SUMMARY

One of the most important aspects of the histopathological assessment of liver biopsies in the setting of chronic liver disease is determination of the degree of fibrosis and architectural change. Most of the work in this regard has been concerned with chronic viral hepatitis. This article attempts to assess critically our current and historical biopsy practice, from subjective fibrosis scoring systems to biopsy sample size; and the appropriate use of the data that scoring systems generate in the research and clinical setting. An understanding of the limitations of each of the components of the fibrosis assessment process can help to devise appropriate protocols to ensure that the information obtained is optimised, and its degree of reliability appreciated. It is only from this starting point that recently promulgated antifibrotic medications and “non-invasive” liver fibrosis assessment techniques can be evaluated properly.

INTRODUCTION

The degree of liver fibrosis is one of the most important diagnostic and prognostic assessments in chronic liver disease. Histological assessment of fibrosis is regarded as the “gold standard” in this respect. Clinical manifestations of liver disease and liver dysfunction accompany architectural changes of the liver parenchyma that are a result of advanced stages of liver fibrosis. Previously, it was thought that liver fibrosis and end stage liver disease (cirrhosis) were irreversible, and therefore crude determinations of liver fibrosis were acceptable because the therapeutic impact of this assessment was relatively minor. Recent work suggests that liver fibrosis may be modified by treatment^1–4 and so critical re-evaluation of the histopathological methods used to assess liver fibrosis is necessary. Liver biopsy assessment of fibrosis is not only the end point for the development of antifibrotic treatments but also forms the benchmark for validation of serum “surrogate markers” of liver fibrosis and other non-invasive techniques that purport to measure liver fibrosis and stage of liver disease.

There are several important issues which make interpretation of the current literature difficult.

• Firstly, the bulk of the literature published recently regarding the adequacy of liver biopsies is based on studies of chronic viral hepatitis. This literature suggests that the majority of liver biopsy samples are too small to be representative and therefore are inadequate for a reliable assessment of histological stage or grade.

• Secondly, that the histopathological scoring of the stage of liver disease is largely an assessment of architectural change and not a measurement of amount or degree of fibrosis.

• Thirdly, that stage scores are often mistaken for measurements, and inappropriate statistical techniques are then applied.

• Lastly, although recognised but often forgotten, that the scores generated are prone to considerable intraobserver and interobserver error.

Thus histopathological liver fibrosis assessment has become a tarnished “gold standard” and it needs to be improved. Progress will depend on recognition of the problem, a better understanding of the need for an adequate sample, and a much better understanding of the nature of “staging scores”—including the use of more appropriate statistical tools and strategies to minimise observer variation. For proper quantification of liver collagen, image analysis (IA) of appropriately stained liver sections or biochemical assay might be necessary.

SCORING SYSTEMS AND VARIABILITY

Before going on to address the important issue of biopsy size and adequacy, we would like first to consider analysis of the biopsy. Currently, most centres carry out descriptive reporting when assessing diagnostic biopsies from an individual patient. However, in the trial and research setting, evaluation of biopsies is carried out using subjective scoring systems that produce shorthand values for various categories of inflammation (grade), and fibrosis and architectural disruption (stage). There are many such systems^2,5,6 but essentially they “score” (that is, categorise) similar features.

The history of these scoring systems dates back to 1981 when the histological features of chronic hepatitis were evaluated for potential importance in determining the prognosis according to the understanding at the time of the pathophysiology of chronic hepatitis B virus (HBV) infection, and organised into a scoring system by Knodell and colleagues.⁵ Evaluations were based on review of 14 biopsies from five patients: one with chronic HBV and four with presumed chronic non-A-non-B infection—at that time, hepatitis C virus (HCV) had not been identified. In the Knodell “histological activity index” system (HAI), each of four histopathological features (periportal ± bridging necrosis, intralobular degeneration/focal necrosis, portal inflammation, and fibrosis) are assessed separately and assigned a score. In order to exaggerate the difference between mild and serious disease, Knodell and colleagues⁵ eliminated the number 2 from their scoring system for each of these features. Some consider this a drawback of this system, as it makes the scores discontinuous, but by omitting the 2, the Knodell system conveniently and simply allows the four key histopathological aspects of chronic liver disease (noted above) to be assigned to categories of: normal or minor change and two degrees of major change for each of the axes of assessment apart from periportal ± bridging necrosis (which is weighted further in the presence of bridging necrosis because of its perceived pathophysiological importance).

