Transient Elastography (FibroScan) for the diagnosis of liver fibrosis: a systematic review and meta-analysis

2009 Agencia de Evaluación de TecnologIas Sanitarias (A

BACKGROUND
Transient elastography (TE) (FibroScan) is a new non-invasive diagnostic method which uses low frequency vibration and ultrasound to assess the degree of liver fibrosis. This technique could benefit a great number of patients as an alternative to hepatic biopsy.
OBJECTIVE
To study the scientific evidence regarding effectiveness and safety of TE in the diagnosis of liver fibrosis as compared to hepatic biopsy, and assess the foreseeable impact of using this test.
METHODS
Systematic review of scientific literature and meta-analysis. Studies were identified using the following databases: MEDLINE (PubMed), EMBASE, The Cochrane Library, Database of Abstracts of Reviews of Effects, Economic Evaluation Database and Health Technology Assessment Database of the CRD of the University of York. In addition, the Indice Médico Espanol, HSRproj, Web of Science, SciSearch, and other databases available through DATASTAR such as PASCAL, EMCare, Elsevier BIOBASE, CAB Global Health or MEDITEC, as well as databases including abstracts of Congresses were also consulted. Web sites from Health Technology Assessment Agencies, both National and International, were also consulted through INAHTA, as well as the main scientific Societies and journals related to the technology of interest. Lastly, a manual search of the bibliographic references of the chosen articles was also carried out. The selection of terms for the search was a combination of free terms and controlled terms in the databases which have a thesaurus available.
Studies were selected up to March 2008, without any language or sample size restriction. All study designs evaluating diagnostic effectiveness and safety of TE in patients diagnosed with any hepatic pathology where TE was used as a procedure to assess liver fibrosis and a hepatic biopsy was used as a reference test were included. The following outcome measures were noted: Indicators of diagnostic effectiveness in terms of True Positive, True Negative, False Positive and False Negative values. At the same time the Sensitivity (Se), Specificity (Sp), Positive and Negative Predictive Values (PPV, NPV), positive and negative Likelihood Ratios (LR+, LR-), Diagnostic Accuracy (Ex), Diagnostic Odds Ratio (DOR) with their corresponding 95% confidence intervals (95% CI) were extracted. Possible complications or adverse effects caused by using TE were considered as test safety indicators.
The exclusion criteria included duplicated studies or studies outdated by later studies from the same institution, studies with insufficient data on the diagnostic effectiveness of TE and case reports.
Identification and selection of studies, as well as data extraction, were carried out exhaustively and independently by two reviewers. Disagreements were resolved by consensus. The quality of studies was evaluated using the QUADAS questionnaire.
Firstly, a qualitative analysis of the data was carried out, followed by a metaanalysis using the random effects model. Contingency tables for the studies were estimated using the corresponding diagnostic parameters (Se, Sp, PPV, NPV) for each stage of liver fibrosis. When different cut-offs were available for the same stage, the meta-analysis was done with the data corresponding to the optimal cut-off value. The fibrosis stage was established, according to the METAVIR scale (or, failing this, using the corresponding equivalence), such that mild fibrosis (F?1); significant fibrosis (F?2), advanced fibrosis (F?3) and cirrhosis (F4). Graphical representation of the pairs (Se, 1-Sp) in the ROC plane was used to assess the presence of a threshold effect, and the Spearman correlation coefficient (rho) for statistical confirmation. Also, heterogeneity was assessed using forest plots of Se, Sp, LRs and DOR for the different levels of fibrosis, together with the Chi-squared test for Se and Sp and Cochrans Q test for LRs. The inconsistency index I2 was also studied. When a threshold effect was evident, the results were expressed as the area under the ROC curve (AUCSROC) and the Q* point; when such effect was absent, the pooled indices for Se, Sp, LRs and DOR with their 95% CI were also calculated. In cases where outliers were detected, the analysis was repeated excluding these studies. A subgroup analysis was also carried out selecting studies with chronic hepatitis C (CHC) patients to determine whether there were differences in the diagnostic performance of TE in this pathology as compared to the whole of hepatic pathologies. Lastly, a meta-regression was performed to assess whether the publication format (original article or abstract) had any influence on the overall results, studying the coefficient and its significance, and using the relative DOR and its 95% CI. The patient was considered the analysis unit. Version 1.4 of MetaDiSc was used for the quantitative analysis.
