FibroTest/Fibrosure for Liver Fibrosis and Cirrhosis in HBV
In this meta-analysis we have summarized the diagnostic accuracy of FT for CHB-related significant fibrosis and liver cirrhosis. This is the largest review of the diagnostic accuracy of FT in CHB patients, and the largest review by independent investigators. In addition, as statistical aspects of a systematic review of DTA still present challenges, it is of utmost importance to use validated and robust methodology that will provide usable results from a clinical standpoint; this is the main reason why we have chosen the methodological approach recommended by the Cochrane DTA Group. Additional strengths of our review include a meticulous search of published studies, formal assessment of methodological quality, assessment of heterogeneity with sensitivity analysis, and assessment of clinical repercussions of our findings.
One of the main limitations is the significant heterogeneity of included studies. A considerable variation between the results of diagnostic studies is a common occurrence, possibly to a greater extent than is seen for therapeutic interventions. This is perhaps one of the main sources of heterogeneity and a direct consequence of the fact that the importance of rigorous design has been less well appreciated for diagnostic studies than for therapeutic interventions, resulting in a poorer adherence to methodological constraints. This is noticeable in many studies that we included, and can be considered as a general problem in many studies dealing with the diagnostic accuracy of liver fibrosis markers, as already noted by others. We have also chosen to evaluate only FibroTest, despite the availability of other tests, mainly because of the limited number of publications describing the accuracy of other such tests in CHB patients. Some potentially eligible studies were excluded because we could not derive data to construct 2 × 2 tables. We also did not take into consideration the prognostic value of FT in this review—there is clear potential for the utility of FT in this area too.
Overall, our results in terms of pooled AUCs for significant fibrosis and cirrhosis are in agreement with findings reported from a previous meta-analysis conducted by FT developers (AUC: 0.79 and 0.84 for significant fibrosis and cirrhosis, respectively). Interestingly, these measurements were similar to the findings reported in an independent meta-analysis of FT in HCV patients (AUC: 0.81 and 0.90 for significant fibrosis and cirrhosis, respectively). It appears that with the current (imperfect) gold standards such as LB, these are the best numbers for FT we can achieve in terms of overall accuracy. This is confirmed by the fact that all direct, indirect, and combined serum markers of liver fibrosis have AUCs clustering around the value of 0.85. Therefore, when discussing the accuracy of any marker in the case of discordant results between biopsy and biomarker such as FT, the cause of discordance can be either failure of fibrosis marker or failure of biopsy to detect true stage.
Even in the best possible conditions, in which LB accuracy is highest (sensitivity and specificity of biopsy are 90%) and the prevalence of significant disease is 40%, the AUC for a perfect liver fibrosis marker would be 0.90, which makes it hard to distinguish a perfect surrogate from a bad one. Owing to the markedly poor risk-to-benefit ratio of LB, it has been argued that it should not be recommended as a first-line procedure but rather as a second-line estimate of liver injury, reserved for cases involving complex disorders and discordance between clinical and noninvasive findings. LB is prone to sampling errors and intra- and interobserver variability, with the additional impact of the level of experience of the pathologist on overall accuracy and reliability. The size of liver sample may also be of importance, but even a 25-mm-long biopsy sample may have up to 25% rate of discordance for fibrosis staging.
In our analysis of liver-biopsy-related factors that may have an impact on accuracy, studies with an average length of LB ≥20 mm did not have better measurements of diagnostic accuracy for either significant fibrosis or cirrhosis, in comparison with those reporting an average length <20 mm. However, studies using histological score other than METAVIR had significantly better sensitivity and DOR for significant fibrosis, although there were no differences in measurements of diagnostic accuracy for liver cirrhosis (Table 2 and Table 3). It is important to emphasize that all five studies that used a non-METAVIR score were conducted on patients of Asian origin; therefore, this may be another confounder. What may make this of even greater significance is the fact that three of the five studies used the Batts and Ludwig scoring system, which is similar to the METAVIR score.
One common problem in the studies included in the present review was the use of FT thresholds not recommended by developers, as more than half the included studies used nonrecommended FT thresholds for both fibrosis and cirrhosis. This is another possible source of bias and a strong source of heterogeneity, as selective reporting of the thresholds identified to optimize test accuracy can introduce bias if they are selected in a data-driven manner. This was confirmed in a sensitivity analysis, in which we demonstrated a significantly higher pooled sensitivity and DOR for significant fibrosis in studies that used FT cutoff <0.48 (Table 2), suggesting better diagnostic accuracy in studies using FT cutoff in the range of 0.31–0.48. However, studies that used the recommended cutoff for cirrhosis had better pooled specificity in comparison with those that used FT threshold <0.74 (Table 3). FibroTest at the proposed cutoff of 0.74 had excellent performance for exclusion of liver cirrhosis.