The currently most widely used system is the Ishak, or “revised Knodell”, system⁶ which attempts to correct the criticism of numerical discontinuity by reintroducing the number 2. The criticism is itself erroneous when it is understood that the “numbers” associated with any histopathological scoring system represent a numerical shorthand for a descriptive categorical assignment, and are neither integers nor numerical measurements along a continuum in a mathematical sense (see below).

The first three axes of the Knodell HAI relate to the necroinflammatory grade of the disease. The fourth feature (“fibrosis”: which incorporates architectural changes) assesses the stage of disease by evaluating the degree of fibrous portal tract expansion, fibrous portal-portal linking, portal-central fibrous bridges, and the formation of fibrous septa and parenchymal nodules.

The Ishak system⁶ assesses fibrosis in seven categories, ranging from normal to cirrhosis (see fig 1) and so has potentially more discriminant descriptive power. All scoring systems basically use the same principles to record liver disease stage. It is obvious that the “stage” or “fibrosis” score is composed of a mixture of features, none of which specifically depends on the amount of fibrous tissue in a liver biopsy sample. The higher Ishak scores particularly depend more on architectural changes and degree of nodularity rather than amount of fibrous tissue. Even the lower scores (for example, “portal tract expansion”) are only partly dependent on the amount of portal tract collagen because portal tracts can be considerably expanded by inflammatory infiltrates as well (fig 2).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

 Stage component of the Ishak system.⁶ *Proportion (%) of area of illustrated section showing Sirius red staining for collagen (collagen proportionate area).

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

 Relationship between Ishak stage scores and measured amount of fibrosis. Examples of the Ishak categories illustrated in fig 1 have been plotted against their individually measured collagen proportionate area (CPA). Clearly, the measured amount of fibrosis, and the corresponding Ishak category, are different evaluations. The increase in fibrosis with Ishak disease stage category is not a straight line.

Although in the original Knodell publication HAI was stated to be “numerical, objective, and reproducible”,⁵ it has, strictly speaking, never been any of these things. The considerable observer variation in the subjective interpretation of the categorical assignments has been often underestimated or ignored as a source of variability, despite the fact that both inter- and intraobserver variability have been well documented with respect to the various scoring systems.^7–9 Indeed, Knodell and colleagues recommended that a change of four points in the aggregate score was required before there was a high probability of a real or significant change, rather than merely representing a shift attributable to reinterpretation or mistaken interpretation of the biopsy features.⁵ Not surprisingly, studies have shown that this variability is reduced if there are fewer categories to choose from within each axis of assessment, but this approach reduces the descriptive power of the more abbreviated scoring systems that have been proposed.

The METAVIR scoring system was designed for the assessment of HCV chronic hepatitis specifically and reproducibly. In its current formulation,¹⁰ HCV hepatitis activity is based on evaluation of piecemeal necrosis and lobular necrosis. Portal inflammation is excluded from the METAVIR algorithm “because this feature is a prerequisite for the definition of chronic hepatitis even without activity.” Nevertheless, the recent study by Rousselet and colleagues,¹¹ using the METAVIR system to examine 254 liver biopsies from patients with chronic viral hepatitis, found that: “the level of experience (specialisation, duration, and location of practice) had more influence on agreement than the characteristics of the specimen (length, fibrosis class number, miscellaneous factors). Agreement can be improved by experienced pathologist or consensus reading”.¹¹ The experience of the histopathologist is an important variable that must be considered and optimised.

