| |
Ultrasound-Based Transient Elastography for the Detection of Hepatic Fibrosis: Systematic Review and Meta-analysis
|
| |
| |
Clinical Gastroenteroly and Hepatology
Jayant A. Talwalkar_, David M. Kurtz, Scott J. Schoenleber, Colin P. West, Victor M. Montori
_ Department of Radiology, Mayo Clinic College of Medicine, Rochester, Minnesota
Division of Gastroenterology and Hepatology, Mayo Clinic College of Medicine, Rochester, Minnesota
Supported by National Institutes of Health grant EB001981. P.J.R. and R.L.E. and their institution hold patents related to this technology and have a potential financial interest in this research
ABSTRACT
Background & Aims: Ultrasound-based transient elastography is a promising noninvasive alternative to liver biopsy for detecting hepatic fibrosis. However, its overall test performance in various settings remains unknown. The aims of this study were to perform a systematic review and meta-analysis of diagnostic accuracy studies comparing ultrasound-based transient elastography with liver biopsy for hepatic fibrosis.
Methods: Electronic and manual bibliographic searches to identify potential studies were performed. Selection of studies was based on reported accuracy of ultrasound-based transient elastography compared with liver biopsy. Data extraction was performed independently by 2 reviewers. Meta-analysis combined the sensitivities, specificities, and likelihood ratios of individual studies. Extent and reasons for heterogeneity were assessed.
Results: Nine studies in full publication were identified. For patients with stage IV fibrosis (cirrhosis), the pooled estimates for sensitivity were 87% (95% confidence interval [CI], 84%-90%), specificity 91% (95% CI, 89%-92%), positive likelihood ratio 11.7 (95% CI, 7.9-17.1), and negative likelihood ratio 0.14 (95% CI, 0.10-0.20). Among 7 investigations reporting patients with stages II-IV fibrosis, the pooled estimates for sensitivity were 70% (95% CI, 67%-73%), specificity 84% (95% CI, 80%-88%), positive likelihood ratio 4.2 (95% CI, 2.4-7.2), and negative likelihood ratio 0.31 (95% CI, 0.23-0.43). Diagnostic threshold (or cut-off value) bias was identified as an important cause of heterogeneity for pooled results in both patient groups.
Conclusions: Ultrasound-based transient elastography appears to be a clinically useful test for detecting cirrhosis.
Discussion
There is great interest in developing and validating noninvasive methods to detect hepatic fibrosis among patients with chronic liver disease. To assess emerging data from independent studies, systematic reviews will be required to identify estimates of test performance based on all available evidence. The impact of methodologic quality with analysis to account for variation in findings between studies then can be explored by meta-analysis.24 For example, this methodology recently was applied in a systematic review examining the performance of serum marker panels for detecting moderate to severe fibrosis in patients with chronic hepatitis C.7
In this systematic review and meta-analysis, we identified and evaluated 9 primary studies from the published literature comparing ultrasound-based transient elastography with liver biopsy for detecting hepatic fibrosis. Summary estimates of sensitivity and specificity to detect cirrhosis (stage IV fibrosis) by ultrasound-based transient elastography were excellent. For identifying patients with moderate to severe (stages II-IV) fibrosis, the summary estimates of sensitivity and specificity were considered good. In particular, the magnitude of summary positive and negative likelihood ratios for detecting cirrhosis were consistent with values seen among tests considered to provide strong diagnostic evidence in clinical decision making (ie, positive likelihood ratio >10 and negative likelihood ratio <0.1). For both patient subgroups, the summary results were associated with varying degrees of statistical heterogeneity between primary studies.
Differences in study design methodology are a well-recognized cause for heterogeneity in meta-analyses of diagnostic tests.25 Although descriptions of patient characteristics and compliance with test and reference standards were uniform overall, the majority of investigations examined in this review were not explicit about methods of patient recruitment (ie, consecutive vs nonconsecutive). Previous work has shown the usefulness of examining consecutive patients or individuals evaluated within the spectrum of clinical practice to gauge test performance.26 Subtle variations in the technical performance of ultrasound-based transient elastography and liver biopsy also may be a potential cause of between-study variation.24 Although optimal performance has been defined by (1) a proportion of successful measurements to total number of acquisitions being at least 60% and (2) the presence of an interquartile range not exceeding 30% of the median value,16 we were unable to identify these criteria in the majority of reviewed studies. Based on a priori decisions before embarking on data extraction, we sought to determine if liver disease etiology or median liver biopsy size might be important contributors to reported test results from primary studies. The absence of effect from these variables, however, may have been from reduced power based on the small number of studies examined in this analysis.
