Transient Elastography for Significant Liver Fibrosis and Cirrhosis in Chronic Hepatitis B: A Meta-Analysis

Background The hepatitis B virus infection is a global health issue and the stage of liver fibrosis affects the prognosis in patients with chronic hepatitis B (CHB). We performed the meta-analysis describing diagnostic accuracy of transient elastography (TE) for predicting CHB-related fibrosis. Methods We performed an adequate literature search to identify studies that assessed the diagnostic accuracy of TE in CHB patients using biopsy as reference standard. Hierarchical summary receiver-operating curves model and the bivariate mixed-effects binary regression model were applied to generate summary receiver-operating characteristic curves and pooled estimates of sensitivity and specificity. Results The area under the summary receiver-operating curve for significant fibrosis and cirrhosis was 0.86 (95% confidence interval (CI): 0.83–0.89) and 0.92 (95% CI: 0.90–0.94), respectively. The sensitivity, specificity, and diagnostic odds ratio of TE for significant fibrosis were 0.78 (95% CI: 0.73–0.81, p < 0.01; I2 = 85.59%), 0.81 (95% CI: 0.77–0.84, p < 0.01; I2 = 88.20%), and 14.44 (95% CI: 10.80–19.31, p < 0.01; I2 = 100%) and for cirrhosis were 0.84 (95% CI: 0.80–0.88, p < 0.01; I2 = 76.67%), 0.87 (95% CI: 0.84–0.90, p < 0.01; I2 = 90.89%), and 36.63 (95% CI: 25.38–52.87, p < 0.01; I2 = 100%), respectively. The optimal cut-off values of TE were 7.25 kPa for diagnosing significant fibrosis and 12.4 kPa for diagnosing cirrhosis, respectively. Conclusion TE is of great value in the detection of patients with CHB-related cirrhosis but has a suboptimal accuracy in the detection of significant fibrosis.


Introduction
Chronic hepatitis B virus infection continues to be a major public health issue worldwide with the prevalence of 3.61% [1]. As well known, liver fibrosis, one of the main prognostic factors in chronic hepatitis B (CHB), was associated with the risk of developing cirrhosis and cirrhosis-related complications [2,3]. Therefore, liver fibrosis stage plays one of the most important roles in diagnostic and prognostic assessments in patients with CHB.
Liver biopsy (LB), as invasive in nature with related risks, is the gold standard for fibrosis assessment. However, LB is associated with obvious patient discomfort and risk of complications ranging from pain to more serious events with hospitalization rate of 1.4-3.2% [4] and mortality varying from 0.0088 to 0.3% [5]. Besides, LB provides only a quite small part of the organ, and thus there is a risk that the small part might not be representative for the live fibrosis in the whole liver [6].
Noninvasive methods of assessing fibrosis and cirrhosis were urgently needed, and serologic tests and novel imaging techniques were recently developed [7,8]. Most of these studied focused on whether noninvasive methods can accurately detect minimal (F0-1), significant (≥F2), or advanced (≥F3-4) fibrosis based on the METAVIR score [9]. Transient elastography (TE), also known as FibroScan, was 2 Canadian Journal of Gastroenterology and Hepatology a device and a well-validated method with advantages of a short procedure time (<5 min), immediate results, and the ability to perform the test at the bedside or in an outpatient clinic [10]. Compared with blood tests, TE has a similar performance to predict significant fibrosis (SF) and higher accuracy to identify cirrhosis [11]. Measurement of liver fibrosis without biopsy is very tempting. In spite of the fact that recommendations suggested that noninvasive tests were still not ready to replace LB [12,13], TE has become widely present in clinical practice. The accuracy of TE for detection of fibrosis has been assessed extensively in a variety of liver diseases [14][15][16][17]. However, it was reported that the presence of an IQR/M > 30% and liver stiffness median ≥7.1 kPa lead to a lower accuracy determined by the area under receiveroperating curve (AUROC) and these cases were considered "poorly reliable" [18]. Another study also indicated that there was a significant discrepancy in up to 20% of cases cirrhosis between different TE devices [19].
In the study, we performed an independent meta-analysis of the diagnostic accuracy of TE for predicting significant liver fibrosis (F2-4 versus F0-1) and cirrhosis (F4 versus F0-3) in CHB patients.

Literature Search Strategy.
PubMed, Web of Science, and EMBASE database were searched to October 10, 2016, as well as Wanfang database and China National Knowledge Infrastructure. The search strategy was "FibroScan or transient elastography" in combination with "liver fibrosis assessment," "significant fibrosis or cirrhosis or advanced liver fibrosis," and "liver stiffness measurement." All eligible studies were retrieved and their reference lists were checked for additional relevant publications.