Procedures to minimise observer variation are required when any scoring system is used in studies of liver fibrosis: pre-trial consensus meetings should be held by the histopathologists involved who should have sufficient experience of the disease to be studied to debate and decide the precise fibrosis assignment categories and the borders between them. Then the entire study series should be assessed by at least two hepatopathologists independently, followed by consensus sessions to discuss significant disagreements; and the assessments should be conducted within as short a period as is practical (that is, the assessment should not be of single biopsies weeks or months apart).¹² An additional component that can be included in study protocols to improve detection of subtle changes is the comparative review of each subject’s paired biopsies, from the beginning and end of a trial, after the initial blinded histological scoring has been performed. This should be performed with limited unblinding only (that is, comparison of biopsy pairs without knowledge of intervention versus control, or time point of biopsy). However, the subtler the change that is detected in this way, the greater the risk of over-interpretation if the samples are unrepresentative.

In reality, few studies are structured in such a way. Routine scoring of liver biopsies is not appropriate for sporadic assessments in daily diagnostic practice because of the likelihood of inter- and intraobserver errors, and promotion of the spurious idea that something quantitative has been achieved. Routine scoring is done with increasing frequency under the impression that the UK National Institute for Health and Clinical Excellence (formerly known as the National Institute for Clinical Excellence) HCV treatment guidelines require this: the guidelines state that “Interferon alpha and ribavirin as combination therapy is recommended for the treatment of moderate to severe hepatitis C (defined as histological evidence of significant scarring (fibrosis) and/or significant necrotic inflammation), at standard doses for patients over the age of 18 years”.¹³ The original guidelines (Appraisal 14, 2000) suggested in a more detailed discussion (than is present in the 2003 revision) that “Liver biopsy is undertaken, if there are no increased risks, in order to assess liver scarring and necro-inflammation according to an accepted severity scale such as the Knodell”. A later document (Appraisal 75, 2004) eliminates reference to the Knodell system. Wisely, there is neither reference to any particular scoring system nor is any score specified that mandates treatment in the 2004 guidelines. The decision of how much histological fibrosis, necrosis, and/or inflammation is “significant” should be a matter for discussion between the patient and their hepatologist, taking into account the potential risks and benefits of treatment versus no treatment in any individual case. Furthermore, recent studies suggest that in non-genotype 1 HCV, the benefit/risk balance currently favours treatment in even histologically mild disease, so that there may no longer be the need for therapeutic decisions to be biopsy driven at all.¹⁴

HISTOPATHOLOGICAL SCORES AND NUMBERS

The central dictum of Pythagoras’ followers was “all is number”. Unfortunately, where histological scoring systems are concerned, all is not number (fig 3). Failure to appreciate this essential point has led to misinterpretation of the histological scores and application of inappropriate statistical techniques. Consequently, the conclusions of most of the relevant literature are questionable.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

 Ishak stage scores are neither numbers nor measurements. Two liver biopsies, each of which is stained with Sirius red for collagen. Both biopsies show parenchymal nodules surrounded by fibrous tissue, fulfilling the histological diagnostic criteria of cirrhosis. Therefore, each biopsy can also be described as “Ishak stage 6” (which is merely a symbol for the histological definition of cirrhosis). The overall area of one of the biopsies consists of 27% collagen, and 12% of the other biopsy is collagenous (collagen proportionate area (CPA)). Obviously, the histopathological diagnosis of cirrhosis (and assignment of “Ishak stage 6”), and the measured amount of fibrosis are entirely different assessments.

Histological scores are at best ordered categorical data and not numbers. Any statistical analysis must be carried out with this in mind. One would no more consider these scores to be treatable numerically than one would add, subtract, and divide different categories under the heading of "colour", or to mix and match apples and oranges. Appropriate statistical treatment of categorical data includes the use of contingency tables to convert categorical data into frequency data, which can then be numerically manipulated.⁸ This important aspect of the nature of scoring has eluded many investigators, and little of the available literature on the subject has converted categorical data into frequency data.

Clinicians, including those who work with drug regulatory authorities (in contrast with hepatopathologists), are less familiar perhaps with the nature of the scores and the underlying definitions which lead to their assignment. For example, the French METAVIR group analysed scores of stage in biopsies from patients with chronic hepatitis C virus infection to produce an “annual fibrosis progression rate”.¹⁵ Not only does this assume a linear progression of fibrosis in HCV infection (which has not been demonstrated), but the work produced annual fibrosis progression rates with bewildering score fractions. One accepts that any histopathological fibrosis scoring scale consists of rather arbitrary points on a continuum between normality and cirrhosis, but when those points have been strictly defined at the start of a study, because the in-between points lack any histological descriptive definition, the sense of the subsequently invented fractions also lacks meaning. Despite this, the “annual fibrosis progression rate” has become popular in some similar studies, and this “measure” is in danger of becoming as entrenched as the aggregated HAI score.

