good overall (Fig. 2). Regarding the risk of bias, the domain of the flow and timing was insufficient in eight of the 12 studies, which lacked clear presentation or a description of the patient selection process. Regarding the applicability, the domain of the patient selection process was insufficient in six of the 12 studies in which they recruited part of the population from patients with chronic liver disease. Meta-analysis for correlation between TE-LSM and HVPG measurement The forest plot of the correlation analysis between TE-LSM and HVPG measurement are presented in Figure 3. As there was heterogeneity among 11 studies (p , 0.0001, Q test), the random effect model should be the preferred model (Bureau et al. 2008; Carrion et al. 2006; Colecchia et al. 2012; Hong et al. 2013; Lemoine et al. 2008; Llop et al. 2012; Reiberger et al. 2012; Salzl et al. 2014; Sanchez-Conde et al. 2011;
6
Ultrasound in Medicine and Biology
Volume -, Number -, 2016
Fig. 2. Quality Assessment of the Diagnostic Accuracy Studies-2 (QUADAS-2) criteria for the included studies.
Schwabl et al., 2015; Vizzutti et al. 2008). The summary correlation coefficient with 95% CI based on the randomeffect model was 0.783 (95% CI, 0.737–0.823) and thus suggested a fairly strong correlation liver stiffness measured by TE using the kilopascal scale and in HVPG as measured using the angio-interventional method with the mmHg scale. Meta-analysis for diagnostic accuracy In the 11 studies with a total of 1399 patients, the descriptive statistics of the diagnostic accuracy of TELSM to diagnose CSPH, including sensitivity, specificity, positive LR, negative LR and DOR, are presented in Table 2 (Bureau et al. 2008; Carrion et al. 2006; Colecchia et al. 2012; Hong et al. 2013; Llop et al. 2012; Reiberger et al. 2012; Salzl et al. 2014; SanchezConde et al. 2011; Schwabl et al., 2015; Vizzutti et al. 2008). The Cochran Q test revealed that there was significant heterogeneity between the studies regarding their sensitivity (c2 5 86.3, p , 0.001) and specificity (c2 5 56.7, p , 0.001). When we drew coupled forest plots of sensitivity and specificity in an ascending order
of sensitivity, there seemed to be an inverse correlation between the sensitivity and specificity, which may indicate a threshold effect (Fig. 4). When we calculated the correlation between the sensitivity and specificity, the Spearman correlation coefficient was 20.508 (95% CI, 20.132 to 0.849), which also suggests that a threshold effect existed. From the bivariate random effects model, the summary estimates of sensitivity, specificity, positive LR, negative LR and DOR were obtained (Table 2). The summary sensitivity of 87.5% (95% CI, 75.8– 93.9%) and specificity of 85.3% (95% CI, 76.9–90.9%) indicate that TE-LSM showed good diagnostic accuracy for the diagnosis of CSPH, which is also supported by the area under the curve (AUC) of HSROC curve of 0.9 (Fig. 5). Bivariate meta-regression analysis using a covariate of cut-off values revealed that the cut-off values of TELSM to diagnose CSPH is the important variable source of heterogeneity for estimating the summary sensitivity (z 5 24.086, p , 0.001) and specificity (z 5 23.231, p 5 0.001).
Fig. 3. Forest plots of the correlation coefficient between liver stiffness measurement using transient elastography and hepatic venous pressure gradient.
Transient elastography for CSPH d M.-W. YOU et al.
7
Table 2. Summary estimates of the diagnostic accuracy of TE to diagnose CSPH Study
Sensitivity (95% CI)
Specificity (95% CI)
Positive LR (95% CI)
Negative LR (95% CI)
DOR (95% CI)
Augustin et al. (2014) Bureau et al. (2008) Carrion et al. (2006) Colecchia et al. (2012) Hong et al. (2013) Llop et al. (2012) Reiberger et al. (2012) Salzl et al. (2014) Sanchez-Conde et al. (2011) Schwabl et al. (2015) Vizzutti et al. (2008) Bivariate summary estimates (95% CI)
65.4% (44.3–82.3%) 90.3% (81.0–96.0%) 00.0% (87.2–00.2%) 52.3% (39.5–64.5%) 82.5% (67.2–92.2%) 53.1% (34.7–70.7%) 82.2% (77.2–86.2%) 90.5% (77.4–97.4%) 92.9% (76.5–99.5%) 94.8% (85.6–98.6%) 97.9% (88.7–99.7%) 87.5% (75.8–93.9%)
92.9% (66.1–99.1%) 93.1% (84.5–97.5%) 60.8% (50.6–70.6%) 97.1% (85.1–99.1%) 73.7% (48.8–90.8%) 91.5% (79.6–97.6%) 83.2% (77.7–87.7%) 76.5% (50.1–93.1%) 50.0% (18.7–81.7%) 86.9% (79.9–92.9%) 92.9% (66.1–99.1%) 85.3% (76.9–90.9%)
9.2 (1.4–61.7) 13.0 (5.6–30.3) 2.5 (1.9–3.2) 18.3 (2.6–128.1) 3.1 (1.5–6.7) 6.2 (2.3–16.8) 4.9 (3.6–6.6) 3.8 (1.62–9.1) 1.9 (1.0–3.5) 7.3 (4.6–11.3) 13.7 (2.07–90.6) 5.9 (3.9–9.0)
0.4 (0.2–0.6) 0.1 (0.1–0.2) 0.02 (0.0–0.5) 0.5 (0.4–0.6) 0.2 (0.1–0.5) 0.5 (0.4–0.7) 0.2 (0.2–0.3) 0.1 (0.04–0.3) 0.1 (0.03–0.6) 0.06 (0.02–0.2) 0.02 (0.00–0.16) 0.1 (0.1–0.3)
24.5 (2.7–219.0) 124.4 (37.5–412.0) 84.8 (5.0–1430.7) 37.2 (4.8–288.8) 13.2 (3.6–48.76) 12.1 (3.5–42.0) 22.9 (14.3–36.5) 30.8 (6.7–141.4) 13.0 (1.9–86.8) 121.8 (34.2–433.4) 598.0 (34.9–10228.7) 40.3 (21.0–77.4)
CI 5 confidence interval; CSPH 5 clinically significant portal hypertension; DOR 5 diagnostic odds ratio; LR 5 likelihood ratio; TE 5 transient elastography.
Among 11 of the included studies, five studies with 422 patients used cut-off values within the range of 21– 25 kPa (Augustin et al. 2014; Bureau et al. 2008; Colecchia et al. 2012; Hong et al. 2013; Llop et al. 2012), and five studies with 848 patients used cut-off values within the range of 13.6–18 kPa (Reiberger et al. 2012; Salzl et al. 2014; Sanchez-Conde et al. 2011; Schwabl et al., 2015; Vizzutti et al. 2008). Therefore, we performed subgroup analysis to calculate the summary estimates of each subgroup. The subgroup using a low cut-off value of 13.6–18 kPa had a higher summary sensitivity (91.2%; 95% CI, 83.1–95.7%) and a lower specificity (81.3%; 95% CI, 70.4–88.8%) compared to the subgroup using high cut-off values of 21–25 kPa with summary sensitivity (71.2%; 95% CI, 52.1–84.9%) and specificity (90.9%; 95% CI, 81.6– 95.8%). We compared the bivariate SROC curves of each subgroup, which revealed that partial AUC at the observed false positive rate was higher in the low cutoff value subgroup than in the high cut-off value subgroup (0.921 vs. 0.769, respectively) (Fig. 6). No publication bias existed among these studies in funnel plot since the shape of funnel plot was symmetric (Fig. 7) and the Egger test result was not significant (p 5 0.308).
DISCUSSION Our study revealed that the liver stiffness values (in the kPa scale) measured by TE were highly correlated with HVPG (in the mmHg scale) measured by catheterization of the hepatic vein (summary correlation coefficient 0.783). The diagnostic accuracy of TE to diagnose CSPH was high, as evidenced by the summary sensitivity and specificity of 87.5% and 85.3%, respectively, and the AUC of the HSROC curve of 0.9. These results indicate that TE is a reliable, non-invasive method to assess portal hypertension and can be used for the screening and diagnosis of CSPH. It may be used for monitoring the disease status during the management of patients with liver cirrhosis as well (Jung and Kim, 2012; Wang et al. 2014). In our meta-analysis, the most critical issue used to calculate the summary estimates of diagnostic accuracy is the between-study heterogeneity due to the threshold effect. Various cut-off values were adopted among the 11 studies, as presented in Table 1, and ranging from 8.74– 25 kPa. Indeed, the inverse relationship between the sensitivity and specificity according to the changes of cut-off values was noted. In general, the higher cut-off values resulted in lower sensitivity and higher specificity. The most statistically significant methods used to
Fig. 4. Coupled forest plots of the sensitivity and specificity of transient elastography for diagnosing clinically significant portal hypertension. Forest plots of the sensitivity and specificity are plotted in ascending order of sensitivity. There seems to be an inverse correlation between the sensitivity and specificity: When the sensitivity increases, the specificity decreases.
