The Biliary Atresia Research Consortium was established to promote clinicopathological and translational research in BA. It was essential to develop a standardized system of histological reporting in the context of a multi-institutional study. The primary goal of the current study was to establish a semi-quantitative assessment system for the histological evaluation of liver biopsy specimens from infants with cholestasis that would lead to a better understanding of the pathogenesis of BA, aid in the recognition of other cholestatic disorders with which BA initially may be confused, and for BA prognostication following Kasai portoenterostomy. The features that were used expanded upon those reported in several retrospective studies11-13
, and employed the Ishak grading system to assess the degree of fibrosis14
. The number of choices for each histologic feature were limited to as few grades as possible to enhance interobserver reproducibility15
The second goal of this study was to evaluate the predictive value of the liver biopsy for the diagnosis of BA and to identify which histologic features were most associated with a diagnosis of BA. It needs to be stressed that the clinical information available to the pathologist was restricted to age at biopsy in order to increase the objectivity of the interpretation of histological findings. Critical key clinical data critical for interpretation of biopsies in infants with cholestasis, such as TPN status and the alpha-1 antitrypsin phenotype, were withheld.
This study shows that inter-observer variability is a problematic for some of the included histological features. This may reflect insufficient emphasis on training before the individual blinded assessments were undertaken, the inclusion in this study of 18 of 97 biopsies from children beyond the usual age for diagnostic purposes (12 weeks), or that the definitions used were ambiguous. On the other hand, despite being blinded to clinical information, consortium pathologists consistently recognized histological features of biliary obstruction in approximately 90% of cases of BA. Furthermore, the assessment of a limited number of key histologic features provides the critical information needed to conclude that obstruction is present. The complete clinical follow-up provided an accurate denominator (49 cases of BA) for calculating diagnostic accuracy. The study pathologists agreed on diagnosis of BA in 36 and of obstruction other than BA in 7 of the 49 cases of proven BA (87.8% accuracy). Either of these histologic “diagnoses” would lead to the need for surgical exploration and thus to confirmation of the diagnosis of BA. Four of the remaining six “false negative” liver biopsies were inadequate specimens for proper evaluation (small and fragmented) and one was from a 2-week-old infant, an age when BA may not be fully expressed. Thus, if the inadequate specimens are excluded from the analysis, the histologic diagnosis would have been BA or obstruction other than BA in 96% of infants eventually shown to have BA. Therefore, if adequate biopsy specimens are provided, it can be expected that there would be a very high degree of accuracy in prediction of BA. Based on the consensus of the pathologists from this study, an adequate liver biopsy for interpretation in an infant should be a minimum of 2.0 cm long, and 0.2 mm wide or contain at least 10 portal areas, and if a surgical wedge, sufficiently deep to include 6 complete portal tracts independent of the liver capsule.
Several published reports from single institutions on the accuracy of liver biopsy for the diagnosis of BA involved a limited number of pathologists5, 16-18
. Brough and Bernstein retrospectively compared the original pathologic diagnosis in 158 consecutive cases with the ultimate clinical diagnosis 8
. The original pathological diagnosis was correct in 148 cases, an accuracy rate of 93.7%. Six of 10 cases of hepatocellular disease histologically misdiagnosed as favoring obstruction could not, even upon review, be differentiated from mechanical obstruction. In a similarly designed retrospective study by Ferry et al, the initial liver biopsy correctly predicted the clinical diagnosis in 94% of 143 cases19
. It was against this background that the current study was designed to evaluate the predictive value of liver biopsy for the diagnosis of BA. The pathologists prospectively interpreted liver biopsies blinded to clinical information and the complete clinical follow-up provided an accurate denominator for calculating diagnostic accuracy.
Zerbini et al. applied logistic regression analysis to 100 liver biopsy specimens and confirmed that bile duct proliferation and bile plugs were the best histologic predictors of obstruction13
. Ferry et al. also reported that bile duct proliferation is the key feature in biopsies from patients with BA19
Similar to these investigators, the current study found that items in the scoring system showing the greatest difference in the gradient of responses between BA and non-BA cases were bile duct proliferation, bile plugs in ducts and canaliculi, and the more severe grades of portal fibrosis. In addition, portal tract edema was a feature recognized more often in BA than in non-BA cases, and conversely the presence of significant lobular (sinusoidal) fibrosis militated against BA. Steatosis, hepatocellular swelling, necrosis, multinucleation, pseudorosette formation, extramedullary hematopoiesis, cholangitis and peri-ductular inflammation were seen as frequently in BA as in non-BA cases. No other features examined showed significant value in diagnosing or excluding BA.
According to Landis and Koch, a kappa value of 0.61-0.8 indicates substantial agreement, 0.41-0.6 moderate, 0.21-0.4 fair and 0-0.2 slight20
. The most reproducible feature was bile plugs in ducts (K = 0.65). Interobserver agreement as measured by kappa values was similar overall for needle and for wedge biopsies. Moderate interobserver agreement was reached for bile duct proliferation, the grade of portal fibrosis, extramedullary hematopoiesis, giant cell transformation and steatosis. Agreement was fair to slight for the rest. Taking into account those features that were predictive of BA and also those that had reasonably good inter-observer agreement, logistic regression resulted in a “best multivariate model” for diagnosing BA that included bile duct proliferation and portal fibrosis and absence of sinusoidal fibrosis.
Kappa values observed in this study for many features such as steatosis, hepatocellular necrosis and ballooning and bile duct inflammation were similar to those observed in other multiobserver studies of liver biopsies, such as for hepatitis C21-23
and non-alcoholic fatty liver disease15
. On the other hand, kappa values are an imperfect measure of agreement, being dependent on the prevalence of a given feature23
. For the ductal plate malformation, for example, the kappa values were low despite a relatively high percent agreement, best explained by the low frequency of that feature in the biopsies evaluated.
The pathologists agreed on the presence of obstructive features (either “BA or “obstruction other than BA”) in 14 of 15 cases of TPN-associated liver disease and in the 3 cases of alpha-1-antitrypsin deficiency. Both these entities are known to present with obstructive histologic features and may be impossible to differentiate from BA in the absence of clinical information and with only H+E and trichrome stains available. Thus, it is important for the pathologist to have this clinical information on hand when interpreting the liver biopsy of infants with cholestasis.
This study did not address whether histological features in a biopsy of a cholestatic infant can provide prognostic information in addition to providing a diagnosis, as suggested by other investigators. Reproducing the current study on a larger scale might permit better use of the scoring system to predict outcome. This should become possible as many-fold more patients have been prospectively enrolled into an ongoing longitudinal study in BARC, into which extensive clinical information is entered and follow-up is closely maintained.
In summary, we have developed a systematic histological evaluation system for assessment of liver biopsies of cholestatic infants and have shown reasonable to substantial inter-observer agreement on a number of features that have diagnostic utility. We have also shown that experienced pediatric pathologists can correctly identify BA with a high degree of sensitivity and good interobserver agreement, even with minimal clinical information. However, distinguishing between BA and disorders such as TPN liver disease and alpha-1 antitrypsin deficiency is not possible on biopsy alone without adequate clinical information provided to the pathologist, which is the standard of practice for clinical interpretation of these biopsies. The BARC scoring system appears to be a useful semi-quantitative assessment tool of liver biopsies from infants with cholestasis.