The same issues are pertinent when considering other histopathological scoring systems. A scoring system suitable for a particular disease process is unlikely to be suitable for a different process. For example, a scheme designed for the assessment of steatohepatitis such as that devised by Brunt and colleagues¹⁶ would be inappropriate for the assessment of chronic viral hepatitis, for which we should use scoring systems such as the Knodell or Ishak, because fibrosis develops and progresses in different ways in these diseases. Such misunderstandings can be confusing until one realises that all scoring systems are based on our understanding of the pathophysiology of the disease in question and the histopathologist is only trying to describe what is seen. The misuse of histopathological scoring systems confounds efforts to understand the natural history of disease which should be based on long term longitudinal follow up and outcome. The danger of misleading others with inappropriate manipulations of histopathological scores is clear and present, and the ease with which one can fool colleagues, and oneself, is alarming.

BIOPSY SIZE

Several years ago it was stated rather arbitrarily, and probably dictated by clinical pragmatism and a compromise with reality, that “a liver biopsy containing six portal tracts satisfies most hepatopathologists”.¹⁷ Now it is timely to move forward from the laudable desire to achieve histopathological happiness to a more scientific approach with regard to liver biopsy adequacy. In the case of liver fibrosis assessment, to determine the size of an adequate (representative) liver biopsy we must know the variability of distribution of liver fibrosis within the liver for the particular disease under consideration, which is likely to change with the stage of disease. An assessment when there is a less than adequate biopsy sample needs to be considered in the light of the degree to which that analysis is reliable or unreliable (that is, the confidence interval).

Sampling error usually results in underestimation of the feature being assessed but, for example, inclusion of capsular/subcapsular tissue overestimates both histological activity and fibrosis, and inclusion of the connective tissue normally seen in large (septal) portal tracts will overestimate chronic viral hepatitis related fibrosis (see fig 4). Colloredo and colleagues¹⁸ showed that reduction of the size of the biopsy available for histological review (both the length and diameter of the needle core) influences histological interpretation, with shorter cores more likely to be scored with lower grade and stage—interpreted as “underestimation” of overall disease activity. Following this study they recommended a biopsy length of 20 mm or greater—with more than 11 complete portal tracts—in order to reliably grade and stage a biopsy for chronic viral hepatitis.¹⁸ When these liver biopsy characteristics are compared with the average biopsy (even in specialised centres), then efforts to improve sample adequacy are clearly necessary.

• Download figure
• Open in new tab
• Download powerpoint

Figure 4

 Normal liver contains more collagen in normal structures than in the small portal tracts which are usually affected by chronic viral hepatitis. This is a section of normal liver stained with Sirius red for collagen. The section includes normal liver capsule (top) and a large portal tract (middle left). In studies of chronic viral hepatitis, these areas should be excluded from measurements of liver fibrosis because chronic viral hepatitis affects small portal tracts primarily. Otherwise, subtle changes due to disease or treatment will be obscured. Uncritical inclusion of structural collagen components confounds attempts to measure disease related collagen accurately, so that editing is necessary of structures (for example, large portal tracts, capsular collagen), and technical staining artefacts that are irrelevant to the pathological process being studied.

Biopsy size is also one of the variables that impact on the reliability of scoring (both grade and stage). Two similar studies of pairs of synchronous liver biopsies from patients with HCV infection have been performed in efforts to assess the reliability of disease stage and grade assessments in chronic HCV infection. In the first study, by Siddique and colleagues,¹⁹ pairs of right lobe biopsies from 29 patients were evaluated. Each biopsy contained 4–5 portal tracts (around half the currently recommended number). Using the Knodell HAI, Siddique et al showed that in approximately 20% of cases the stage score differed between the synchronous biopsy pairs by two categories or more.¹⁹ Persico et al compared Ishak stage scores of synchronous left and right lobe biopsies.²⁰ The left and right lobe biopsies had mean lengths of 2.5 (±0.9) cm and 2.8 (±1.1) cm, respectively. There was minimal variation of stage score between the paired biopsies.²⁰ These two studies together show the need for adequate biopsies to minimise variation in stage assessment due to sampling error. The studies demonstrate that inadequate samples give unreliable results and adequate samples give reproducible results.