8
Ultrasound in Medicine and Biology
Volume -, Number -, 2016
Fig. 6. Subgroup analysis according to the cut-off values. In the bivariate summary receiver operating characteristic curves of each subgroup, partial area under the curve was higher in the low cut-off value subgroup (0.921) than in the high cut-off value subgroup (0.769). Fig. 5. Hierarchical summary receiver operating characteristic (HSROC) curve of the diagnostic accuracy of transient elastography to diagnose clinically significant portal hypertension. The summary point of the sensitivity and specificity, HSROC curve, 95% confidence region and 95% prediction region are provided. The size of the dots of each study estimate was derived from the respective sample size. The area under the curve of the HSROC was 0.90.
value of 13.6–18 kPa to decrease false negative cases and increase sensitivity to enable early detection of CSPH. In patients with chronic liver disease, CSPH should be diagnosed as early as possible in order to strictly control liver function and aggressively prevent further deterioration and complications such as
determine this threshold effect are the bivariate model and the HSROC model as they model the study-based binomial structure of the data while accounting for the between-study heterogeneity (Reitsma et al. 2005; Rutter and Gatsonis, 2001). Indeed, major leading centers of meta-analysis research, including the Cochrane Collaboration and the National Institute for Health Research in the UK, concluded that the bivariate/HSROC model is highly recommended, together with an analysis of the summary ROC curves, credibility and prediction regions (Centre for Reviews and Dissemination 2009; Cochrane Collaboration 2013). According to Baveno VI consensus, higher TE-LSM value (i.e., $20–25 kPa) alone or combined platelet count and spleen size can diagnose CSPH non-invasively (de Franchis and Baveno, 2015). However, in our subgroup analysis, the subgroup with the lower cut-off value of 13.6–18 kPa showed better summary estimates (sensitivity 91.2%, specificity 81.3% and partial AUC 0.921) than the higher cut-off value of 21–25 kPa (sensitivity 71.2%, specificity 90.9% and partial AUC 0.769). Although Baveno VI consensus applies higher TE-LSM values to diagnose CSPH, our data show a lower cut-off
Fig. 7. Funnel plot of the diagnostic odds ratio of liver stiffness measurement using transient elastography to diagnose clinically significant portal hypertension. The x-axis represents the study’s result (i.e., diagnostic odds ratio). The y-axis represents study precision (i.e., standard error of diagnostic odds ratio). Each dot represents a study.
Transient elastography for CSPH d M.-W. YOU et al.
esophageal varix. Therefore, TE-LSM would be more appropriate for screening and monitoring tool rather than accurate diagnosis in patients with chronic liver disease; we recommend applying a lower cut-off value of TE for sensitive detection and screening as well as monitoring of CSPH in patients with chronic liver disease. TE is a well-established diagnostic elastographic method used to measure liver stiffness and is regarded as a standard method among various methods such as acoustic radiation force impulse, real-time elastography, MR elastography and shear-wave elastography (Bota et al. 2013; Chon et al. 2012; Ferraioli et al. 2013; Friedrich-Rust et al. 2008; Kobayashi et al. 2015; Stebbing et al. 2010). Most of the studies performed to determine the diagnostic accuracy of elastography for diagnosing liver cirrhosis or portal hypertension have focused on TE, and the other methods lack sufficient data. In their meta-analysis, Shi et al (2013) also reported high accuracy of TE in the diagnosis of CSPH (sensitivity 90%, specificity 79% and AUC 0.93), comparable to the results of our study (sensitivity 87.5%, specificity 85.3% and AUC 0.90). However, they included only five studies, which were also included in this meta-analysis, and did not consider the threshold effect. The strengths of our study include its validated systematic review methods and the reporting of results according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. We also followed the guidance of the Handbook for Diagnostic Test Accuracy Reviews published by the Cochran Collaboration (2013). We also extracted the maximum amount of information from the included studies by a thorough qualitative review and quantitative meta-analysis. Our study has several limitations. First, it included a relatively small number of studies (i.e., 11 studies). Among these 11 studies, five were retrospective studies with low-level evidence. However, the subject of our study comparing the TE value and HVPG measurement is not suitable for a randomized controlled trial, and meta-analysis can be the most appropriate alternative method to generate high-level evidence. Second, the eligible population of the included studies differs slightly in its etiology; however, all patients had the common factor of virus-related chronic liver disease. We excluded study populations with other etiologies of CLD, such as alcoholic hepatitis (Lemoine et al. 2008). Furthermore, as we primarily investigated the technical aspect of the index test, there would be little significance regarding the clinical aspect affected by a heterogeneous, eligible population. Third, various cut-off values of TE-LSM was found among included studies. However, we analyzed summary estimates of diagnostic accuracy with HSROC model and performed bivariate meta-regression analysis
9
to explain threshold heterogeneity.
effect
and
between-study
CONCLUSIONS In conclusion, the TE-LSM correlated well with the HVPG measurement and showed excellent diagnostic performance for diagnosing CSPH. Considering the advantages of TE, such as its non-invasiveness and short scan time, we believe that TE can be an effective diagnostic tool for CSPH. The cut-off value of TE-LSM for diagnosing CSPH should be standardized, and we suggest that a cut-off value of 13.6–18 kPa would be better for screening and monitoring than a cut-off value of 21– 25 kPa. Acknowledgments—This research was supported by the Basic Science Research Program through the National Research Foundation of Korea (NRF) (No. 2014 R1 A1 A1006823) and a grant from the Ministry of Health & Welfare, Republic of Korea (Grant No. HI14 C1090).
REFERENCES Augustin S, Millan L, Gonzalez A, Martell M, Gelabert A, Segarra A, Serres X, Esteban R, Genesca J. Detection of early portal hypertension with routine data and liver stiffness in patients with asymptomatic liver disease: A prospective study. J Hepatol 2014;60:561–569. Berzigotti A, Seijo S, Arena U, Abraldes JG, Vizzutti F, Garcia-Pagan JC, Pinzani M, Bosch J. Elastography, spleen size, and platelet count identify portal hypertension in patients with compensated cirrhosis. Gastroenterology 2013;144:102–111.e1. Bota S, Herkner H, Sporea I, Salzl P, Sirli R, Neghina AM, Peck-Radosavljevic M. Meta-analysis: ARFI elastography versus transient elastography for the evaluation of liver fibrosis. Liver Int 2013;33:1138–1147. Bureau C, Metivier S, Peron JM, Selves J, Robic MA, Gourraud PA, Rouquet O, Dupuis E, Alric L, Vinel JP. Transient elastography accurately predicts presence of significant portal hypertension in patients with chronic liver disease. Aliment Pharmacol Ther 2008;27: 1261–1268. Carrion JA, Navasa M, Bosch J, Bruguera M, Gilabert R, Forns X. Transient elastography for diagnosis of advanced fibrosis and portal hypertension in patients with hepatitis C recurrence after liver transplantation. Liver Transpl 2006;12:1791–1798. Centre for Reviews and Dissemination. Systematic reviews: CRD’s guidance for undertaking reviews in health care. York: University of York; 2009. Chon YE, Choi EH, Song KJ, Park JY, Kim do Y, Han KH, Chon CY, Ahn SH, Kim SU. Performance of transient elastography for the staging of liver fibrosis in patients with chronic hepatitis B: A meta-analysis. PloS One 2012;7:e44930. Colecchia A, Montrone L, Scaioli E, Bacchi-Reggiani ML, Colli A, Casazza G, Schiumerini R, Turco L, Di Biase AR, Mazzella G, Marzi L, Arena U, Pinzani M, Festi D. Measurement of spleen stiffness to evaluate portal hypertension and the presence of esophageal varices in patients with HCV-related cirrhosis. Gastroenterology 2012;143:646–654. Cochrane Screening and Diagnostic Tests Methods Group. Handbook for diagnostic test accuracy reviews. London, UK: The Cochrane Collaboration; 2013. de Franchis R, Baveno VIF. Expanding consensus in portal hypertension: Report of the Baveno VI Consensus Workshop: Stratifying risk and individualizing care for portal hypertension. J Hepatol 2015;63:743–752. Deeks JJ. Systematic reviews in health care: Systematic reviews of evaluations of diagnostic and screening tests. BMJ 2001;323:157–162.