The systematic assessment of methodologic quality for primary studies may identify biases that contribute to heterogeneity within summary estimates of test performance. We used the quality assessment tool QUADAS to describe the frequency of methodologic features associated with high-quality diagnostic accuracy studies in our systematic review. The majority of our studies met 10 of 14 criteria within the QUADAS checklist. Although QUADAS was developed using explicit methods,13 its reliability and validity still are evolving because recent studies are noted for low to moderate inter-rater reliability (kappa) values between 0.24 and 0.57.26, 27, 28 In our experience, however, we observed excellent degrees of agreement (kappa = 0.78) between 2 independent reviewers who assessed all 14 items for each included study. Variations in reviewer compliance with recommendations for item assessment may be an explanation for the disparate results between reported studies and our experience.28 Notably, all of the examined studies in this systematic review met 3 of 4 evidence-based study criteria (items 5, 6, and 11, but not 10) from the QUADAS checklist. However, the ultimate relevance of this distinction requires additional study to determine its significance.29
Variation in the diagnostic threshold value used for detecting a target condition can be a major cause of heterogeneity in diagnostic accuracy studies. The choice of diagnostic threshold may be explicit, identified through natural observation, or derived on the basis of disease prevalence.24 Estimates of specificity, sensitivity, and likelihood ratios, however, will differ if studies do not use a similar cut-off value for recognizing a positive test result.30, 31 This is a common observation among investigations of diagnostic testing as was seen in our systematic review. However, the high summary estimate value for sensitivity and its relative consistency across studies suggests that a negative test result may be of potential clinical use in excluding cirrhosis among patients with chronic liver disease. Similarly, a positive test result may be informative as well given the summary positive likelihood ratio exceeds 10 for detecting cirrhosis. The presence or absence of at least stage 2 fibrosis was not associated with similar likelihood ratios yet this traditionally has been more difficult to identify even with liver histology staging.
Summary ROC curves were constructed to assess the impact of diagnostic threshold bias on reported test parameters.30 Although the impact was less appreciable for the detection of cirrhosis, there was statistical heterogeneity associated with diagnostic threshold bias for the detection of stages II-IV fibrosis. Although summary ROC curves compile the diagnostic odds ratios for each individual study, there is no single measure that best captures the sensitivity and specificity with this approach. As stated previously,24, 32 the applicability of summary ROC curves remains limited based on its use of diagnostic odds ratios, which are difficult to interpret in the context of clinical practice.
Meta-regression techniques have been used for exploring factors beyond diagnostic threshold bias that may be responsible for heterogeneity in accuracy studies.33 A priori, we selected uniformity of liver disease etiology (viral or nonviral) and median liver biopsy length as potential study level covariates that might contribute additional degrees of heterogeneity. Among patients with cirrhosis, neither of these variables was found to provide additional heterogeneity to summary test results. Unfortunately, we could not assess these variables reliably for detecting stages II-IV fibrosis because only 2 investigations shared the same threshold value in this patient subgroup.
Although attempting to minimize bias against primary study selection by including non-English contributions to the literature, the prospect of publication and reporting bias affecting our results remains highly likely. In contrast to interventional studies, the extent and impact of publication bias on diagnostic accuracy studies remains less well defined.34 Although the number of published studies examining ultrasound-based transient elastography remains small given the recent emergence of this technology, it remains possible that investigations showing poor reproducibility or accuracy have not been published because of negative results. In addition, reporting bias remains a serious problem when trying to identify all the data necessary to conduct a methodologically rigorous systematic review.35 In this review, a number of studies did not report the proportion of liver biopsies that may not have been suitable for assessment of fibrosis stage. Moreover, the exclusion of ultrasound-based transient elastography data from patients with high body mass index values based on technical failure36 also was noted. The removal of these uninterpretable results would tend to overestimate the diagnostic performance of index tests compared with its reference standard.