Bedossa and colleagues²¹ studied large liver tissue sections of resections from patients with chronic HCV infection (including both liver resections and livers removed at the time of transplantation), which were given a METAVIR stage score. Internal variation within each tissue section was not directly addressed in this part of the study, and a single stage score was ascribed to each large section. Picro Sirius Red F3BA (Sirius red) staining and IA—measuring the area of Sirius red staining per unit area of liver tissue (see discussion of IA below)—were used to determine the “virtual” biopsy size that adequately reflected the overall stage of liver disease. “Virtual” biopsies of 25 mm length (1 mm diameter, which approximately corresponds to a needle of internal diameter 1.2 mm) “correctly” staged 75% of cases. Considering that the same data set was used both to construct the IA range for each stage score and to test it, this is perhaps a little disappointing. This could be because this attempt to use IA to validate the stage score ignores the disparate nature of the techniques—scoring being to a large extent a subjective architectural assessment and IA being a measurement of fibrosis. Variation attributable to sample size decreased progressively up to a (1 mm diameter) biopsy length of 40 mm but intrinsic variability of liver fibrosis persisted in virtual biopsies of greater than 40 mm.²¹

Given these recent revelations with regard to liver biopsy adequacy, there has been concern regarding the feasibility of safely obtaining adequate samples for reliable grading and staging of chronic liver disease. Sample size is probably related to operator experience and efforts should be made to improve operator performance. Percutaneous (PLB) and transjugular (TJLB) liver biopsies are the two main techniques used to obtain liver specimens. Reviewing the literature, 11 studies were identified that have information about both the length and number of complete portal tracts (CP) in PLB ^19,22–31. In order to adjust for the different cohort sizes, mean length and number of CP were summated relative to the number of biopsies in each study (n, that is, overall mean length  = Σ ∑[n × mean length]/∑Σn). The mean number of CP in all of the studies was evaluated similarly (that is, overall mean CP  = Σ ∑[n × mean CP]/∑Σn). The raw CP data were not available from the studies to enable calculation of an overall median, which would have been more appropriate for assessment of complete portal tracts (table 1). Overall mean length was 13.5 mm, and overall mean number of CP was 6.58. Some studies did not specify the type of needle used. Where specified, Menghini-type PLB were significantly longer compared with Tru-cut type PLB (16.2 v 12.1 mm), but with almost the same mean number of CP (6.09 v 6.35). Insufficient information was included in these papers regarding the number of needle passes, which is a significant omission, as it has been shown that the complication rate correlates with the number of passes.³² Fewer studies had information for both the length and number of CP obtained with TJLB using Tru-cut and/or Menghini-type needles.^25,33–35 Based on the data given by these four studies and following the same methodology as above, these show that with an average of 2.9 passes, the total mean length and CP were 20.9 mm and 8.09, respectively (table 2).

The implication of these findings is that, in well documented series of PLB and TJLB, the average length and mean number of portal tracts for over half the patients are well below the current recommendations for reliable grading and staging of biopsies taken for the assessment of chronic viral hepatitis.³⁶ Thus in order to obtain an adequate sample, a standard percutaneous approach will require two passes on average, increasing the risk of complications.³² Via a transjugular route, several passes can be made without increasing complications. In a series of 326 TJLB from our centre, using three passes as standard, a median length of 22 mm and a median number of 8 CP was achieved without complications.³⁵ It seems reasonable to suppose that an adequate number of CP (⩾11) could be achieved safely by performing four or more passes.

