2D shear wave liver elastography by Aixplorer to detect portal hypertension in cirrhosis: An individual patient data meta‐analysis

Liver stiffness measured with 2‐dimensional shear wave elastography by Supersonic Imagine (2DSWE‐SSI) is well‐established for fibrosis diagnostics, but non‐conclusive for portal hypertension.


| INTRODUC TI ON
Chronic liver disease costs 39 million disability-adjusted life years annually, with decompensated cirrhosis driving excess morbidity and mortality. 1 The consequence of a change from compensated to decompensated cirrhosis is most obvious in 5-year mortality rates, which increase from 1.5% in compensated cirrhosis to 88% in cirrhosis with multiple decompensating events. 2 In compensated cirrhosis, the risk of developing varices, and the risk of decompensation, increases dramatically when the hepatic venous pressure gradient (HVPG) is ≥10 or ≥12 mmHg, marking the thresholds for clinically significant portal hypertension (CSPH) and severe portal hypertension (SPH). 3,4 Liver vein catheterization with measurement of HVPG is the diagnostic gold standard for the indirect portal pressure assessment in patients with cirrhosis, but the procedure is restricted to highly specialized centres. 5 Therefore, non-invasive tools with high accuracy to rule in or rule out CSPH, SPH and varices needing treatment in patients with cirrhosis are needed.
Transient elastography (TE) with FibroScan® (Echosens, France) alone or in combination with platelet count and/or spleen diameter is widely used for portal hypertension. 6,7 TE below 20 kPa is a key feature of the Baveno VI criteria to rule out varices needing treatment in patients with compensated advanced chronic liver disease. 4 The applicability of TE is, however, limited in patients with ascites, narrow intercostal space, obesity or displaced liver, and has an upper measurement limit of 75 kPa. 8 Handling Editor: Jian Sun and upper endoscopy as reference. We used meta-analytical integration of diagnostic accuracies with optimized rule-out (sensitivity-90%) and rule-in (specificity-90%) cut-offs.
Results: Five studies from seven centres shared data on 519 patients. After exclusion, we included 328 patients. Eighty-nine (27%) were compensated and 286 (87%) had CSPH. 2DSWE-SSI < 14 kPa ruled out CSPH with a summary AUROC (sROC), sensitivity and specificity of 0.88, 91% and 37%, and correctly classified 85% of patients, with minimal between-study heterogeneity. The false negative rate was 60%, of which decompensated patients accounted for 78%. 2DSWE-SSI ≥ 32 kPa ruled in | 1437 This allows the operator to visualize the liver while performing the elastography. Furthermore, 2DSWE-SSI has no upper limit and measurements can be obtained in patients with ascites.
This makes the technique suited for patients with advanced disease, where portal hypertension is the main driver of prognosis. 8 However, the studies investigating 2DSWE-SSI to diagnose portal hypertension and its severity have resulted in heterogeneous cut-off values, and the role of potential confounders has not been investigated.
We hypothesized that 2DSWE-SSI correlates with HVPG and can be used as a diagnostic marker of portal hypertension and oesophageal varices. We therefore performed an individual patient data meta-analysis to aggregate existing evidence. Our primary aim was to evaluate 2DSWE-SSI for ruling in and ruling out clinically significant portal hypertension in patients with cirrhosis, including subgroup and sensitivity analyses (compensated cirrhosis, patients without ascites, alcohol and viral aetiology, body mass index (BMI) below 25 kg/m 2 , beta-blocker treatment and drinking status). Secondary aims were to assess the correlation between 2DSWE-SSI and the hepatic venous pressure gradient and evaluate the diagnostic accuracy of 2DSWE-SSI for severe portal hypertension and oesophageal varices.

| ME THODS
This meta-analysis was based on a multicentre collaboration with data from individual patients in previously published studies. The study protocol was sent to all authors prior to data collection. Our report follows the preferred reporting items for systematic reviews and meta-analyses extension for diagnostic test accuracy studies (PRISMA-DTA) checklist. 9 All included studies were approved by their local ethics board.

| Studies
We searched MEDLINE and Web of Sciences papers written in English that reported diagnostic studies with concurrent HVPG and 2DSWE-SSI measurements using the Supersonic Aixplorer system (Supersonic Imagine), in patients with cirrhosis. Last search update was 16 May 2019. We combined electronic searches with manual searches by scanning relevant reference lists. We invited corresponding authors to participate in this individual patient data metaanalysis. The search strategy and results can be seen in Table S1.