10
Ultrasound in Medicine and Biology
Doebler P, Holling H. Meta-analysis of diagnostic accuracy with mada; 2012. Available at: https://cran.r-project.org/web/packages/ mada/vignettes/mada.pdf. Egger M, Davey Smith G, Schneider M, Minder C. Bias in meta-analysis detected by a simple, graphical test. BMJ 1997;315:629–634. Ferraioli G, Tinelli C, Dal Bello B, Zicchetti M, Lissandrin R, Filice G, Filice C, Above E, Barbarini G, Brunetti E, Calderon W, Di Gregorio M, Gulminetti R, Lanzarini P, Ludovisi S, Maiocchi L, Malfitano A, Michelone G, Minoli L, Mondelli M, Novati S, Patruno SF, Perretti A, Poma G, Sacchi P, Zanaboni D, Zaramella M. Performance of liver stiffness measurements by transient elastography in chronic hepatitis. World J Gastroenterol 2013; 19:49–56. Friedrich-Rust M, Ong MF, Martens S, Sarrazin C, Bojunga J, Zeuzem S, Herrmann E. Performance of transient elastography for the staging of liver fibrosis: A meta-analysis. Gastroenterology 2008;134:960–974. Hedges LV, Olkin I. Statistical methods for meta-analysis. London: Academic Press; 1985. Higgins JP, Thompson SG, Deeks JJ, Altman DG. Measuring inconsistency in meta-analyses. BMJ 2003;327:557–560. Hong WK, Kim MY, Baik SK, Shin SY, Kim JM, Kang YS, Lim YL, Kim YJ, Cho YZ, Hwang HW, Lee JH, Chae MH, Kim HA, Kang HW, Kwon SO. The usefulness of non-invasive liver stiffness measurements in predicting clinically significant portal hypertension in cirrhotic patients: Korean data. Clin Mol Hepatol 2013;19: 370–375. Jung KS, Kim SU. Clinical applications of transient elastography. Clin Mol Hepatol 2012;18:163–173. Kobayashi K, Nakao H, Nishiyama T, Lin Y, Kikuchi S, Kobayashi Y, Yamamoto T, Ishii N, Ohashi T, Satoh K, Nakade Y, Ito K, Yoneda M. Diagnostic accuracy of real-time tissue elastography for the staging of liver fibrosis: A meta-analysis. Eur Radiol 2015; 25:230–238. Laliberte E., Metacor: Meta-analysis with correlation coefficients. R package version 2011:1.0–2. Available at: https://cran.r-project. org/web/packages/metacor/metacor.pdf. Lemoine M, Katsahian S, Ziol M, Nahon P, Ganne-Carrie N, Kazemi F, Grando-Lemaire V, Trinchet JC, Beaugrand M. Liver stiffness measurement as a predictive tool of clinically significant portal hypertension in patients with compensated hepatitis C virus or alcohol-related cirrhosis. Aliment Pharmacol Ther 2008;28: 1102–1110. Llop E, Berzigotti A, Reig M, Erice E, Reverter E, Seijo S, Abraldes JG, Bruix J, Bosch J, Garcia-Pagan JC. Assessment of portal hypertension by transient elastography in patients with compensated cirrhosis and potentially resectable liver tumors. J Hepatol 2012;56: 103–108.
Volume -, Number -, 2016 Procopet B, Tantau M, Bureau C. Are there any alternative methods to hepatic venous pressure gradient in portal hypertension assessment? J Gastrointestin Liver Dis 2013;22:73–78. Reiberger T, Ferlitsch A, Payer BA, Pinter M, Homoncik M, Peck-Radosavljevic M. Non-selective beta-blockers improve the correlation of liver stiffness and portal pressure in advanced cirrhosis. J Gastroenterol 2012;47:561–568. Reitsma JB, Glas AS, Rutjes AW, Scholten RJ, Bossuyt PM, Zwinderman AH. Bivariate analysis of sensitivity and specificity produces informative summary measures in diagnostic reviews. J Clin Epidemiol 2005;58:982–990. Rutter CM, Gatsonis CA. A hierarchical regression approach to metaanalysis of diagnostic test accuracy evaluations. Stat Med 2001; 20:2865–2884. Salzl P, Reiberger T, Ferlitsch M, Payer BA, Schwengerer B, Trauner M, Peck-Radosavljevic M, Ferlitsch A. Evaluation of portal hypertension and varices by acoustic radiation force impulse imaging of the liver compared to transient elastography and AST to platelet ratio index. Ultraschall Med 2014;35:528–533. Sanchez-Conde M, Miralles P, Bellon JM, Rincon D, Ramirez M, Gutierrez I, Ripoll C, Lopez JC, Cosin J, Clemente G, Lo Iacono O, Banares R, Berenguer J. Use of transient elastography (FibroScan) for the noninvasive assessment of portal hypertension in HIV/HCV-coinfected patients. J Viral Hepat 2011;18:685–691. Schwabl P, Bota S, Salzl P, Mandorfer M, Payer B, Ferlitsch A, Stift J, Wrba F, Trauner M, Peck-Radosavljevic M, Reiberger T. New reliability criteria for transient elastography increase the number of accurate measurements for screening of cirrhosis and portal hypertension. Liver Int 2015;35:381–390. Shi KQ, Fan YC, Pan ZZ, Lin XF, Liu WY, Chen YP, Zheng MH. Transient elastography: A meta-analysis of diagnostic accuracy in evaluation of portal hypertension in chronic liver disease. Liver Int 2013;33:62–71. Stebbing J, Farouk L, Panos G, Anderson M, Jiao LR, Mandalia S, Bower M, Gazzard B, Nelson M. A meta-analysis of transient elastography for the detection of hepatic fibrosis. J Clinical Gastroenterol 2010;44:214–219. Vizzutti F, Arena U, Rega L, Pinzani M. Non invasive diagnosis of portal hypertension in cirrhotic patients. Gastroenterol Clin Biol 2008;32: 80–87. Wang JH, Chuah SK, Lu SN, Hung CH, Kuo CM, Tai WC, Chiou SS. Baseline and serial liver stiffness measurement in prediction of portal hypertension progression for patients with compensated cirrhosis. Liver Int 2014;34:1340–1348. Whiting P, Rutjes AW, Reitsma JB, Bossuyt PM, Kleijnen J. The development of QUADAS: A tool for the quality assessment of studies of diagnostic accuracy included in systematic reviews. BMC Med Res Methodol 2003;3:25.
http://dx.doi.org/10.1016/j.ultrasmedbio.2016.07.025
d
Review A META-ANALYSIS FOR THE DIAGNOSTIC PERFORMANCE OF TRANSIENT ELASTOGRAPHY FOR CLINICALLY SIGNIFICANT PORTAL HYPERTENSION MYUNG-WON YOU,* KYUNG WON KIM,*y JUNHEE PYO,z JIMI HUH,* HYOUNG JUNG KIM,*y SO JUNG LEE,*y and SEONG HO PARK* * Department of Radiology and Research Institute of Radiology, Asan Medical Center, University of Ulsan College of Medicine, Seoul, Korea; y Asan Liver Center, Asan Medical Center, University of Ulsan College of Medicine, Seoul, Korea; and z WHO Collaborating Center for Pharmaceutical Policy and Regulation, Department of Pharmaceutical Science, Utrecht University, Netherlands (Received 8 November 2015; revised 18 July 2016; in final form 31 July 2016)
Abstract—We aimed to evaluate the correlation between liver stiffness measurement using transient elastography (TE-LSM) and hepatic venous pressure gradient and the diagnostic performance of TE-LSM in assessing clinically significant portal hypertension through meta-analysis. Eleven studies were included from thorough literature research and selection processes. The summary correlation coefficient was 0.783 (95% confidence interval [CI], 0.737–0.823). Summary sensitivity, specificity and area under the hierarchical summary receiver operating characteristic curve (AUC) were 87.5% (95% CI, 75.8–93.9%), 85.3 % (95% CI, 76.9–90.9%) and 0.9, respectively. The subgroup with low cut-off values of 13.6–18 kPa had better summary estimates (sensitivity 91.2%, specificity 81.3% and partial AUC 0.921) than the subgroup with high cut-off values of 21–25 kPa (sensitivity 71.2%, specificity 90.9% and partial AUC 0.769). In summary, TE-LSM correlated well with hepatic venous pressure gradient and represented good diagnostic performance in diagnosing clinically significant portal hypertension. For use as a sensitive screening tool, we propose using low cut-off values of 13.6–18 kPa in TE-LSM. (E-mail: medimash@ gmail.com) Ó 2016 World Federation for Ultrasound in Medicine & Biology. Key Words: Transient elastography, Fibroscan, Portal hypertension, Liver stiffness measurement, Chronic liver disease.