MEASUREMENT

As the clinical importance of liver fibrosis assessment grows, the limitations of biopsy scoring systems are becoming more apparent. The changes described in the scoring categories are largely architectural, with little reference to the actual amount of collagen (fibrosis) in the liver sample. Routine (diagnostic) fibrosis assessment is usually carried out on a trichrome or reticulin stain. These are not specific collagen stains, and no measurement is made. If, after due consideration, it is felt to be desirable to determine precisely how much liver fibrosis there is (if liver collagen is what is meant), to validate studies of surrogate markers of fibrosis for example, or changes with antifibrotic treatment, then proper measurement of liver collagen is unavoidable. This will require time, effort, adequate sampling, and appropriate methodology. We cannot continue to pretend that subjective scoring has measured something when it has not.

Lord Kelvin said “In physical science the first essential step in the direction of learning any subject is to find principles of numerical reckoning and practicable methods for measuring some quality connected with it. I often say that when you can measure what you are speaking about, and express it in numbers, you know something about it; but when you cannot measure it, when you cannot express it in numbers, your knowledge is of a meagre and unsatisfactory kind; it may be the beginning of knowledge, but you have scarcely in your thoughts advanced to the state of Science, whatever the matter may be”.³⁷

His reference to suitable “principles of numerical reckoning” alludes to the use of appropriate statistical analysis, which we have already touched on. As the value of routine histopathology demonstrates, clinically relevant data do not have to be numerical to be useful, but to pretend that they are numerical (when they are not) is wrong. It may be informative to measure what is seen, but to imagine it has been measured without doing so is self-delusion.

This brings us to the finding of a “practicable method for measuring” liver fibrosis. It is possible to use biochemical methods to measure the amount of collagen in tissue homogenates, but this requires the destruction of the tissue, so no further information can be gathered, which is unsatisfactory in the assessment of a diagnostic liver biopsy (in which context most human study material is gathered). In a recent study, Lee et al took multiple biopsies from the same livers (each of 12 livers sampled after transplantation, resection, or autopsy), and showed that hydroxyproline measurements of the biopsies from each liver had a coefficient of variance between 13% and 36% (mean 23%), and that this variation was present at all stages of liver disease.³⁸ In this study, 16 gauge Tru-cut needle biopsies with biopsy lengths of 10–15 mm were used and “obviously identifiable large vessels on the biopsy samples were excised and excluded from hydroxyproline measurements”. Because of the nature of the biochemical assay, we cannot know the number of complete portal tracts in the samples but this study illustrates that even biochemical approaches to the quantification of liver biopsy fibrosis will be subject to appreciable sampling variability, and dependent on sample size.

Computer assisted IA of histochemically stained sections is a method for measuring fibrosis morphologically, it does not hinder the other necessary diagnostic evaluations, and collagenous structures irrelevant to the disease process (and which contribute to the variability between samples) can be excluded precisely (fig 4). Broadly speaking, IA uses segmentation of digital images to measure the area of collagen and the area of tissue, producing a “fibrosis ratio” or collagen proportionate area (CPA).

To quantify hepatic collagen histologically, it is necessary first to stain it specifically. Sirius red has an affinity for most hepatic collagens, including types I and III (the major components of liver collagen)^39,40 and this binding correlates with chemical hydroxyproline assay under standardised laboratory conditions.⁴¹ Although the nature of the binding of the dye is not fully understood, staining of liver sections appears to be reliable and reproducible. Thus Sirius red staining is the preferred histochemical method when quantifying liver fibrosis, even though the staining may not be stoichiometric.⁴² Some groups interested in IA have used reticulin or trichrome stains, and others have suggested the use of immunohistochemistry, but due to variability in staining and difficulty in accurate thresholding, these techniques can suffer from poor reproducibility.

Normal human liver is estimated to contain approximately 5.5 mg/g of collagen, and cirrhotic liver contains of the order of 30 mg/g.³⁹ Other estimates of normal liver collagen concentration suggest 2–8 mg/g wet weight of liver.⁴³ We must also consider that different disease processes probably lead to different amounts of collagen deposition at different stages of disease. Even the same disease process at the same stage is likely to produce different amounts of collagen in different individual patients. This aspect of collagen deposition has not yet been fully evaluated.⁴⁴

IA studies have looked at fibrosis in many conditions, producing overall similar results ranging from 1–4% fibrous tissue in normal liver to 15–35% fibrous tissue in cirrhosis.^21,45–50 However, the differences in the technology and methodology used in these studies means that the results are not directly comparable.