Inclusion
Criteria. All diagnostic cross-sectional studies, cohort studies, and randomized studies that compared TE accuracy with biopsy in diagnosis fibrosis grade were eligible for inclusion. Studies that met all the following criteria were included: (i) studies which reported that all patients had undergone biopsy and TE; (ii) having enough data to create 2 × 2 table of test performance (with numbers of true and false positives and negatives); and (iii) studies which reported the method of definition of the fibrosis grade.

Exclusion Criteria.
The exclusion criteria were as follows: (i) the patients belonging to the pediatric population, hepatitis C/hepatitis B virus coinfected patients, mixed chronic liver disease patients (but not CHB and nonalcoholic fatty liver disease), and liver/kidney transplant patients; (ii) studies that were clearly extensions of previously published cohorts; and (iii) studies unable to obtain sufficient data for statistical analysis.

Methodological
Assessment. Methodological quality was assessed by the Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) tool. QUADAS-2 was designed to assess the internal and external validity. Any differences between two authors were resolved with discussion between the two review authors and the third author was final arbiter.

Data Extraction and Management.
As for each study, the following information was extracted: year of publication, study design, sample size, presence of HIV coinfection, the QUADAS-2 methodological items, prevalence of each fibrosis stage on biopsy, along with total prevalence of SF and cirrhosis, interval between biopsy and TE, size of biopsy sample, type of scoring system used for histology (METAVIR versus other), and AUROC. Two authors performed the data extraction independently. Disagreement was resolved with discussion between the two review authors, with a third author as final arbiter.
2.6. Statistical Analysis and Data Synthesis. Initial analysis was performed with the Review Manager (RevMan) 5.0. Stata 12.0 was used for meta-analysis of diagnostic accuracy studies, to compute the pooled sensitivity and specificity and to plot the summary receiver-operating characteristics curve (SROC) with summary point and corresponding 95% confidence interval (CI). Regression analysis was performed by Stata 12.0, with each time point providing another covariate to verify the influence of the chosen covariate on the accuracy estimates. We used hierarchical SROC model and the bivariate random efforts model to produce SROC and pooled estimates of sensitivity and specificity. We performed Fagan test to detect clinical significant by Stata 12.0. Heterogeneity was assessed with the inconsistency index ( 2 ) and 2 values over 50% indicated substantial heterogeneity. Heterogeneity from threshold effect was explored by meta-disc 1.4.

Characteristics of Included
As presented in Figure 2, the results of methodological quality assessment based on the QUADAS-2 scale were depicted for all of the 44 eligible studies. The majority of the methodological concern lies within the index test, because TE in ten studies interpreted with knowledge of the results of the biopsy [     Canadian Journal of Gastroenterology and Hepatology  was conducted with assistance by a time-motion ultrasound image [40]. Another possible issue was addressed in patient selection that participants might be enrolled consecutively with confirmed diagnosis in three studies [31,50,55]. Both of these concerns might be located in heterogeneity and sensitivity analyses.

Diagnosis of SF.
We included 35 studies ( = 6,202) in the analysis for SF (F2-F4) [15-23, 25-27, 29-35, 37-40, 43, 56-59]. Summary representation of the overall analysis was presented in Figure 3 Figure 1(C)). However, it must be carefully considered as they were not pooled from studies with identical TE threshold. Overall, there was heterogeneity as graphically illustrated on the forest plot in Supplementary Figure 1. The cut-off value for SF (F2-4) ranged from 5.2 to 10.3 kPa with a mean value of 8.6 kPa and a median of 7.25 kPa.
As shown in Figure 3(b) and Table 2, in the analysis of LB-related factors with an impact on accuracy, there was no significant difference (joint = 0.47 for classification criteria; joint = 0.29 for interval time; joint = 0.77 for average sample size). 26 studies conducted in Asian presented a better both pooled sensitivity (0.78, 95% CI: 0.73-0.82) and specificity (0.83, 95% CI: 0.79-0.87) than in Caucasian (joint = 0.03). As presented in Figure 3(c), it was indicated that posttest probability of LR+ increased to 86% and LR− decreased to 29% after TE was performed based on Fagan test.

Diagnosis of Cirrhosis.
41 studies were included in the cirrhotic analysis with a total of 7,205 patients, as four studies did not have any cases of liver cirrhosis (METAVIR F4) [21,24,35,48]. The overall prevalence of METAVIR F4 and the AUROCs in the included studies ranged from 5% to 69.2% and from 0.80 to 0.98 (Table 1), respectively.
Summary representation of the overall analysis was shown in Figure 4(a). The area under the SROC for liver cirrhosis was 0.92 (95% CI: 0.90-0.94). Sensitivity ranged from 49% to 100%, much more widely than specificity which ranged from 62% to 99% (Supplementary Figure 2).  Figure 2(C)). Again, these measures must be carefully considered without identical TE thresholds. The cut-off value for cirrhosis ranged from 9 kPa to 18.2 kPa with both a mean value and a median of 12.4 kPa.