| Participants
We included patients for the meta-analysis if the following inclusion

| Index test and outcomes
The index test was 2DSWE-SSI, reported in kilopascals (kPa). The description of the technique can be found elsewhere. 10 The primary outcome was CSPH defined as HVPG ≥ 10 mmHg, measured by hepatic vein catheterization according to standard. 11 Secondary outcomes were SPH defined as HVPG ≥ 12 mmHg, and oesophageal varices needing treatment assessed by endoscopy at the time of 2DSWE-SSI measurement. 4

| Data extraction and quality assessment
Authors who agreed to participate shared baseline characteristics, HVPG-and 2DSWE-SSI measurements for individual study participants. Shared data were compared to published results from the individual studies by MBH and MT.
Four authors (MBH, MT, MI and AK) assessed study quality using the Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) tool, which consists of four components: patient selection, index tests, reference standard and flow and timing ( Figure 1). 12 For each component, we labelled the risk of bias and applicability as low, high or unclear. We contacted authors for clarification, if papers had unclear risk components.

| Statistics
We report summary statistics on pooled data as mean ± SD and median ± IQR, depending on distribution. For between-group differences, we report Student's t test in terms of normally distributed data, Wilcoxon rank sum test in terms of non-normally distributed data and Fisher's exact test in terms of proportions. The correlation between 2DSWE-SSI and HVPG was assessed by Pearson's correlation coefficient. We used area under the receiver operating characteristics (AUROC) curve statistics on pooled data to find the optimal 2DSWE-SSI cut-offs to rule out (90% sensitivity) and rule in (90% specificity) CSPH, SPH and oesophageal varices. Cut-offs were rounded to the nearest integer. Next, we used the rule-in and rule-out cut-offs to create 2x2 tables for the distribution of true positives, false positives, false negatives and true negatives within each included study. Next, we used a meta-analytical diagnostic accuracy model, based on bivariate logistic regressions, to combine a hierarchical summary receiver operating characteristics (sROC) model and a bivariate random-effects model of sensitivity and specificity, visualized in sROC plots. 13 We report I 2 as a measure of heterogeneity between studies. An I 2 value above 30% indicates substantial heterogeneity. We performed subgroup analyses on pooled data, owing to the low number of studies. Subgroup analyses included compensated cirrhosis, patients without ascites, alcohol and viral etiologies, and BMI below 25 kg/m 2 . We also used pooled data to compare the diagnostic accuracy of 2DSWE-SSI with other non-invasive markers (Meld-Na, liver TE, spleen TE and liver stiffness-spleen size-to-platelet ratio risk score [LSPS]) using the DeLong test. We considered a P-value below 0.05 as statistically significant. We used STATA 15.0 for all analyses, with the MIDAS module for meta-analyses (Statacorp LP).

| Studies
We identified 13 references referring to six studies which fulfilled the inclusion criteria of the meta-analysis and invited the corresponding authors to share individual patient data (see trial flow in Figure 2). Authors from five studies responded, 14-25 while authors of one study did not reply. 26 Characteristics of the five included studies can be seen in Table S1, and a list of excluded studies and cause for exclusion can be seen in Table S2. We included four single centre studies (Korea, France, Spain and Italy) and one multicentre study (Denmark, Germany and Belgium). 14-25 Most patients were enrolled prospectively, 16,18,[22][23][24][25] with only one centre including patients enrolled both prospectively and retrospectively. 17,[19][20][21] For the quality assessment (Figure 1), we considered the included studies at low risk of bias for all domains, except regarding patient selection. We labelled all studies as having a high risk of bias for patient selection, because no study restricted inclusions to compensated cirrhosis patients only (for inclusion criteria in the individual studies, see Table S1). All studies have reported experienced 2DSWE-SSI operators with no more than three different operators per study (Table S1). Since portal hypertension is a key driver of hepatic decompensation, the inclusion of decompensated patients causes a high pretest risk of CSPH.

| Patients
We received data on 519 participants of whom we included 328 for the individual patient data meta-analysis, while 191 met our exclusion criteria ( Figure 2). The majority of patients had experienced at least one decompensating event prior to inclusion (n = 237, 73%).