The developers of the FT recommend the application of FT as a continuous rather than binary variable in order to maximize its effectiveness. However, in clinical practice, knowing the exact fibrosis stage is not as important as knowing whether the patient has mild or advanced liver disease (F0–F1 vs. F2–F4 and F0–F3 vs. F4). This is of potentially crucial clinical interest in cases where the patient is to be evaluated for antiviral treatment, as the decision to start treatment and the choice of medications may be influenced by fibrosis stage. For example, differentiation between the F1 and F2 stage may be of utmost importance when deciding not to initiate long-term treatment with analogs in patients with the risk of developing resistance. Therefore, it may be a better strategy to concentrate on finding the best-performing FT thresholds for detection and exclusion of both significant fibrosis and cirrhosis.
Obviously, at least for a CHB population, there is a clear need for further refinement of proposed FT thresholds, especially for significant fibrosis, and their adaptation to clinical needs. This, however, may prove to be a hard task; Parkes and colleagues, in a review of 14 studies with 10 different tests, showed that cutoff levels with clinically relevant predictive values for the presence or absence of significant fibrosis were applicable to only 35% of the population of patients with chronic hepatitis C. Others have argued, however, that sensitivities and specificities above 85% can be considered as adequate for identifying patients with significant fibrosis, as there are no relevant clinical consequences of false positives or false negatives.
The prevalence of advanced fibrosis and cirrhosis may have an impact on overall accuracy because of a spectrum effect, and it has been observed that in cases where spectrum bias is present, either sensitivity or specificity would be expected to change. Therefore, one could expect higher sensitivity and specificity in populations in which extreme stages of fibrosis (F0 or F4) are present than in populations with higher prevalence of intermediate stages (F1–F3). This was recognized as an issue in studies dealing with accuracy of fibrosis markers; as mentioned above, Poynard et al. proposed standardization of the reported AUCs according to the prevalence of fibrosis stages. We also investigated whether the prevalence of significant fibrosis and cirrhosis may have an impact on accuracy by evaluating studies in which reported prevalence of significant fibrosis and cirrhosis was above 54 and 18%, respectively. We have chosen these two thresholds as, again, Poynard et al. reported natural prevalence of advanced fibrosis and cirrhosis in a cohort of 2,235 CHC patients to be 54 and 18%, respectively. As expected, both pooled sensitivity and specificity, and especially DOR, were better in studies with prevalence of significant fibrosis ≥54%, but without statistical significance.
Calculated values of prevalence of significant fibrosis and cirrhosis in a total of 2,494 patients with CHB included in this meta-analysis were 55.33 and 16.92% respectively—similar to the above-mentioned values. In addition, in the 2,394 patients from 15 studies with sufficient data, the value of DANA was 2.105, again similar to the DANA reported in the CHC patients. Therefore, the natural distribution of stages of fibrosis in CHB and CHC patients appears to be similar.
There was no difference between the performances of FibroTest on the basis of the independence of authors; however, although there was no statistical difference, pooled DOR and AUC were higher in studies conducted by independent authors. Although there were four studies in which we were not able to obtain data on time interval between LB and FT, studies with blood sampling for FT and LB performed on the same day had better pooled sensitivity, specificity, and DOR for significant fibrosis than those that reported a time difference between FT and LB. However, such difference was not observed in the analysis for liver cirrhosis. Inclusion of a special population of HIV-coinfected patients also did not have any impact on diagnostic accuracy, although there was a tendency for better sensitivity in studies with HIV-coinfected patients. However, owing to the small number of included studies, this result should be considered with caution and leaves space for additional exploration in future studies.
There are several implications for future research. More studies on diagnostic accuracy for liver fibrosis are needed in populations of CHB patients, regardless of the noninvasive methods being explored. Future authors of studies exploring the performance of FT in CHB patients should be encouraged to insist on a rigorous design and methodology. Although reporting AUCs is usual, it is also important to report data for diagnostic accuracy at a recommended threshold(s) in order to allow for easier extraction of data for future meta-analyses (2 × 2 table). There is a need for studies exploring best-performing FT thresholds for both dichotomous outcomes, as well as studies in which the central research question explores the diagnostic accuracy of FT in Asian populations, but with the use of the MEATVIR scoring system. More studies in HIV–HBV-coinfected patients are also warranted. A future review in which diagnostic performance of FT is compared with other noninvasive markers could also be of tremendous use, as could a review of studies using a combination of two or more noninvasive markers.
Implications for practice deriving from our results suggest that FibroTest is of excellent utility for excluding cirrhosis in patients with CHB, but has suboptimal performance in detection of significant fibrosis and cirrhosis and in exclusion of significant fibrosis. It is of importance for clinicians to adhere to the recommended thresholds for dichotomous outcomes until better ones are derived and to sample blood for FT on the day of LB, as this obviously may have an impact on accuracy.