In conducting this systematic review, we recognize a number of limitations that require further discussion. Lack of uniform reporting on liver biopsy quality as well as failures of both techniques limit the ability to assess diagnostic performance fully. Our analyses observed higher performance rates for cirrhosis but the relative difference in performance vs liver biopsy is unknown. However, the relative difference in performance between advanced and nonadvanced fibrosis could be similar when adjusted on the spectrum bias and on the quality of biopsy. Recent advances in methodology using sensitivity or standardization analyses,37 which can highlight differences in stage prevalence, which influence test performance, should be used in future analyses as more studies are performed.
Because of the spectrum bias, it is automatic that the diagnostic value of any marker will be higher for cirrhosis than for advanced fibrosis. This always has been observed in comparing stage 4 vs stages 0-3 as compared with stages II-IV vs 0-1 on explant analysis.38 In addition, the absence of a large control group of healthy individuals with normative data from elastography precluded us from comparing this group with individuals with hepatic disease. Strong clinical end points such as morbidity and mortality have been used for validated biomarkers39 and also should be recommended to assess the diagnostic performance of ultrasound-based transient elastography.
This systematic review describes the diagnostic test performance of ultrasound-based transient elastography compared with liver biopsy for detection of hepatic fibrosis in patients with chronic liver disease. This noninvasive method appears to be clinically useful in assessing the presence or, more specifically, the absence of cirrhosis. Methodologic differences and varying diagnostic thresholds among published investigations suggest that additional studies are needed to further establish the precision of this emerging technology. Continued advances in data analysis with the subsequent evaluation of patients in community-based settings also will be required to achieve this goal.40
BACKGROUND
Cirrhosis and its disease-related complications are responsible for more than 40,000 deaths annually in the United States.1 In addition, health care resource utilization for survivors is expected to increase based on trends in aging, obesity, and the emergence of overt manifestations related to chronic hepatitis C infection.2, 3 These clinical outcomes are related directly to the development of progressive hepatic fibrosis. To date, liver biopsy has been regarded as the gold standard for detecting hepatic fibrosis. However, there continues to be evidence that sampling error (up to 25%-40%) remains a problem in accurate fibrosis staging for individual patients.4, 5, 6 In addition, the extent of variation between histopathologist interpretation for a particular stage may be as high as 20%.4 Furthermore, the ability to procure an optimal specimen for assessment (biopsy length ≥ 25 mm with ≥ 11 complete portal tracts) is not guaranteed with either percutaneous or transjugular approaches.5 Coupled with high costs and reduced patient acceptance, there remains a need for developing noninvasive detection methods for hepatic fibrosis in addition to liver biopsy.
To date, the noninvasive diagnostic tests available from clinical practice are not sensitive or specific enough for detecting occult fibrosis. In terms of novel approaches, a recent systematic review of serum fibrosis markers was noted for only moderate performance in detecting moderate to severe fibrosis among patients with chronic hepatitis C after pooling results of high-quality studies.7 A number of serum markers representing the process of hepatic fibrosis have been studied to date. However, the frequency of indeterminate results from individual markers or panels ranges between 25% and 40% with the need for subsequent liver biopsy for diagnosis.7 Emerging data have shown that a sequential algorithm-type approach to improve the posttest likelihood of detecting advanced fibrosis may be more effective than application of a single test alone.8
Recently, the method of ultrasound-based transient elastography was reported to identify patients with hepatic fibrosis.9 Transient elastography is based on the principle of Hooke's law, which characterizes a material's strain response to external stress. By using an ultrasound transducer probe mounted on the axis of a vibrator, the transmission of low-frequency vibrations from the right intercostal space creates an elastic shear wave that propagates into the liver. A pulse-echo ultrasound acquisition then is used to detect the velocity of wave propagation. This velocity is proportional to tissue stiffness, with faster wave progression occurring through stiffer material. Measurement of liver stiffness then is performed and measured in kilopascals. Ex vivo investigations have confirmed the association between tissue elasticity and degree of fibrosis.10
Since the initial report in 2003,9 a number of investigations examining ultrasound-based transient elastography in patients with chronic liver disease have been published. With the upcoming prospect of this technique's expansion into the United States, we sought to conduct a systematic review and meta-analysis to characterize the diagnostic performance of ultrasound-based transient elastography as compared with the reference standard of liver biopsy for the detection of hepatic fibrosis.