F I G U R E 3
Hierarchical summary receiver operating characteristic plots of sensitivity and specificity. Hierarchical summary receiver operating characteristic plots (HSROC) of sensitivity and specificity using (A) a 14 kPa cut-off for 2DSWE-SSI to rule out CSPH (derived by optimizing to 90% sensitivity in the pooled data). B, A 32 kPa cut-off to rule in clinically significant portal hypertension (derived by optimizing to 90% specificity). C, A 18 kPa cut-off to rule out severe portal hypertension. D, A 36 kPa cut-off to rule in severe portal hypertension. The summary operating point presents the summary sensitivity and specificity encircled by a 95% prediction contour of the confidence interval. Individual studies are presented with a circle and a number:

| Subgroup analyses of 2-dimensional shear wave elastography by Supersonic Imagine to diagnose clinically significant portal hypertension
Subgroup analyses showed that rule out CSPH at the cut-off 14 kPa had similar sensitivity and higher specificity for patients with compensated cirrhosis, without ascites, viral aetiology or BMI < 25 kg/ m 2 (Table 3). We compared patients receiving beta-blockers (n = 110) versus those without beta-blockers (n = 132), and found that those on beta-blockers had significantly higher 2DSWE-SSI (P = .002), higher rate of varices needing treatment (P < .001), but not significantly higher MELD-Na (P = .412). By ROC comparison, no significant difference was found between the diagnostic accuracy of 2DSWE-SSI in patients receiving beta-blockers versus those without beta-blocker treatment (P = .312 for CSPH; P = .589 for SPH). We had insufficient data on antiviral treatment to justify subgroup analyses. We compared abstinent (n = 130) and drinking patients (n = 96) and found no difference in liver stiffness (P = .480) or diagnostic accuracy of 2DSWE-SSI (P = .864 for CSPH; P = .592 for SPH).
To test the robustness of our sensitivity analyses taking into account that patients were from different studies, we performed a post-hoc multilevel logistic regression with study identifier as a random effect. This did not change the subgroup analyses (data not shown).
In a post-hoc analysis, we similarly did not find evidence of a cor-   (Table S3). In post-hoc analyses, we also applied the Baveno VI criteria to 2DSWE-SSI instead of TE, to check how many patients had platelet count >150*10 9 /L and 2DSWE-SSI < 14 kPa.

| D ISCUSS I ON
In this meta-analysis based on data from five studies and 328 individual patients, we found that 2-dimensional shear wave elastography by Supersonic Imagine had a high sensitivity to rule out clinically significant portal hypertension at cut-off 14 kPa, but TA B L E 3 Subgroup diagnostic accuracy, sensitivity and specificity of 2-dimensional real-time shear wave elastography by Supersonic Imagine to rule out (<14 kPa) and rule in (≥32 kPa) clinically significant portal hypertension in patients with cirrhosis Note: All subgroup analyses were performed on pooled data from the full cohort. We derived the 2DSWE-SSI rule out cut-offs by optimizing to 90% sensitivity and rule in cut-offs by optimizing to 90% specificity. All data are reported with 95% confidence interval in parenthesis. Between-study heterogeneity cannot explain our findings, as I 2 was low.
Currently, TE is the only elastography technique recommended by the Baveno group to identify CSPH and, in combination with platelet count, to rule out varices needing treatment. 4 The Baveno VI criteria saves approximately 30% of variceal screening endoscopies in compensated cirrhosis patients. 28 In our cohort, we did not find similar results for 2DSWE-SSI, as only 2% of patients had 2DSWE-SSI below 14 kPa and platelet count above 150*10 9 /L. However, this discrepancy may not be due to differences between TE and 2DSWE-SSI, but rather due to the high proportion of decompensated patients in our pooled data, in whom variceal screening should always be done. Existing evidence suggests that 2DSWE-SSI and TE exhibit similar diagnostic accuracies for staging fibrosis across aetiologies. 10,29 Two of the studies included in this meta-analysis similarly found no difference in diagnostic accuracy of the two techniques, albeit one study suggested that TE may have a higher failure rate than 2DSWE-SSI. 22 Between-study variation in cut-off values is considered one of the main barriers for implementation of new diagnostic techniques.
The individual patient data meta-analytical approach enabled us to derive consistent cut-offs from the pooled data, to be used in future studies. Another advantage was the ability to exclude patients who did not fulfil our inclusion and exclusion criteria, thereby limiting bias from patients with concomitant hepatic malignancy, severe liver inflammation or an extended time between 2DSWE-SSI and HVPG measurements.
In conclusion, this individual patient data meta-analysis is the largest series to date on 2-dimensional shear wave elastography by Supersonic Imagine for portal hypertension in patients with advanced liver disease. Our findings indicate that a cut-off of 14 kPa may be tested in future studies including patients with compensated advanced chronic liver disease to rule out clinically significant portal hypertension.