RESULTS
The review includes a total of twenty six publications (23 articles and 3 abstracts), of the 372 references initially identified, all of them being case series. According to the QUADAS questionnaire, the quality of the studies was good. The review includes 3,365 patients studied between November 1995 and September 2007. Only one study was carried out with children, with a mean age of 9.5, among the rest, the age ranged between 16 and 86 years old. The most frequent cause of liver fibrosis was CHC with or without co-infection with HIV or HBV, although there were also other causes such as sclerosing cholangitis, autoimmune hepatitis, haemochromatosis or Wilsons disease.
With regards to TE, the average performance time was less than 5 minutes. The number of valid measures which allow the test to be considered appropriate was between 5 and 20.The success rate of TE was over 60%. Patients were excluded in 13 studies due to a failure in the TE measurements. Only 3 studies provided information regarding inter/intra observer agreement when carrying out a TE, obtaining an intraclass correlation coefficient or Kappa Index greater than 0.95.
None of the studies chosen for this review described secondary effects.
The correlation between elasticity measured using TE and fibrosis diagnosed using a biopsy was analysed in 25 studies. Twenty one of these described a statistically significant correlation (p<0.05). A total of 7 studies confirmed the relationship between TE and biopsy, and the significant increase in elasticity as a function of fibrosis stage. The length of the hepatic biopsy sample was very variable (range 2-60 mm). Most studies used the METAVIR scale to measure the degree of liver fibrosis. The percentage of patients with F0-1 varied between 18-65%; with F2 between 15.4-50%; with F3-4 between 12.5-54%; and with cirrhosis (F4) between 8-38%.
The following meta-analyses were carried out to study the ability of TE to discriminate different fibrosis stages:
- Meta-analysis for fibrosis diagnosis F?1:
4 studies were included (N=288) with a median cut-off=6.10 kPa. A threshold effect was observed (rho=0.80) and statistically significant heterogeneity for Sp and LR+, significant heterogeneity for Se(Chi-squared=5.10; p=0.165), LR- (Cochran-Q=4.15; p=0.246) and for DOR (Cochran-Q=3.22; p=0.359) was ruled out, confirmed using the I2
statistic (41.2% for Se, 27.2% for LR-, and 6.7% for DOR). The pooled values of Se, LR- and DOR were 0.81 (95% CI: 0.74-0.87), 0.24 (0.16-0.37) and 23.09 (8.74- 60.98) respectively. The value of AUC-SROC was 0.899 and Q* point 0.830.
- Meta-analysis for fibrosis diagnosis F?2:
18 studies were included (N=2,288) with a median cut-off=7.70 kPa. A threshold effect was ruled out (rho=0.161), with great heterogeneity among studies for all the diagnostic estimators, statistically confirmed (p<0.001; I2 =57% for DOR and >80% for Se, Sp and LRs). Two outliers were observed which were excluded and a new analysis was carried out (N=2,122). The median cut-off increased to 7.90 kPa. No threshold effect was again observed (rho=0.096) and significant heterogeneity was again observed for all diagnostic estimators. After adjusting data to a SROC curve, an AUCSROC value of 0.896 and Q* point of 0.827 were observed.
To assess possible TE performance differences in the F?2 stage, a meta-analysis was carried out including only the 8 studies (N=860) which included CHC patients, co-infected or not with HBV or HIV. In this case, no threshold effect was observed, but there was great heterogeneity (p<0.001) for all estimators, with an AUC-SROC of 0.888 and Q* point of 0.819.
After excluding an outlier which was encountered, the analysis was repeated (N=788) with a median cut-off of 8.50 kPa. The results were similar; no threshold effect and great heterogeneity for all estimators except DOR, with an pooled value of 22.21 (11.33-43.55). The SROC curve values were AUC=0.892 and Q*point=0.823.