Methods
Literature Search
A computer-aided literature search of PubMed (MEDLINE), EMBASE, the Cochrane Library, Database of Abstracts of Reviews of Effects, Web of Science, SCOPUS, American College of Physicians Journal Club, and Google Scholar databases was conducted from database inception through January 29, 2007. Initial search strategy using free-text words ("hepatic fibrosis AND elastography") was conducted. We also used a sensitive and precise search strategy in the PubMed database for locating any existing systematic reviews on ultrasound-based transient elastography (to identify additional studies missed by our search), of which none were identified.11 Manual searching of reference lists from primary studies then was performed to locate any potential studies missed by electronic search strategies. Abstracts and proceedings from the American Association for the Study of Liver Disease, European Association for the Study of the Liver, and Digestive Disease Week annual meetings between 2003 and 2006 also were reviewed. This time period is noted for an increased submission and acceptance rate for investigations examining ultrasound-based transient elastography to detect hepatic fibrosis. Consultation with experts in the field also was performed to identify additional published and unpublished primary studies.
Study Selection
Two independent reviewers (D.M.K. and S.J.S.) read all abstracts of candidate articles and retrieved the full text of published manuscripts that could not be narrowed down based on title and abstract alone. These articles were read and checked for inclusion criteria independently, with disagreements resolved through consensus with a third reviewer (J.A.T.). Primary studies that reported data required for meta-analysis were identified and included.
Study Inclusion/Exclusion Criteria
Inclusion criteria for primary studies required the following features: (1) detailed description of human subjects under study, (2) description of ultrasound-based transient elastography as the index test, and (3) description of liver biopsy as the reference standard. The inclusion of non-English language studies was allowed. Studies in which transient elastography was compared with other noninvasive methods of hepatic fibrosis (ie, serum markers) were allowed if discrete information on transient elastography alone could be extracted from the data. Exclusion criteria were limited to duplicate publications of a primary study that contained all or some of the original data. In this situation, the updated manuscript was to be chosen, assuming the relevant data for meta-analysis were available.
Quality Assessment of Primary Studies
Each of the studies meeting inclusion criteria was analyzed in duplicate by independent reviewers (D.M.K. and S.J.S.) for quality using the Quality Assessment of Diagnostic Accuracy Studies (QUADAS) checklist12, 13 (Table 1). This tool is a 14-item instrument that allows for the identification of important design elements in diagnostic accuracy studies such as patient spectrum, the presence or absence of observer blinding and verification bias, handling of indeterminate results, and reporting of patient loss to follow-up evaluation. Where there were discrepancies between the reviewers, a consensus reviewer (J.A.T.) resolved the differences.
Table 1. QUADAS Assessment Items
Was the spectrum of patients representative of the patients who will receive the test in practice? (Generalizability item)
Were selection criteria clearly described? (Clarity item)
Is the reference standard likely to correctly classify the target condition? (Validity item)
Is the time between reference standard and index test short enough to be reasonably sure that the target condition did not change between the two tests? (Validity item)
Did the whole sample or a random selection of the sample receive verification using a reference standard of diagnosis? (Validity item)
Did patients receive the same reference standard regardless of the index test result? (Validity item)
Was the reference standard independent of the index test (ie, the index test did not form part of the reference standard)? (Validity item)
Was the execution of the index test described in sufficient detail to permit replication of the test? (Clarity item)
Was the execution of the reference standard described in sufficient detail to permit its replication? (Clarity item)
Were the index test results interpreted without knowledge of the results of the reference standard? (Validity item)
Were the reference standard results interpreted without knowledge of the results of the index test? (Validity item)
Were the same clinical data available when test results were interpreted as would be available when the test is used in practice? (Validity item)
Were uninterpretable/intermediate test results reported? (Clarity item)
Were withdrawals from the study explained? (Clarity item)
Data Extraction
Two reviewers (D.M.K. and S.J.S.) independently extracted the required information from primary studies. Prespecified data elements for collection included patient age, sex, and body mass index; underlying chronic liver disease etiology, details of test and reference standards, histologic fibrosis stage, average liver biopsy size, average number of portal tracts per liver biopsy, and duration of time between biopsy and ultrasound-based transient elastography. Other variables that were sought included diagnostic threshold (or cut-off) values used for detecting hepatic fibrosis, test performance characteristics, reasons for participant exclusion, and methods for handling indeterminate or missing data.