gradient between the wedged (i.e., balloon-occluded) hepatic venous pressure and the free hepatic venous pressure and represents pressure from the hepatic sinusoids (Carrion et al. 2006; Colecchia et al. 2012; Hong et al. 2013). In patients with liver cirrhosis with sinusoidal portal hypertension, the inter-sinusodal communications are blocked due to fibrosis, and the hepatic sinusoidal pressure equilibrates with the portal perfusion pressure. A normal range of HVPG is 0–5 mmHg. In general, portal hypertension is subclinical when the HVPG is 6–9 mmHg and becomes clinically significant, resulting in complications, when the HVPG is $10 mmHg. Therefore, the term clinically significant portal hypertension (CSPH) is usually defined as a portal pressure gradient $10 mmHg, as estimated by HVPG in many reports (Berzigotti et al. 2013; Procopet et al. 2013). Measurement of HVPG requires a specialized angio-intervention unit. In addition, the invasive nature of the procedure and its high cost render HVPG measurement difficult to routinely perform in every medical
INTRODUCTION Portal hypertension is an increase in blood pressure in the portal venous system and is a primary consequence of liver fibrosis and cirrhosis. As the severity of portal hypertension increases, it can result in complications such as the development of esophageal varices, variceal bleeding, ascites, spontaneous bacterial peritonitis and hepatorenal syndrome (Bureau et al. 2008). Therefore, it is important to estimate the severity of portal hypertension to ensure appropriate patient management. Currently, the gold-standard method for estimating portal hypertension is measurement of the hepatic venous pressure gradient (HVPG) by catheterizing the hepatic vein using a balloon-tipped catheter. The HVPG is the
Address correspondence to: Kyung Won Kim, Department of Radiology and Research Institute of Radiology, University of Ulsan College of Medicine, Asan Medical Center, 88 Olympic-ro 43-gil, Songpagu, Seoul 138-736, Korea. E-mail: [email protected] 1
2
Ultrasound in Medicine and Biology
center for the purpose of making a diagnosis or providing longitudinal follow up. Therefore, there has been growing interest in developing novel, non-invasive approaches to assess the severity of portal hypertension. Transient elastography (TE) is an ultrasonographybased technology for liver stiffness measurement (LSM; hereafter referred as to TE-LSM) using the kilopascal scale and allows us to diagnose advanced liver fibrosis and cirrhosis (Augustin et al. 2014; Bureau et al. 2008). Recent studies have demonstrated that TE-LSM is well correlated with HVPG, can be used to estimate the severity of portal hypertension and can diagnose CSPH. Owing to the advantages of TE, including its noninvasiveness and ease of use, TE-LSM has begun to replace HVPG for diagnosing CSPH, at least at our medical institution. Although several studies have evaluated the correlation of TE-LSM and HVPG and the diagnostic performance of TE-LSM to detect CSPH, the majority were retrospective studies with low-level evidence. There is a previous meta-analysis to examine TE accuracy in the diagnosis of CSPH; however, only five studies were analyzed (Shi et al. 2013). Generating a more evidencebased systematic summary would be of great help for more evidence-based, standardized management of liver cirrhosis patients. For this purpose, we performed this systemic review and meta-analysis to determine the correlation between TE-LSM and HVPG and to evaluate the performance of TE-LSM for the screening and monitoring of CSPH.
MATERIALS AND METHODS Literature search strategy A comprehensive search of the PubMed MEDLINE and EMBASE databases was conducted to identify relevant, original publications regarding the correlation between TE-LSM and HVPG measurement and the diagnostic performance of TE-LSM to diagnose CSPH in patients with chronic liver disease. The following search terms were used: (liver cirrhosis OR portal hypertension) AND (elastography) AND (HVPG OR ‘‘hepatic vein pressure gradient’’). ‘‘Liver cirrhosis OR portal hypertension’’ was intended for the patient group and ‘‘elastography and hepatic vein pressure gradient’’ was intended for the imaging modality and its comparator. The search for bibliographies of included articles was also conducted. No beginning date limit was used. We searched the literature published before 30 August 2014. Our search was restricted to human patients and English-language studies. For management of the searched literature, Endnote version X7 (Thomson Reuters, New York, NY) was used.
Volume -, Number -, 2016
Inclusion criteria Studies investigating the correlation or diagnostic performance of TE for CSPH in patients with chronic liver disease were included. Studies that satisfied all of the following criteria were included: (i) population—patients who were evaluated for chronic liver disease and portal hypertension with both TE and HVPG; (ii) reference standard—CSPH should be defined as HVPG $10 mmHg; (iii) study designs—all observational studies (retrospective or prospective) and clinical trials that have received approval by an ethics committee or institutional review board; and (iv) outcomes—study results available to extract data such as correlation coefficient regarding the correlation between TE-LSM and HVPG and sensitivity, specificity regarding the diagnostic performance of TE-LSM to diagnose CSPH. Exclusion criteria The exclusion criteria were as follows: (i) case reports and case series with the sample size smaller than eight patients; (ii) review articles, editorials, letters, comments and conference proceedings; (iii) studies using imaging tools other than TE for liver stiffness measurement; (iv) studies that did not use HVPG as the reference standard; (v) studies that were not within the field of interest of this study; and (vi) studies with overlapping patients and data. Because CSPH is defined as HVPG $10 mmHg in this systematic review, we excluded studies which defined CSPH as HVPG $12 mmHg or other criteria. We included studies that used the transient TE using Fibroscan (EchoSens, Paris, France) and excluded other kinds of elastography, such as real time elastography, shear wave elastography, acoustic radiation force impulse elastography or magnetic resonance (MR) elastography, as TE has been widely used for the evaluation of liver fibrosis and was used in the majority of the included studies. We also excluded the study population with alcoholic hepatitis in the Lemoine et al. (2008) study (n 5 48) because most of the included study population had the virus-related chronic liver disease. Titles and abstracts identified by the search strategy were independently reviewed by two reviewers (K.W.K. and M.W.Y.). For all abstracts that met the inclusion/ exclusion criteria or were potentially eligible, full articles were retrieved, independently reviewed and a consensus was reached in cases of discrepancy by the same reviewers. Data extraction From the selected studies, we extracted the following data to standardized data forms: (i) study characteristics such as authors, year of publication, hospital or medical school and study design; (ii) demographic and clinical characteristics of the patients (i.e., age, sex, cause
Transient elastography for CSPH d M.-W. YOU et al.
of chronic liver disease, population size); (iii) imaging characteristics including definition of CSPH as the reference standard and the cut-off value of TE for diagnosing CSPH defined in each study; and (iv) outcomes, that is, the statistical method and results for investigating the correlation between TE-LSM and HVPG, a 2 3 2 table composed of the number of true positives, false positives, false negatives and true negatives of patients for diagnosing CSPH. Two reviewers (K.W.K. and M.W.Y.) independently extracted data from the studies, and all discrepancies were resolved by consensus. Quality assessment The methodological quality of the included studies was assessed by the Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) criteria in consensus between the two reviewers (K.W.K. and M.W.Y.) (Whiting et al. 2003). QUADAS-2 consists of four key domains covering patient selection, index test, reference standard and ‘‘flow and timing’’. ‘‘Flow and timing’’ includes flow of patients through the study and timing of the index test and reference standard. Each domain is assessed in terms of the ‘‘risk of bias’’ and except for ‘‘flow and timing.’’ The rest of the three domains are also assessed in terms of ‘‘concerns regarding applicability.’’ The studies are judged either ‘‘low,’’ ‘‘high’’ or ‘‘unclear’’ in each domain. ‘‘Low’’ indicates ‘‘low risk of bias’’ or ‘‘low concern regarding applicability,’’ and ‘‘high’’ or ‘‘unclear’’ indicate ‘‘at risk of bias’’ or ‘‘concerns regarding applicability.’’ Data synthesis for assessing the correlation between TE-LSM and HVPG To calculate the summary correlation coefficient (r) between TE-LSM and HVPG, we used the Hedges-Olkin method using a Fisher Z transformation of the correlation coefficients (Hedges and Olkin, 1985). With this method, the Pearson or Spearman correlation coefficients were converted to Z transforms after which a mean transformed correlation weighted by the sample size was calculated. Once a 95% confidence interval (CI) was obtained for the pooled Z score, it was transformed back to a 95% CI for the pooled r with both fixed and random effects models. Heterogeneity was assessed using the Cochran Q test based on the chi-square statistic. The pvalue ,0.10 of the Q test was considered to indicate substantial heterogeneity (Higgins et al. 2003). If heterogeneity existed, a random-effect model was used to obtain the summary correlation coefficient and its 95% CI. For the statistical analysis, R version 3.0.1 (R Foundation, Vienna, Austria) with the ‘‘metacor’’ package (Laliberte, 2011) and MedCalc version 14.8.1 (MedCalc Software, Mariakerke, Belgium) were used.