Many of these studies have also attempted to correlate the image analysis measurements of fibrosis with the categorical stage scores. Manabe et al found that CPA did not correlate well with the components of the Knodell scoring system.⁵⁰ Pilette et al used the Knodell and a modified METAVIR scoring system and found that these correlated with liver biopsy percentage area of fibrosis.⁴⁸ Kage et al found that CPA correlated with their modified staging system based on the Desmet and Scheuer staging systems.⁴⁶ This collection of inconsistent results can be explained by the notion that IA and histopathological stage scores assess quite different things, and in different ways. In assigning histological stage score categories, much depends on architectural changes. It may be in some (or many) cases that the amount of fibrous tissue increases with these architectural derangements, but that is not specifically addressed in any of the scoring systems. No heed is paid to the width of a fibrous septum or the size of a nodule, for example, so two cirrhotic livers scoring 6 on Ishak staging might contain vastly different amounts of collagen, and thus have different IA measurements (fig 3). Despite this argument, O’Brien et al showed that IA measurements correlated well with Ishak stage scores, but only in the higher stages.⁴⁵ Results such as these may be chance occurrences, or only applicable to the limited framework of the individual study.

We are in the early days of elucidating the best way of using IA reliably, and the clinical utility or otherwise of IA has yet to be established. At present, a range of computer and camera hardware, software, and experimental procedures (and liver sample sizes) are being used, and so it is not surprising that some of the results are contradictory and confusing. When some of the methodological aspects have been resolved, IA may find a useful place in the armamentarium available to assess liver biopsy fibrosis.^51,52 Just as in biopsy scoring, there are elements of study design that can be put in place to minimise variability (see above), so with IA it is important to pay attention to various aspects of the methodology (including section thickness and staining, light source stability, camera characteristics, and accurate thresholding⁵³ of the image) to ensure stability and reproducibility of measurements. In addition, just as a mental note is made of capsular collagen, large septal portal tracts, and blood vessels when assessing fibrosis by scoring (so that these normal collagenous structures can be excluded from the disease stage scoring because they do not represent disease related collagen), so too in IA these structures need to be edited out of the measurement, particularly in the analysis of biopsy specimens, so as not to overshadow small disease or treatment related changes in the measured amount of fibrosis (figs 4, 5).

• Download figure
• Open in new tab
• Download powerpoint

Figure 5

 Cumulative mean of collagen proportionate area of normal liver stained with Sirius red (unedited versus editing of structural collagen). A smaller sample is adequate and representative if collagen that is not relevant to the disease process being studied is removed from the measurement. A traditional morphometric method of determining representative sample size is by measuring the feature of interest (in this case, liver fibrosis) in consecutive microscopic fields, and calculating the “cumulative mean”. When a stable cumulative mean is achieved, the overall area measured can be regarded as an adequate (representative) sample. Editing (exclusion) of large portal tracts (>0.5 mm diameter), capsule, and technical artefacts in this experiment (measurement of collagen proportionate area) achieves a stable cumulative mean with a much smaller sample size. Without editing, fluctuations of the cumulative mean persist at sample sizes equivalent to quite large liver biopsies. The graph also illustrates the dominant proportion of collagen that resides in normal structures in normal liver, and the relatively small amount of fibrous tissue that is normally present in small (<0.5 mm diameter) portal tracts.

It is important to remember that IA in its simplest form is a measurement of area, and is unable to evaluate architectural changes such as nodularity, fibrous portal linking, and portal-central fibrous bridging, which are the histological features of architectural change included in stage scoring systems. Thus IA and stage scoring are different assessments (although they may in some circumstances be linked) and should be understood as complementary aspects of liver biopsy assessment (figs 2, 3).

NON-INVASIVE ASSESSMENT OF FIBROSIS

The procedure of obtaining a liver biopsy is not without its attendant risk of morbidity and even mortality. The risk of complications is of the order of 1%, and the risk of mortality has been put at between 0.1% and 0.01%. The advent of the transjugular approach may reduce these further, particularly in high risk patients, and may allow multiple samples to be taken without the increased risk that is seen with multiple percutaneous passes in standard liver biopsies.^54,55 With this small but significant danger in mind and taking into account intra/interobserver error as well as sample variability, non-invasive methods of liver fibrosis assessment have been developed. These include using various imaging techniques and serological markers—“surrogate” markers of fibrosis.