- Meta-analysis for fibrosis diagnosis F?3:
15 studies were included (N=1,695) with a median cut-off=10.30 kPa. A slight threshold effect was observed (rho=0.39) and great heterogeneity between studies for all diagnostic estimators (p<0.001). The values of the SROC curve were AUC=0.896 and Q* point=0.827.
The meta-analysis carried out with the 4 studies (N=410) which only included CHC patients co-infected or not with HBV or HIV, ruled out the presence of a threshold effect (rho=0.20) and statistical heterogeneity (p>0.05) between studies for all estimators. The pooled values for Se, LR+, LR- and DOR were 0.82 (0.74-0.89), 7.50 (4.35-12.91), 0.21 (0.14-0.31) and 35.66 (19.23-66.14), respectively. The value for AUCSROC was 0.916 and Q* point was 0.849.
- Meta-analysis for fibrosis diagnosis F4:
20 studies were included (N=3,052) with a median cut-off of 15.60 kPa. Threshold effect was ruled out (rho=0.145) although great heterogeneity between studies for all diagnostic estimators was observed (p<0.001). The values of the SROC curve were AUC=0.953 and Q* point=0.896.
The analysis performed with 8 studies (N=834) which included only CHC patients co-infected or not with HVB or HIV with stage F4, ruled out a threshold effect and heterogeneity for Se and LR-, while heterogeneity for Sp, LR+ and DOR persisted (p<0.001). The pooled value for Se was 0.89 (0.84-0.93) and for LR- of 0.16 (0.09- 0.26). With the graphical representation, 2 outliers were detected and excluded from the study, and the analysis was repeated (N=757) with a median cut-off of 14.55 kPa. A significant threshold effect was observed (rho=0.696) and no significant heterogeneity was found between studies for all estimators except Sp (p=0.013). The value of AUCSROC was 0.968 and Q* point was 0.917.
Lastly, the meta-regression carried out including the publication format (original articles or abstracts) as a co-variable in the model, showed that there was a statistically significant influence (p<0,001) on the DOR value. The relative DOR value was 8.76 (1.81-42.42).
CONCLUSIONS
1) Current scientific evidence regarding TE shows that it is a valid technology for diagnosing liver fibrosis due to different aetiologies, especially to confirm the presence of cirrhosis. In patients diagnosed with cirrhosis using FibroScan, the performance of a biopsy could be avoided. However, its use to discern between early and intermediate stages of fibrosis is limited.
2) To achieve maximum accuracy in results using FibroScan, the procedure should be carried out following the manufacturers indications. Technological improvements are being carried out to try to resolve some of the limitations which have appeared in using this technology, such as new probes to use with obese patients or for individuals with small intercostal spaces.
3) The use of FibroScan could avoid using biopsies in a significant number of patients, thus reducing the risk associated which such test.
4) TE is a safe test, well accepted by patients, and not associated with complications nor adverse effects.
5) The introduction of FibroScan as a diagnostic technology in health services does not imply relevant organizational changes. TE could be carried out, as well as at the hospital level, in specialised consultations as it is an easily carried out procedure and does not need to be performed by medical personnel.
6) FibroScan is not an expensive technology. It is foreseeable that its use will lead to a reduction in costs, although cost-effectiveness and cost-utility studies are needed to assess this.
7) It would be advisable to study the use of TE with other serum biomarkers to properly assess its diagnostic effectiveness in this combined use.
8) Due to its safety and good acceptance by the patient, TE could be a useful tool to study liver fibrosis in children and to screen for high risk patients.
9) At the same time, other possible clinical indications for FibroScan should be explored, such as its prognostic capability and its repercussion in therapeutic decision-making. It is foreseeable that FibroScan could be used to select patients which should undergo treatment and indicate the best time to begin such treatment. In addition, TE could become the chosen technique to follow up fibrosis and monitor treatment, assessing fibrosis progression or regression.