3
Data synthesis for calculating the diagnostic accuracy of TE-LSM We calculated descriptive statistics for the diagnostic accuracy of TE-LSM in order to diagnose CSPH, including the sensitivity, specificity, positive likelihood ratio (LR), negative LR and diagnostic odds ratio (DOR) from 2 3 2 tables of true positives, false positives, false negatives and true negatives. If 0 counts occurred in any cell in the 2 3 2 table, 0.5 was added to all the cell values for a continuity correction. The 95% CIs of these descriptive statistics were calculated based on the generic inverse variance method for calculating weights (Deeks, 2001). Heterogeneity of the sensitivity and specificity variables was assessed using the Cochran Q test, and a p-value ,0.10 was considered to indicate substantial heterogeneity (Higgins et al. 2003). As the threshold effect is an important cause of heterogeneity in meta-analysis of studies of diagnostic test accuracy, we explored whether the threshold effect existed by creating coupled forest plots of sensitivity and specificity. If there is a threshold effect, as the cut-off value of the diagnostic test is varied in order to increase the sensitivity, the specificity generally decreases (i.e., inverse correlation) (Reitsma et al. 2005). Therefore, we calculated the Spearman correlation coefficient between the sensitivity and specificity. Summary sensitivity and specificity were calculated using hierarchical logistic regression models including a bivariate model and a hierarchical summary receiver operating characteristic (HSROC) model. As the sensitivity and specificity are often interrelated according to the cut-off values, separate pooling of the sensitivity and specificity without consideration of correlation between them can be misleading. The traditional MosesLittenberg summary receiver operating characteristic (SROC) curve uses the DOR as the main outcome measures without incorporating the effect of a possible threshold, which also can be misleading. Therefore, to estimate the summary sensitivity and specificity, we used the bivariate model which preserves the 2-D nature of data in that pairs of sensitivity and specificity are jointly analyzed and incorporating any correlation between these two measures (Reitsma et al. 2005). The HSROC curve with a 95% confidence region and prediction region was also plotted to graphically present the results. This was particularly important for our meta-analysis as individual literature included in this study used different cutoff values of TE for the diagnosis of CSPH and the threshold effect was thereby expected. To explore the effect of cut-off values that differ in each study, a bivariate meta-regression with a covariate of the cut-off value of TE-LSM was performed. We then performed subgroup analysis based on similar cutoff values and set two groups as the lower cut-off value
4
Ultrasound in Medicine and Biology
group (13.6–18 kPa) versus the higher cut-off value group (21–25 kPa) (de Franchis and Baveno, 2015). Summary sensitivity, specificity and the bivariate SROC curve of each subgroup were calculated and compared based on the bivariate random effects model. Publication bias, which indicates that published studies might be systematically different from those of unpublished studies (mainly due to statistically nonsignificant results not being published), was visually assessed using the funnel plot that plots a DOR (i.e., diagnostic accuracy) against the standard error of DOR (i.e., measure of the study sample size). When there is no publication bias, the funnel plot shows symmetric shape; in the presence of publication bias, the funnel plot shows asymmetric shape. The statistical significance of the publication bias was tested using the Egger test (Egger et al. 1997). For the statistical analysis, we used ‘‘midas’’ and ‘‘metandi’’ modules in Stata 10.0 (StataCorp LP, College Station, TX, USA) and R version 3.0.1 with the ‘‘mada’’ package (Doebler and Holling, 2012).
RESULTS Literature search Our literature search process is illustrated in Figure 1. In the literature search using Ovid-MEDLINE and EMBASE databases, after removing duplicates, 103 articles were screened for eligibility. Of these, 76 articles were
Volume -, Number -, 2016
excluded after review of the titles and abstracts: 20 review articles; one case report; 52 letters, editorials or conference abstracts; and three original articles whose topics were outside the correlation between TE-LSM and HVPG. Full texts of the remaining 27 articles were retrieved. Search of bibliographies of these articles did not find additional eligible studies. The following 15 articles were further excluded after reviewing the full texts: seven articles whose topics were outside the field of interest of our study including studies evaluating for hepatic encephalopathy, decompensated liver cirrhosis, esophageal varices, histologic collagen proportionate area, two different pattern of fibrosis progression, effect of meal ingestion and outcome of antiviral therapy; four articles with insufficient data to generate a 2 3 2 table for the reference standard (i.e., CSPH) in this study; two articles evaluating the liver stiffness measurement using an imaging tool such as MR elastography or real-time elastography other than TE; and one article that contained a 2 3 2 table for portal hypertension-related complications other than CSPH. The other article that presented data using a study population overlapped with other studies (Berzigottis et al. 2013). The remaining 12 studies were used for qualitative and quantitative review (Augustin et al. 2014; Bureau et al. 2008; Carrion et al. 2006; Colecchia et al. 2012; Hong et al. 2013; Lemoine et al. 2008; Llop et al. 2012; Reiberger et al. 2012; Salzl et al. 2014; Sanchez-Conde et al. 2011; Schwabl et al., 2015; Vizzutti et al. 2008).
Fig. 1. Flow diagram of the study selection process.
188 61
Prospective Prospective Prospective
Retrospective Retrospective
Austria (NA) Austria (Feb. 2009 to Apr. 2010) Spain (Jan. 2007 to Jan. 2009)
Austria (NA) Italy (Mar. 2005 to Jun. 2006)
Reiberger et al. (2012) Salzl et al. (2014) Sanchez-Conde et al. (2011)
Schwabl et al. (2015) Vizzutti et al. (2008)
CLD HCV-related CLD
Prospective Retrospective Retrospective Prospective Italy (Sept. 2009 to Feb. 2011) Korea (Feb. 2009 to Feb. 2010) France (Jan. 2004 to Sept. 2006) Spain (Jun. 2007 to Apr. 2011) Colecchia et al. (2012) Hong et al. (2013) Lemoine et al. (2008) Llop et al. (2012)
CLD 5 chronic liver disease; HCV 5 hepatitis C virus; HIV 5 human immunodeficient virus; HVPG 5 hepatic venous pressure gradient; LC 5 liver cirrhosis; TE-LSM 5 liver stiffness measurement by transient elastography; NA 5 not available.
94.8%/86.9% 97%/92% 16.1 13.6
18 16.8 14 502 59 38
100 59 92 79
Prospective Prospective Prospective Spain (Jan. 2010 to Apr. 2014) France (Nov. 2005 to Oct 2006) Spain (Jul. 2004 to Oct. 2005) Augustin et al. (2014) Bureau et al. (2008) Carrion et al. (2006)
Asymptomatic CLD LC with esophageal varix HCV-infected liver transplant recipients HCV-induced LC LC Compensated LC Child-Pugh A with resectable liver tumors CLD LC HIV/HCV-coinfected patients
40 144 129
NA
21
82.2%/83.4% 89.7%/75% 92.8%/50%
Pearson, r 5 0.836 Spearman, r 5 0.704 Spearman, r 5 0.728 Linear regression analysis, r 5 0.552 Spearman, r 5 0.799 Spearman, r 5 0.765 Linear regression analysis, r 5 0.678 Spearman, r 5 0.846 Spearman, r 5 0.781 52.3%/97.1% 82.5%/73.7% NA 53%/91% 24.2 21.95
NA Spearman, r 5 0.858 Pearson, r 5 0.84 65%/93% 89.9%/93.2% 100%/60.8% 25 21 8.74
Correlation between LSM and HVPG Study design Author (y)
Location (Study period)
Eligible population
Sample size
Table 1. Characteristics of the included studies
Cut-off of TE-LSM (kPa)
Sensitivity/specificity for diagnosis of CSPH
Transient elastography for CSPH d M.-W. YOU et al.