Data Analysis/Synthesis
The primary outcome for analysis was diagnostic test performance of ultrasound-based transient elastography for the detection of cirrhosis (stage IV) fibrosis vs no cirrhosis (stages 0-III fibrosis) compared with the reference standard of liver biopsy. Assessing the diagnostic test performance of ultrasound-based transient elastography for detecting stages II-IV fibrosis vs stage 0-I fibrosis was analyzed as a secondary outcome. Sensitivity, specificity, likelihood ratios, and diagnostic odds ratios with 95% confidence interval (CI) values were reported for individual studies. The diagnostic odds ratio is defined as the odds of having a positive test result in patients with disease compared with the odds of a positive test result in patients without disease. If zero cells were identified in calculating likelihood ratios, then a value of 0.5 was added to all cells to facilitate analysis.
The heterogeneity of all diagnostic test parameters was evaluated initially by graphic examination of Forrest plots for each parameter. Statistical assessment then was performed using a chi-square test of homogeneity and the inconsistency index (I2). The I2 statistic is defined as the percentage of variability as a result of heterogeneity beyond that from chance, with values greater than 50% representing the possibility for substantial heterogeneity. Pooled summary statistics for sensitivities, specificities, likelihood ratios, and diagnostic odds ratios of the individual studies then were reported. Analyses were conducted to include diagnostic threshold values corresponding to the maximum sensitivity and specificity values from receiver operating characteristic (ROC) curve analysis. Other diagnostic threshold values reported to maximize sensitivity or specificity were not considered. Because of a priori assumptions about the likelihood for heterogeneity between primary studies, the random-effects model of DerSimonian and Laird14 was used for pooled analyses.
If significant heterogeneity from diagnostic threshold effect was observed based on ROC plot analysis and Spearman rank correlation testing of sensitivity and specificity values for primary studies, then a summary receiver operating curve was constructed to express test parameter results as diagnostic odds ratios. After the assessment for presence of diagnostic threshold bias, the use of meta-regression techniques to explore the possibility that homogeneity of liver disease etiology (viral vs mixed) and mean liver biopsy size were additional causes of between-study heterogeneity was intended. Analyses were performed using the statistical software Meta-DiSc (Ramn y Cajal Hospital, Madrid, Spain) (version 1.1.1)
Results
The abstracts and titles of 120 primary studies were identified for initial review based on described search strategies. Full-text review was required for 78 publications to determine study eligibility. Subsequently, a total of 18 articles were selected with excellent agreement between the 2 reviewers (kappa = 0.89).
Quality assessment revealed that all 18 studies fulfilled greater than 10 of 14 QUADAS items describing methodologic quality. All reported studies were performed in tertiary referral centers. Specific criteria such as verification of diagnosis in all studies with liver biopsy and blinding to ultrasound-based transient elastography results when interpreting liver biopsies was reported. However, none of the 18 studies described interpreting results of ultrasound-based transient elastography while blinded to liver biopsy results. The results of 9 studies providing data for 2 _ 2 table construction to analyze diagnostic test performance for patients with stage IV fibrosis were available for meta-analysis.15, 16, 17, 18, 19, 20, 21, 22, 23 In addition, a total of 7 studies were able to provide data to analyze diagnostic test performance for patients with stages II-IV fibrosis.
The demographic and clinical features of patients in the retained studies are listed in Table 2. The median sample size for all studies was 193 patients (range, 101-1257). The median age of patients was 50 years (range, 42-60 y), with men comprising 52% of reported individuals (range, 25%-67%). The median body mass index reported in 6 studies was 25 kg/m2 (range, 22.4-25 kg/m2). Chronic hepatitis C with or without viral co-infection (hepatitis B or human immunodeficiency virus) was the major disease etiology of interest in 8 studies. The median proportion of individuals with stage IV fibrosis was 15% (range, 10%-33%). Patients with stages II-IV fibrosis accounted for a median proportion of 56% (range, 31%-76%) in 7 studies reporting these data. The median length of liver biopsies reported in 6 studies was 17 mm (range, 16.5-20 mm). Information on average portal tract number per liver biopsy rarely was described. Diagnostic threshold values to define a positive test varied between studies for stage IV fibrosis (range, 11.7-17.6) and stages II-IV fibrosis (range, 4.0-8.8).
Summary Estimates of Primary Studies
For studies examining the presence or absence of stage IV fibrosis (cirrhosis) on liver biopsy, there was no qualitative evidence for heterogeneity of reported sensitivities and specificities by Forrest plot inspection. However, there was evidence for borderline statistical heterogeneity for sensitivities ranging between 78% and 100% (P < .05; I2 statistic, 49%) (Figure 1). In contrast, a greater degree of heterogeneity was observed between specificity values ranging from 82% to 96% (P < .001; I2 statistic, 82%).