Discussion
In this meta-analysis we have summarized the diagnostic accuracy of FT for CHB-related significant fibrosis and liver cirrhosis. This is the largest review of the diagnostic accuracy of FT in CHB patients, and the largest review by independent investigators. In addition, as statistical aspects of a systematic review of DTA still present challenges, it is of utmost importance to use validated and robust methodology that will provide usable results from a clinical standpoint; this is the main reason why we have chosen the methodological approach recommended by the Cochrane DTA Group. Additional strengths of our review include a meticulous search of published studies, formal assessment of methodological quality, assessment of heterogeneity with sensitivity analysis, and assessment of clinical repercussions of our findings.
One of the main limitations is the significant heterogeneity of included studies. A considerable variation between the results of diagnostic studies is a common occurrence, possibly to a greater extent than is seen for therapeutic interventions. This is perhaps one of the main sources of heterogeneity and a direct consequence of the fact that the importance of rigorous design has been less well appreciated for diagnostic studies than for therapeutic interventions, resulting in a poorer adherence to methodological constraints. This is noticeable in many studies that we included, and can be considered as a general problem in many studies dealing with the diagnostic accuracy of liver fibrosis markers, as already noted by others. We have also chosen to evaluate only FibroTest, despite the availability of other tests, mainly because of the limited number of publications describing the accuracy of other such tests in CHB patients. Some potentially eligible studies were excluded because we could not derive data to construct 2 × 2 tables. We also did not take into consideration the prognostic value of FT in this review—there is clear potential for the utility of FT in this area too.
Overall, our results in terms of pooled AUCs for significant fibrosis and cirrhosis are in agreement with findings reported from a previous meta-analysis conducted by FT developers (AUC: 0.79 and 0.84 for significant fibrosis and cirrhosis, respectively). Interestingly, these measurements were similar to the findings reported in an independent meta-analysis of FT in HCV patients (AUC: 0.81 and 0.90 for significant fibrosis and cirrhosis, respectively). It appears that with the current (imperfect) gold standards such as LB, these are the best numbers for FT we can achieve in terms of overall accuracy. This is confirmed by the fact that all direct, indirect, and combined serum markers of liver fibrosis have AUCs clustering around the value of 0.85. Therefore, when discussing the accuracy of any marker in the case of discordant results between biopsy and biomarker such as FT, the cause of discordance can be either failure of fibrosis marker or failure of biopsy to detect true stage.
Even in the best possible conditions, in which LB accuracy is highest (sensitivity and specificity of biopsy are 90%) and the prevalence of significant disease is 40%, the AUC for a perfect liver fibrosis marker would be 0.90, which makes it hard to distinguish a perfect surrogate from a bad one. Owing to the markedly poor risk-to-benefit ratio of LB, it has been argued that it should not be recommended as a first-line procedure but rather as a second-line estimate of liver injury, reserved for cases involving complex disorders and discordance between clinical and noninvasive findings. LB is prone to sampling errors and intra- and interobserver variability, with the additional impact of the level of experience of the pathologist on overall accuracy and reliability. The size of liver sample may also be of importance, but even a 25-mm-long biopsy sample may have up to 25% rate of discordance for fibrosis staging.
In our analysis of liver-biopsy-related factors that may have an impact on accuracy, studies with an average length of LB ≥20 mm did not have better measurements of diagnostic accuracy for either significant fibrosis or cirrhosis, in comparison with those reporting an average length <20 mm. However, studies using histological score other than METAVIR had significantly better sensitivity and DOR for significant fibrosis, although there were no differences in measurements of diagnostic accuracy for liver cirrhosis (Table 2 and Table 3). It is important to emphasize that all five studies that used a non-METAVIR score were conducted on patients of Asian origin; therefore, this may be another confounder. What may make this of even greater significance is the fact that three of the five studies used the Batts and Ludwig scoring system, which is similar to the METAVIR score.
One common problem in the studies included in the present review was the use of FT thresholds not recommended by developers, as more than half the included studies used nonrecommended FT thresholds for both fibrosis and cirrhosis. This is another possible source of bias and a strong source of heterogeneity, as selective reporting of the thresholds identified to optimize test accuracy can introduce bias if they are selected in a data-driven manner. This was confirmed in a sensitivity analysis, in which we demonstrated a significantly higher pooled sensitivity and DOR for significant fibrosis in studies that used FT cutoff <0.48 (Table 2), suggesting better diagnostic accuracy in studies using FT cutoff in the range of 0.31–0.48. However, studies that used the recommended cutoff for cirrhosis had better pooled specificity in comparison with those that used FT threshold <0.74 (Table 3). FibroTest at the proposed cutoff of 0.74 had excellent performance for exclusion of liver cirrhosis.