5
Among the 12 articles, the Lemoine et al. (2008) study was used only for evaluating the correlation between TE-LSM and HVPG, and not for evaluating the diagnostic performance for CSPH as this study provided a 2 3 2 table extracted from a portion of the entire population. Augustin et al.’s (2014) study was used only for evaluating the diagnostic performance for CSPH as no relevant information regarding the correlation was provided in this study. Study characteristics and quality The basic characteristics of the final 12 studies included are summarized in Table 1. Almost all studies were conducted in Europe (four studies in Spain, two studies in France, two studies in Italy and three studies in Austria) except for one study conducted in Korea. Eight studies were prospective studies, and the remaining four studies were retrospective observational studies. The included patient populations were heterogeneous among the 12 studies. Half of the studies included patients with liver cirrhosis, and the other half of the studies included patients with chronic liver disease (CLD), according to the diagnostic criteria defined in each corresponding study. Among six studies with liver cirrhosis populations, one study included patients with liver cirrhosis with esophageal varix and another study included patients with Child-Pugh A liver cirrhosis with resectable hepatic tumor. Among the six studies with a CLD population, one study included patients with asymptomatic CLD, one study included HIV/HCV-coinfected patients and one study included HCV-infected patients after liver transplantation. The cut-off values of TE-LSM to diagnose CSPH varied among the studies and ranging from 8.74–25 kPa. The quality of the included studies, as assessed using the QUADAS-2 tool, was good overall (Fig. 2). Regarding the risk of bias, the domain of the flow and timing was insufficient in eight of the 12 studies, which lacked clear presentation or a description of the patient selection process. Regarding the applicability, the domain of the patient selection process was insufficient in six of the 12 studies in which they recruited part of the population from patients with chronic liver disease. Meta-analysis for correlation between TE-LSM and HVPG measurement The forest plot of the correlation analysis between TE-LSM and HVPG measurement are presented in Figure 3. As there was heterogeneity among 11 studies (p , 0.0001, Q test), the random effect model should be the preferred model (Bureau et al. 2008; Carrion et al. 2006; Colecchia et al. 2012; Hong et al. 2013; Lemoine et al. 2008; Llop et al. 2012; Reiberger et al. 2012; Salzl et al. 2014; Sanchez-Conde et al. 2011;
6
Ultrasound in Medicine and Biology
Volume -, Number -, 2016
Fig. 2. Quality Assessment of the Diagnostic Accuracy Studies-2 (QUADAS-2) criteria for the included studies.
Schwabl et al., 2015; Vizzutti et al. 2008). The summary correlation coefficient with 95% CI based on the randomeffect model was 0.783 (95% CI, 0.737–0.823) and thus suggested a fairly strong correlation liver stiffness measured by TE using the kilopascal scale and in HVPG as measured using the angio-interventional method with the mmHg scale. Meta-analysis for diagnostic accuracy In the 11 studies with a total of 1399 patients, the descriptive statistics of the diagnostic accuracy of TELSM to diagnose CSPH, including sensitivity, specificity, positive LR, negative LR and DOR, are presented in Table 2 (Bureau et al. 2008; Carrion et al. 2006; Colecchia et al. 2012; Hong et al. 2013; Llop et al. 2012; Reiberger et al. 2012; Salzl et al. 2014; SanchezConde et al. 2011; Schwabl et al., 2015; Vizzutti et al. 2008). The Cochran Q test revealed that there was significant heterogeneity between the studies regarding their sensitivity (c2 5 86.3, p , 0.001) and specificity (c2 5 56.7, p , 0.001). When we drew coupled forest plots of sensitivity and specificity in an ascending order
of sensitivity, there seemed to be an inverse correlation between the sensitivity and specificity, which may indicate a threshold effect (Fig. 4). When we calculated the correlation between the sensitivity and specificity, the Spearman correlation coefficient was 20.508 (95% CI, 20.132 to 0.849), which also suggests that a threshold effect existed. From the bivariate random effects model, the summary estimates of sensitivity, specificity, positive LR, negative LR and DOR were obtained (Table 2). The summary sensitivity of 87.5% (95% CI, 75.8– 93.9%) and specificity of 85.3% (95% CI, 76.9–90.9%) indicate that TE-LSM showed good diagnostic accuracy for the diagnosis of CSPH, which is also supported by the area under the curve (AUC) of HSROC curve of 0.9 (Fig. 5). Bivariate meta-regression analysis using a covariate of cut-off values revealed that the cut-off values of TELSM to diagnose CSPH is the important variable source of heterogeneity for estimating the summary sensitivity (z 5 24.086, p , 0.001) and specificity (z 5 23.231, p 5 0.001).
Fig. 3. Forest plots of the correlation coefficient between liver stiffness measurement using transient elastography and hepatic venous pressure gradient.
Transient elastography for CSPH d M.-W. YOU et al.
7
Table 2. Summary estimates of the diagnostic accuracy of TE to diagnose CSPH Study
Sensitivity (95% CI)
Specificity (95% CI)
Positive LR (95% CI)
Negative LR (95% CI)
DOR (95% CI)
Augustin et al. (2014) Bureau et al. (2008) Carrion et al. (2006) Colecchia et al. (2012) Hong et al. (2013) Llop et al. (2012) Reiberger et al. (2012) Salzl et al. (2014) Sanchez-Conde et al. (2011) Schwabl et al. (2015) Vizzutti et al. (2008) Bivariate summary estimates (95% CI)
65.4% (44.3–82.3%) 90.3% (81.0–96.0%) 00.0% (87.2–00.2%) 52.3% (39.5–64.5%) 82.5% (67.2–92.2%) 53.1% (34.7–70.7%) 82.2% (77.2–86.2%) 90.5% (77.4–97.4%) 92.9% (76.5–99.5%) 94.8% (85.6–98.6%) 97.9% (88.7–99.7%) 87.5% (75.8–93.9%)
92.9% (66.1–99.1%) 93.1% (84.5–97.5%) 60.8% (50.6–70.6%) 97.1% (85.1–99.1%) 73.7% (48.8–90.8%) 91.5% (79.6–97.6%) 83.2% (77.7–87.7%) 76.5% (50.1–93.1%) 50.0% (18.7–81.7%) 86.9% (79.9–92.9%) 92.9% (66.1–99.1%) 85.3% (76.9–90.9%)
9.2 (1.4–61.7) 13.0 (5.6–30.3) 2.5 (1.9–3.2) 18.3 (2.6–128.1) 3.1 (1.5–6.7) 6.2 (2.3–16.8) 4.9 (3.6–6.6) 3.8 (1.62–9.1) 1.9 (1.0–3.5) 7.3 (4.6–11.3) 13.7 (2.07–90.6) 5.9 (3.9–9.0)
0.4 (0.2–0.6) 0.1 (0.1–0.2) 0.02 (0.0–0.5) 0.5 (0.4–0.6) 0.2 (0.1–0.5) 0.5 (0.4–0.7) 0.2 (0.2–0.3) 0.1 (0.04–0.3) 0.1 (0.03–0.6) 0.06 (0.02–0.2) 0.02 (0.00–0.16) 0.1 (0.1–0.3)
24.5 (2.7–219.0) 124.4 (37.5–412.0) 84.8 (5.0–1430.7) 37.2 (4.8–288.8) 13.2 (3.6–48.76) 12.1 (3.5–42.0) 22.9 (14.3–36.5) 30.8 (6.7–141.4) 13.0 (1.9–86.8) 121.8 (34.2–433.4) 598.0 (34.9–10228.7) 40.3 (21.0–77.4)
CI 5 confidence interval; CSPH 5 clinically significant portal hypertension; DOR 5 diagnostic odds ratio; LR 5 likelihood ratio; TE 5 transient elastography.