Imaging techniques such as ultrasound, computed tomography, and magnetic resonance scanning as yet cannot detect small changes in fibrosis in an individual patient although some studies have shown good correlation with stage scoring. Newer techniques may be more sensitive but are still experimental. Sandrin et al have shown good correlation between a non-invasive “liver stiffness” measurement (based on transient elastography of the liver) with the METAVIR stage score, a correlation which was shown by Ziol et al to improve with larger biopsy specimens.^56–58

Traditional serological markers of liver function—“liver function tests”—give little indication of the various underlying pathological processes, including fibrosis. More complex panels of markers have been evaluated to examine the possibility of non-invasive fibrosis assessment.^59,60 The panel of markers developed by Rosenberg et al has limited application because it was only reliable for estimating the absence of fibrosis and not its extent.⁶¹ Poynard et al have developed a patented algorithm (the “Fibrotest”) which they have shown can predict significant fibrosis in chronic HCV infected patients according to the METAVIR stage score, with an “area under the receiver operator curve” of 0.73–0.87 and a negative predictive value (excluding significant fibrosis) of 91%.⁶⁰ However, we must consider what these calculations mean biologically, and how they have been validated. We have already mentioned that the use of stage scoring systems—particularly those with few categories, such as the METAVIR system—has well recognised limitations, and we must also consider the issue of sample adequacy. In the study by Rosenberg et al, a liver biopsy length of ⩾12 mm length or content of ⩾5 portal tracts (it is not specified if the portal tracts were complete or incomplete) was necessary for inclusion of the specimen in the study.⁵⁹ Generally, the performance of non-invasive tests of liver fibrosis has been evaluated using reference histological samples which were biopsies of suboptimal size by current standards (<20–25 mm length and/or containing <11 complete portal tracts) or the quality of which was not mentioned.^58,62–65 In a recent editorial, Thuluvath and Krok remind us that the validity of non-invasive serum marker assessments is yet to be established in longitudinal (as opposed to cross sectional) studies, and the results of already published studies need to be replicated in different laboratories before these assessments can play an accepted part in the clinical management of patients.⁶⁶

The utility of each of the new “non-invasive” analytical approaches must be validated against long term patient follow up and precisely measured end points. If one of the end points is liver fibrosis, this must be measured specifically, reliably, reproducibly, accurately, and using adequate samples. Until this has been done, surrogate markers of liver fibrosis must be regarded as qualitative approximations, and not quantitative assessments.

CONCLUSION

With the emergence of potentially efficacious antiviral and antifibrotic agents, much depends on making sure that assessment of liver fibrosis is reliable. The present method of subjective assessment of liver fibrosis and architecture by a single histopathologist is reasonable in the daily diagnostic situation, but application of grading and staging scoring systems is inappropriate routinely. Adequate samples according to current standards for the disease in question must be obtained to avoid error, and if these are not available, the limitations of suboptimal samples must be recognised, acknowledged, and stated clearly.

Histopathological stage scoring is sufficient for many clinical trials, and is the correct approach for observational studies. However, better design, improved organisation, and more appropriate statistical evaluation than has generally been the case up to now is needed in studies involving liver biopsy analysis. The use of fibrosis stage scores as numerical data is an unacceptable statistical technique in studies using subjective histopathological categories as an end point. Reviewers, researchers, and readers require education with regard to this matter, not only with respect to the design and interpretation of current and future studies, but also in the interpretation of studies already published. Formal quantification of liver collagen may be necessary in some studies.

There are great dangers attendant upon a failure to recognise the pitfalls in our current liver fibrosis assessment practice and if on recognition of such we do not make the required improvements. There are dangers from the point of view of future patients, who may be inappropriately treated or denied treatment that has been mistakenly discarded because of ignorance of the histopathological process. The “golden standard” of liver biopsy assessment of fibrosis has become tarnished with age, but with effort it can be polished to regain its rightful place as a more worthy gold standard.⁶⁷