The developers of the FT recommend the application of FT as a continuous rather than binary variable in order to maximize its effectiveness. However, in clinical practice, knowing the exact fibrosis stage is not as important as knowing whether the patient has mild or advanced liver disease (F0–F1 vs. F2–F4 and F0–F3 vs. F4). This is of potentially crucial clinical interest in cases where the patient is to be evaluated for antiviral treatment, as the decision to start treatment and the choice of medications may be influenced by fibrosis stage. For example, differentiation between the F1 and F2 stage may be of utmost importance when deciding not to initiate long-term treatment with analogs in patients with the risk of developing resistance. Therefore, it may be a better strategy to concentrate on finding the best-performing FT thresholds for detection and exclusion of both significant fibrosis and cirrhosis.
Obviously, at least for a CHB population, there is a clear need for further refinement of proposed FT thresholds, especially for significant fibrosis, and their adaptation to clinical needs. This, however, may prove to be a hard task; Parkes and colleagues, in a review of 14 studies with 10 different tests, showed that cutoff levels with clinically relevant predictive values for the presence or absence of significant fibrosis were applicable to only 35% of the population of patients with chronic hepatitis C. Others have argued, however, that sensitivities and specificities above 85% can be considered as adequate for identifying patients with significant fibrosis, as there are no relevant clinical consequences of false positives or false negatives.
The prevalence of advanced fibrosis and cirrhosis may have an impact on overall accuracy because of a spectrum effect, and it has been observed that in cases where spectrum bias is present, either sensitivity or specificity would be expected to change. Therefore, one could expect higher sensitivity and specificity in populations in which extreme stages of fibrosis (F0 or F4) are present than in populations with higher prevalence of intermediate stages (F1–F3). This was recognized as an issue in studies dealing with accuracy of fibrosis markers; as mentioned above, Poynard et al. proposed standardization of the reported AUCs according to the prevalence of fibrosis stages. We also investigated whether the prevalence of significant fibrosis and cirrhosis may have an impact on accuracy by evaluating studies in which reported prevalence of significant fibrosis and cirrhosis was above 54 and 18%, respectively. We have chosen these two thresholds as, again, Poynard et al. reported natural prevalence of advanced fibrosis and cirrhosis in a cohort of 2,235 CHC patients to be 54 and 18%, respectively. As expected, both pooled sensitivity and specificity, and especially DOR, were better in studies with prevalence of significant fibrosis ≥54%, but without statistical significance.
Calculated values of prevalence of significant fibrosis and cirrhosis in a total of 2,494 patients with CHB included in this meta-analysis were 55.33 and 16.92% respectively—similar to the above-mentioned values. In addition, in the 2,394 patients from 15 studies with sufficient data, the value of DANA was 2.105, again similar to the DANA reported in the CHC patients. Therefore, the natural distribution of stages of fibrosis in CHB and CHC patients appears to be similar.
There was no difference between the performances of FibroTest on the basis of the independence of authors; however, although there was no statistical difference, pooled DOR and AUC were higher in studies conducted by independent authors. Although there were four studies in which we were not able to obtain data on time interval between LB and FT, studies with blood sampling for FT and LB performed on the same day had better pooled sensitivity, specificity, and DOR for significant fibrosis than those that reported a time difference between FT and LB. However, such difference was not observed in the analysis for liver cirrhosis. Inclusion of a special population of HIV-coinfected patients also did not have any impact on diagnostic accuracy, although there was a tendency for better sensitivity in studies with HIV-coinfected patients. However, owing to the small number of included studies, this result should be considered with caution and leaves space for additional exploration in future studies.
There are several implications for future research. More studies on diagnostic accuracy for liver fibrosis are needed in populations of CHB patients, regardless of the noninvasive methods being explored. Future authors of studies exploring the performance of FT in CHB patients should be encouraged to insist on a rigorous design and methodology. Although reporting AUCs is usual, it is also important to report data for diagnostic accuracy at a recommended threshold(s) in order to allow for easier extraction of data for future meta-analyses (2 × 2 table). There is a need for studies exploring best-performing FT thresholds for both dichotomous outcomes, as well as studies in which the central research question explores the diagnostic accuracy of FT in Asian populations, but with the use of the MEATVIR scoring system. More studies in HIV–HBV-coinfected patients are also warranted. A future review in which diagnostic performance of FT is compared with other noninvasive markers could also be of tremendous use, as could a review of studies using a combination of two or more noninvasive markers.
Implications for practice deriving from our results suggest that FibroTest is of excellent utility for excluding cirrhosis in patients with CHB, but has suboptimal performance in detection of significant fibrosis and cirrhosis and in exclusion of significant fibrosis. It is of importance for clinicians to adhere to the recommended thresholds for dichotomous outcomes until better ones are derived and to sample blood for FT on the day of LB, as this obviously may have an impact on accuracy.
SHARE