Among 11 of the included studies, five studies with 422 patients used cut-off values within the range of 21– 25 kPa (Augustin et al. 2014; Bureau et al. 2008; Colecchia et al. 2012; Hong et al. 2013; Llop et al. 2012), and five studies with 848 patients used cut-off values within the range of 13.6–18 kPa (Reiberger et al. 2012; Salzl et al. 2014; Sanchez-Conde et al. 2011; Schwabl et al., 2015; Vizzutti et al. 2008). Therefore, we performed subgroup analysis to calculate the summary estimates of each subgroup. The subgroup using a low cut-off value of 13.6–18 kPa had a higher summary sensitivity (91.2%; 95% CI, 83.1–95.7%) and a lower specificity (81.3%; 95% CI, 70.4–88.8%) compared to the subgroup using high cut-off values of 21–25 kPa with summary sensitivity (71.2%; 95% CI, 52.1–84.9%) and specificity (90.9%; 95% CI, 81.6– 95.8%). We compared the bivariate SROC curves of each subgroup, which revealed that partial AUC at the observed false positive rate was higher in the low cutoff value subgroup than in the high cut-off value subgroup (0.921 vs. 0.769, respectively) (Fig. 6). No publication bias existed among these studies in funnel plot since the shape of funnel plot was symmetric (Fig. 7) and the Egger test result was not significant (p 5 0.308).
DISCUSSION Our study revealed that the liver stiffness values (in the kPa scale) measured by TE were highly correlated with HVPG (in the mmHg scale) measured by catheterization of the hepatic vein (summary correlation coefficient 0.783). The diagnostic accuracy of TE to diagnose CSPH was high, as evidenced by the summary sensitivity and specificity of 87.5% and 85.3%, respectively, and the AUC of the HSROC curve of 0.9. These results indicate that TE is a reliable, non-invasive method to assess portal hypertension and can be used for the screening and diagnosis of CSPH. It may be used for monitoring the disease status during the management of patients with liver cirrhosis as well (Jung and Kim, 2012; Wang et al. 2014). In our meta-analysis, the most critical issue used to calculate the summary estimates of diagnostic accuracy is the between-study heterogeneity due to the threshold effect. Various cut-off values were adopted among the 11 studies, as presented in Table 1, and ranging from 8.74– 25 kPa. Indeed, the inverse relationship between the sensitivity and specificity according to the changes of cut-off values was noted. In general, the higher cut-off values resulted in lower sensitivity and higher specificity. The most statistically significant methods used to
Fig. 4. Coupled forest plots of the sensitivity and specificity of transient elastography for diagnosing clinically significant portal hypertension. Forest plots of the sensitivity and specificity are plotted in ascending order of sensitivity. There seems to be an inverse correlation between the sensitivity and specificity: When the sensitivity increases, the specificity decreases.
8
Ultrasound in Medicine and Biology
Volume -, Number -, 2016
Fig. 6. Subgroup analysis according to the cut-off values. In the bivariate summary receiver operating characteristic curves of each subgroup, partial area under the curve was higher in the low cut-off value subgroup (0.921) than in the high cut-off value subgroup (0.769). Fig. 5. Hierarchical summary receiver operating characteristic (HSROC) curve of the diagnostic accuracy of transient elastography to diagnose clinically significant portal hypertension. The summary point of the sensitivity and specificity, HSROC curve, 95% confidence region and 95% prediction region are provided. The size of the dots of each study estimate was derived from the respective sample size. The area under the curve of the HSROC was 0.90.
value of 13.6–18 kPa to decrease false negative cases and increase sensitivity to enable early detection of CSPH. In patients with chronic liver disease, CSPH should be diagnosed as early as possible in order to strictly control liver function and aggressively prevent further deterioration and complications such as
determine this threshold effect are the bivariate model and the HSROC model as they model the study-based binomial structure of the data while accounting for the between-study heterogeneity (Reitsma et al. 2005; Rutter and Gatsonis, 2001). Indeed, major leading centers of meta-analysis research, including the Cochrane Collaboration and the National Institute for Health Research in the UK, concluded that the bivariate/HSROC model is highly recommended, together with an analysis of the summary ROC curves, credibility and prediction regions (Centre for Reviews and Dissemination 2009; Cochrane Collaboration 2013). According to Baveno VI consensus, higher TE-LSM value (i.e., $20–25 kPa) alone or combined platelet count and spleen size can diagnose CSPH non-invasively (de Franchis and Baveno, 2015). However, in our subgroup analysis, the subgroup with the lower cut-off value of 13.6–18 kPa showed better summary estimates (sensitivity 91.2%, specificity 81.3% and partial AUC 0.921) than the higher cut-off value of 21–25 kPa (sensitivity 71.2%, specificity 90.9% and partial AUC 0.769). Although Baveno VI consensus applies higher TE-LSM values to diagnose CSPH, our data show a lower cut-off
Fig. 7. Funnel plot of the diagnostic odds ratio of liver stiffness measurement using transient elastography to diagnose clinically significant portal hypertension. The x-axis represents the study’s result (i.e., diagnostic odds ratio). The y-axis represents study precision (i.e., standard error of diagnostic odds ratio). Each dot represents a study.
Transient elastography for CSPH d M.-W. YOU et al.
esophageal varix. Therefore, TE-LSM would be more appropriate for screening and monitoring tool rather than accurate diagnosis in patients with chronic liver disease; we recommend applying a lower cut-off value of TE for sensitive detection and screening as well as monitoring of CSPH in patients with chronic liver disease. TE is a well-established diagnostic elastographic method used to measure liver stiffness and is regarded as a standard method among various methods such as acoustic radiation force impulse, real-time elastography, MR elastography and shear-wave elastography (Bota et al. 2013; Chon et al. 2012; Ferraioli et al. 2013; Friedrich-Rust et al. 2008; Kobayashi et al. 2015; Stebbing et al. 2010). Most of the studies performed to determine the diagnostic accuracy of elastography for diagnosing liver cirrhosis or portal hypertension have focused on TE, and the other methods lack sufficient data. In their meta-analysis, Shi et al (2013) also reported high accuracy of TE in the diagnosis of CSPH (sensitivity 90%, specificity 79% and AUC 0.93), comparable to the results of our study (sensitivity 87.5%, specificity 85.3% and AUC 0.90). However, they included only five studies, which were also included in this meta-analysis, and did not consider the threshold effect. The strengths of our study include its validated systematic review methods and the reporting of results according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. We also followed the guidance of the Handbook for Diagnostic Test Accuracy Reviews published by the Cochran Collaboration (2013). We also extracted the maximum amount of information from the included studies by a thorough qualitative review and quantitative meta-analysis. Our study has several limitations. First, it included a relatively small number of studies (i.e., 11 studies). Among these 11 studies, five were retrospective studies with low-level evidence. However, the subject of our study comparing the TE value and HVPG measurement is not suitable for a randomized controlled trial, and meta-analysis can be the most appropriate alternative method to generate high-level evidence. Second, the eligible population of the included studies differs slightly in its etiology; however, all patients had the common factor of virus-related chronic liver disease. We excluded study populations with other etiologies of CLD, such as alcoholic hepatitis (Lemoine et al. 2008). Furthermore, as we primarily investigated the technical aspect of the index test, there would be little significance regarding the clinical aspect affected by a heterogeneous, eligible population. Third, various cut-off values of TE-LSM was found among included studies. However, we analyzed summary estimates of diagnostic accuracy with HSROC model and performed bivariate meta-regression analysis
9
to explain threshold heterogeneity.
effect
and
between-study
CONCLUSIONS In conclusion, the TE-LSM correlated well with the HVPG measurement and showed excellent diagnostic performance for diagnosing CSPH. Considering the advantages of TE, such as its non-invasiveness and short scan time, we believe that TE can be an effective diagnostic tool for CSPH. The cut-off value of TE-LSM for diagnosing CSPH should be standardized, and we suggest that a cut-off value of 13.6–18 kPa would be better for screening and monitoring than a cut-off value of 21– 25 kPa. Acknowledgments—This research was supported by the Basic Science Research Program through the National Research Foundation of Korea (NRF) (No. 2014 R1 A1 A1006823) and a grant from the Ministry of Health & Welfare, Republic of Korea (Grant No. HI14 C1090).