Discussion
TE can provide a reliable detection of liver fibrosis in patients with CHB and thus has been recommended by the American Association for the Study of Liver Diseases (AASLD) and European Association for the Study of the Liver (EASL) [60,61]. This meta-analysis was conducted in a total of 7,808 CHB patients to summarize the diagnostic accuracy of TE for CHB-related SF, with optimal statistical method SROC. In addition, regression analysis was carried out to further explore sources of heterogeneity. In our study, TE performed well in both SF (F2-4) and cirrhosis (F4) with pooled sensitivity of 78% and 84%, summary specificity of 81% and 87%, DOR of 14.44 and 36.63, LR+ of 4.01 and 6.66, LR− of 0.28 and 0.18, respectively. Study by Li et al. [62] with hierarchical SROC model was also performed in CHB patients, with summary sensitivity and specificity for SF (F2-4) and cirrhosis (F4) of 80% and 86%, 82%, and 88%, however, without DOR, LR+ and LR−. Interestingly, the pooled specificity for diagnosis SF (F2-4) and cirrhosis (F4) in both studies were higher than Sensitivity (95% CI) * * * length Yes * * * interval Yes * * * length Yes * * * interval Yes * * * patients Yes * * * metavir Yes * * * metavir Yes * p < 0.05, * * p < 0.01, * * * p < 0.001 * p < 0.05, * * p < 0.01, * * * p < 0.001   summary sensitivity, which suggested that the currently cutoff values of TE performed better in excluding diseases rather than confirming diseases. Furthermore, the areas under the SROC were 0.86 for SF (F2-4) and 0.92 for cirrhosis (F4), respectively, which indicated that TE was performed well in staging fibrosis in CHB patients. In addition, TE performed better for cirrhosis than SF with a higher value of AUC, sensitivity, specificity, DOR, LR+, and a lower value of LR−. Although the diagnostic accuracy was higher for cirrhosis, TE could also increase the diagnostic accuracy for SF based on Fagan test with increased LR+ and decreased LR−. The higher TE values were used to confirm diagnosis, while the lower one was used to exclude the false positive diagnosis. However, if the TE value located between the values for rule in and rule out, biopsy was then recommended. Based on the descriptive statistics of enrolled studies, the cut-off values for diagnosing SF (F2-4) and cirrhosis (F4) ranged from 5.2 to 10.3 kPa and 9 to 18.2 kPa, respectively. The optimal cut-off values of TE in CHB patients in our study were 7.25 kPa for SF (F2-4) and 12.4 kPa for cirrhosis (F4). In the previous meta-analysis by Li et al., the weighted mean cut-off values of TE were comparable with 7.2 kPa for SF (F2-4) and 12.2 kPa for cirrhosis (F4) [62]. However, since there was no optimal statistical method to pool different cutoff values in individual studies, the optimal cut-off values in our meta-analysis were simply summarized as median, which could eliminate the impact resulting from the maximum and minimum values that was better than the mean value in previous study [62].
Elevated ALT levels might affect the predictive accuracy of TE [16,24,45,50,55,56]; however, the study by Cardoso et al. reported that the use of TE cut-off values adjusted to ALT level did not improve the performance of liver stiffness in CHB patients [49]. Although elevated ALT might be the most important confounder on liver stiffness measurement, the synthesis analysis of ALT elevation could not be conducted due to insufficient data. Therefore, it would be beneficial if more clinical studies focused on the correlation between ALT elevation and TE in CHB patients.
One of the main limitations in this meta-analysis was the significant heterogeneity of the included studies. Spearman correlation coefficient for SF and cirrhosis were 0.055 ( = 0.755) and 0.057 ( = 0.723), and no threshold effect was presented. Therefore, regression analysis was carried out. Besides, TE value could be applied as diagnosis criteria for both SF and cirrhosis in Asian. However, for Caucasian, it was noted that TE was valid to diagnosis of cirrhosis, while it was less precise for SF. Unfortunately, the regression analysis was not conducted owing to the small size of HIV-and non-HIVcoinfected patients. It should be noted that the overlapped cut-off values from included studies might also result in the heterogeneity.
In conclusion, TE is of great value for detection CHBrelated cirrhosis, however, with a suboptimal performance in detection of SF. Further studies should focus on the TE cut-off value and the effect of ALT elevation in patients with CHB.