Figure 1. Forrest plots and meta-analyses of studies evaluating the (A) sensitivity and (B) specificity of ultrasound-based transient elastography compared with liver biopsy for the detection of stage IV hepatic fibrosis in patients with chronic liver disease.
The pooled sensitivity estimate for the 9 studies was 87% (95% CI, 84%-90%), whereas the pooled specificity estimate was 91% (95% CI, 89%-92%). A pooled positive likelihood ratio of 11.7 (95% CI, 7.9-17.1) with evidence of statistical heterogeneity (P < .001; I2 statistic, 76%) was observed. A pooled negative likelihood ratio was 0.14 (95% CI, 0.10-0.20) with no statistical heterogeneity observed (P = .19; I2 statistic, 28%). The summary diagnostic odds ratio for all studies was 87 (95% CI, 60.0-127.9) without evidence of statistical heterogeneity (P = .74; I2 statistic, 0%).
Among the 7 primary studies assessing patients with stages II-IV hepatic fibrosis, both Forrest plot inspection and statistical testing confirmed the presence of heterogeneity for sensitivity values ranging between 53% and 93% (P < .001; I2 statistic, 92%) (Figure 2). Statistical heterogeneity also was observed for reported specificity values between 33% and 90% (P < .001; I2 statistic, 76%). The pooled sensitivity estimate was 70% (95% CI, 67%-73%), whereas the pooled specificity estimate was 84% (95% CI, 80%-88%). A pooled positive likelihood ratio of 4.2 (95% CI, 2.4-7.2) with evidence for statistical heterogeneity (P < .001; I2 statistic, 85%) was observed. A pooled negative likelihood ratio was 0.31 (95% CI, 0.22-0.43) with statistical heterogeneity also noted (P < .001; I2 statistic, 82%). The summary diagnostic odds ratio was 15 (95% CI, 9.8-24.6) without evidence for statistical heterogeneity (P = .12; I2 statistic, 40%).
Figure 2. Forrest plots and meta-analysis of studies evaluating the (A) sensitivity and (B) specificity of ultrasound-based transient elastography compared with liver biopsy for the detection of stages II-IV hepatic fibrosis in patients with chronic liver disease.
Diagnostic Threshold Bias and Meta-Regression Assessment
To assess for diagnostic threshold bias as a cause for heterogeneity in test performance by ultrasound-based transient elastography, an ROC plot of sensitivity vs 1 - specificity was performed. Among the 9 primary studies reporting data for detecting stage IV hepatic fibrosis, there was some evidence for diagnostic threshold bias as a major cause of statistical heterogeneity (Spearman rank correlation coefficient = 0.567; P = .112). Based on the presence of various diagnostic threshold (cut-point) levels used in reported investigations, a subgroup of 6 studies using a value of 12.5 kPa or greater to define test positivity was analyzed further. Re-examination of these studies, however, did not identify any additional diagnostic threshold effect. Meta-regression analysis did not identify liver disease etiology (viral vs mixed) or median liver biopsy size of 2 cm or more as other variables that influenced the presence of between-study heterogeneity. Summary ROC curve construction also suggested the possibility of diagnostic threshold effect among the primary studies (Figure 3).
Figure 3. Summary ROC curve for all 9 studies examining ultrasound-based transient elastography compared with liver biopsy for the detection of stage IV hepatic fibrosis.
Among patients with stages II-IV fibrosis, the ROC plot of sensitivity vs 1 -specificity revealed statistical evidence for diagnostic threshold bias as a major cause of heterogeneity between the 6 reported studies (Spearman rank correlation coefficient = 0.821; P = .023). Given the reduced number of studies examining stages II-IV fibrosis, subgroup and meta-regression analyses could not be performed reliably to explore additional causes of heterogeneity. Summary ROC curve analysis also confirmed the presence of statistical heterogeneity from diagnostic threshold effect for the detection of stages II-IV fibrosis (Figure 4).
Figure 4. Summary ROC curve for all 7 studies examining ultrasound-based transient elastography compared with liver biopsy for the detection of stages II-IV hepatic fibrosis.
|
|
| |
| |
|
|
|