REFERENCES Augustin S, Millan L, Gonzalez A, Martell M, Gelabert A, Segarra A, Serres X, Esteban R, Genesca J. Detection of early portal hypertension with routine data and liver stiffness in patients with asymptomatic liver disease: A prospective study. J Hepatol 2014;60:561–569. Berzigotti A, Seijo S, Arena U, Abraldes JG, Vizzutti F, Garcia-Pagan JC, Pinzani M, Bosch J. Elastography, spleen size, and platelet count identify portal hypertension in patients with compensated cirrhosis. Gastroenterology 2013;144:102–111.e1. Bota S, Herkner H, Sporea I, Salzl P, Sirli R, Neghina AM, Peck-Radosavljevic M. Meta-analysis: ARFI elastography versus transient elastography for the evaluation of liver fibrosis. Liver Int 2013;33:1138–1147. Bureau C, Metivier S, Peron JM, Selves J, Robic MA, Gourraud PA, Rouquet O, Dupuis E, Alric L, Vinel JP. Transient elastography accurately predicts presence of significant portal hypertension in patients with chronic liver disease. Aliment Pharmacol Ther 2008;27: 1261–1268. Carrion JA, Navasa M, Bosch J, Bruguera M, Gilabert R, Forns X. Transient elastography for diagnosis of advanced fibrosis and portal hypertension in patients with hepatitis C recurrence after liver transplantation. Liver Transpl 2006;12:1791–1798. Centre for Reviews and Dissemination. Systematic reviews: CRD’s guidance for undertaking reviews in health care. York: University of York; 2009. Chon YE, Choi EH, Song KJ, Park JY, Kim do Y, Han KH, Chon CY, Ahn SH, Kim SU. Performance of transient elastography for the staging of liver fibrosis in patients with chronic hepatitis B: A meta-analysis. PloS One 2012;7:e44930. Colecchia A, Montrone L, Scaioli E, Bacchi-Reggiani ML, Colli A, Casazza G, Schiumerini R, Turco L, Di Biase AR, Mazzella G, Marzi L, Arena U, Pinzani M, Festi D. Measurement of spleen stiffness to evaluate portal hypertension and the presence of esophageal varices in patients with HCV-related cirrhosis. Gastroenterology 2012;143:646–654. Cochrane Screening and Diagnostic Tests Methods Group. Handbook for diagnostic test accuracy reviews. London, UK: The Cochrane Collaboration; 2013. de Franchis R, Baveno VIF. Expanding consensus in portal hypertension: Report of the Baveno VI Consensus Workshop: Stratifying risk and individualizing care for portal hypertension. J Hepatol 2015;63:743–752. Deeks JJ. Systematic reviews in health care: Systematic reviews of evaluations of diagnostic and screening tests. BMJ 2001;323:157–162.
10
Ultrasound in Medicine and Biology
Doebler P, Holling H. Meta-analysis of diagnostic accuracy with mada; 2012. Available at: https://cran.r-project.org/web/packages/ mada/vignettes/mada.pdf. Egger M, Davey Smith G, Schneider M, Minder C. Bias in meta-analysis detected by a simple, graphical test. BMJ 1997;315:629–634. Ferraioli G, Tinelli C, Dal Bello B, Zicchetti M, Lissandrin R, Filice G, Filice C, Above E, Barbarini G, Brunetti E, Calderon W, Di Gregorio M, Gulminetti R, Lanzarini P, Ludovisi S, Maiocchi L, Malfitano A, Michelone G, Minoli L, Mondelli M, Novati S, Patruno SF, Perretti A, Poma G, Sacchi P, Zanaboni D, Zaramella M. Performance of liver stiffness measurements by transient elastography in chronic hepatitis. World J Gastroenterol 2013; 19:49–56. Friedrich-Rust M, Ong MF, Martens S, Sarrazin C, Bojunga J, Zeuzem S, Herrmann E. Performance of transient elastography for the staging of liver fibrosis: A meta-analysis. Gastroenterology 2008;134:960–974. Hedges LV, Olkin I. Statistical methods for meta-analysis. London: Academic Press; 1985. Higgins JP, Thompson SG, Deeks JJ, Altman DG. Measuring inconsistency in meta-analyses. BMJ 2003;327:557–560. Hong WK, Kim MY, Baik SK, Shin SY, Kim JM, Kang YS, Lim YL, Kim YJ, Cho YZ, Hwang HW, Lee JH, Chae MH, Kim HA, Kang HW, Kwon SO. The usefulness of non-invasive liver stiffness measurements in predicting clinically significant portal hypertension in cirrhotic patients: Korean data. Clin Mol Hepatol 2013;19: 370–375. Jung KS, Kim SU. Clinical applications of transient elastography. Clin Mol Hepatol 2012;18:163–173. Kobayashi K, Nakao H, Nishiyama T, Lin Y, Kikuchi S, Kobayashi Y, Yamamoto T, Ishii N, Ohashi T, Satoh K, Nakade Y, Ito K, Yoneda M. Diagnostic accuracy of real-time tissue elastography for the staging of liver fibrosis: A meta-analysis. Eur Radiol 2015; 25:230–238. Laliberte E., Metacor: Meta-analysis with correlation coefficients. R package version 2011:1.0–2. Available at: https://cran.r-project. org/web/packages/metacor/metacor.pdf. Lemoine M, Katsahian S, Ziol M, Nahon P, Ganne-Carrie N, Kazemi F, Grando-Lemaire V, Trinchet JC, Beaugrand M. Liver stiffness measurement as a predictive tool of clinically significant portal hypertension in patients with compensated hepatitis C virus or alcohol-related cirrhosis. Aliment Pharmacol Ther 2008;28: 1102–1110. Llop E, Berzigotti A, Reig M, Erice E, Reverter E, Seijo S, Abraldes JG, Bruix J, Bosch J, Garcia-Pagan JC. Assessment of portal hypertension by transient elastography in patients with compensated cirrhosis and potentially resectable liver tumors. J Hepatol 2012;56: 103–108.
Volume -, Number -, 2016 Procopet B, Tantau M, Bureau C. Are there any alternative methods to hepatic venous pressure gradient in portal hypertension assessment? J Gastrointestin Liver Dis 2013;22:73–78. Reiberger T, Ferlitsch A, Payer BA, Pinter M, Homoncik M, Peck-Radosavljevic M. Non-selective beta-blockers improve the correlation of liver stiffness and portal pressure in advanced cirrhosis. J Gastroenterol 2012;47:561–568. Reitsma JB, Glas AS, Rutjes AW, Scholten RJ, Bossuyt PM, Zwinderman AH. Bivariate analysis of sensitivity and specificity produces informative summary measures in diagnostic reviews. J Clin Epidemiol 2005;58:982–990. Rutter CM, Gatsonis CA. A hierarchical regression approach to metaanalysis of diagnostic test accuracy evaluations. Stat Med 2001; 20:2865–2884. Salzl P, Reiberger T, Ferlitsch M, Payer BA, Schwengerer B, Trauner M, Peck-Radosavljevic M, Ferlitsch A. Evaluation of portal hypertension and varices by acoustic radiation force impulse imaging of the liver compared to transient elastography and AST to platelet ratio index. Ultraschall Med 2014;35:528–533. Sanchez-Conde M, Miralles P, Bellon JM, Rincon D, Ramirez M, Gutierrez I, Ripoll C, Lopez JC, Cosin J, Clemente G, Lo Iacono O, Banares R, Berenguer J. Use of transient elastography (FibroScan) for the noninvasive assessment of portal hypertension in HIV/HCV-coinfected patients. J Viral Hepat 2011;18:685–691. Schwabl P, Bota S, Salzl P, Mandorfer M, Payer B, Ferlitsch A, Stift J, Wrba F, Trauner M, Peck-Radosavljevic M, Reiberger T. New reliability criteria for transient elastography increase the number of accurate measurements for screening of cirrhosis and portal hypertension. Liver Int 2015;35:381–390. Shi KQ, Fan YC, Pan ZZ, Lin XF, Liu WY, Chen YP, Zheng MH. Transient elastography: A meta-analysis of diagnostic accuracy in evaluation of portal hypertension in chronic liver disease. Liver Int 2013;33:62–71. Stebbing J, Farouk L, Panos G, Anderson M, Jiao LR, Mandalia S, Bower M, Gazzard B, Nelson M. A meta-analysis of transient elastography for the detection of hepatic fibrosis. J Clinical Gastroenterol 2010;44:214–219. Vizzutti F, Arena U, Rega L, Pinzani M. Non invasive diagnosis of portal hypertension in cirrhotic patients. Gastroenterol Clin Biol 2008;32: 80–87. Wang JH, Chuah SK, Lu SN, Hung CH, Kuo CM, Tai WC, Chiou SS. Baseline and serial liver stiffness measurement in prediction of portal hypertension progression for patients with compensated cirrhosis. Liver Int 2014;34:1340–1348. Whiting P, Rutjes AW, Reitsma JB, Bossuyt PM, Kleijnen J. The development of QUADAS: A tool for the quality assessment of studies of diagnostic accuracy included in systematic reviews. BMC Med Res Methodol 2003;3:25.