Serum markers detect the presence of liver fibrosis: A cohort study

Background & Aims: Histologic examination of a liver biopsy specimen is regarded as the reference standard for detecting liver fibrosis. Biopsy can be painful and hazardous, and assessment is subjective and prone to sampling error. We developed a panel of sensitive automated immunoassays to detect matrix constituents and mediators of matrix remodeling in serum to evaluate their performance in the detection of liver fibrosis. Methods: In an international multicenter cohort study, serum levels of 9 surrogate markers of liver fibrosis were compared with fibrosis stage in liver biopsy specimens obtained from 1021 subjects with chronic liver disease. Discriminant analysis of a test set of samples was used to identify an algorithm combining age, hyaluronic acid, amino-terminal propeptide of type III collagen, and tissue inhibitor of matrix metalloproteinase 1 that was subsequently evaluated using a validation set of biopsy specimens and serum samples. Results: The algorithm detected fibrosis (sensitivity, 90%) and accurately detected the absence of fibrosis (negative predictive value for significant fibrosis, 92%; area under the curve of a receiver operating characteristic plot, .804; standard error, .02; P < .0001; 95% confidence interval, .758–.851). Performance was excellent for alcoholic liver disease and nonalcoholic fatty liver disease. The algorithm performed equally well in comparison with each of the pathologists. In contrast, pathologists’ agreement over histologic scores ranged from very good to moderate (κ = .97–.46). Conclusions: Assessment of liver fibrosis with multiple serum markers used in combination is sensitive, specific, and reproducible, suggesting they may be used in conjunction with liver biopsy to assess a range of chronic liver diseases. Background & Aims: Histologic examination of a liver biopsy specimen is regarded as the reference standard for detecting liver fibrosis. Biopsy can be painful and hazardous, and assessment is subjective and prone to sampling error. We developed a panel of sensitive automated immunoassays to detect matrix constituents and mediators of matrix remodeling in serum to evaluate their performance in the detection of liver fibrosis. Methods: In an international multicenter cohort study, serum levels of 9 surrogate markers of liver fibrosis were compared with fibrosis stage in liver biopsy specimens obtained from 1021 subjects with chronic liver disease. Discriminant analysis of a test set of samples was used to identify an algorithm combining age, hyaluronic acid, amino-terminal propeptide of type III collagen, and tissue inhibitor of matrix metalloproteinase 1 that was subsequently evaluated using a validation set of biopsy specimens and serum samples. Results: The algorithm detected fibrosis (sensitivity, 90%) and accurately detected the absence of fibrosis (negative predictive value for significant fibrosis, 92%; area under the curve of a receiver operating characteristic plot, .804; standard error, .02; P < .0001; 95% confidence interval, .758–.851). Performance was excellent for alcoholic liver disease and nonalcoholic fatty liver disease. The algorithm performed equally well in comparison with each of the pathologists. In contrast, pathologists’ agreement over histologic scores ranged from very good to moderate (κ = .97–.46). Conclusions: Assessment of liver fibrosis with multiple serum markers used in combination is sensitive, specific, and reproducible, suggesting they may be used in conjunction with liver biopsy to assess a range of chronic liver diseases. Liver fibrosis is a common consequence of almost all chronic liver diseases. Progression of disease accounts for over 34,000 deaths per year from chronic liver disease and cirrhosis in the United Kingdom and USA alone.1World Health Organization Mortality Statistics 1999. http://www3.who.int/whosis.Google Scholar Diagnosis of liver fibrosis is usually made by the histologic analysis of liver biopsy specimens. A single biopsy specimen can be highly informative in determining diagnosis, prognosis, and appropriate management.2Desmet V. Fevery J. Liver biopsy.Baillieres Clin Gastroenterol. 1995; 9: 811-828Abstract Full Text PDF PubMed Scopus (17) Google Scholar, 3Scheuer P.J. Chronic hepatitis what is activity and how should it be assessed?.Histopathology. 1997; 30: 103-105Crossref PubMed Scopus (28) Google Scholar The role of surrogate markers in the detection of liver fibrosis is not yet established. Accordingly, liver biopsy is currently regarded as the “reference standard” index of liver fibrosis. Obtaining biopsy specimens, however, is costly4Pasha T. Gabriel S. Therneau T. Dickson E.R. Lindor K.D. Cost-effectiveness of ultrasound-guided liver biopsy.Hepatology. 1998; 27: 1220-1226Crossref PubMed Scopus (123) Google Scholar and is associated with hazards for the patient, including pain,5Gilmore I.T. Burroughs A. Murray-Lyon I.M. Williams R. Jenkins D. Hopkins A. Indications, methods, and outcomes of percutaneous liver biopsy in England and Wales an audit by the British Society of Gastroenterology and the Royal College of Physicians of London.Gut. 1995; 36: 437-441Crossref PubMed Scopus (351) Google Scholar hemorrhage, or death.6McGill D.B. Rakela J. Zinsmeister A.R. Ott B.J. A 21-year experience with major hemorrhage after percutaneous liver biopsy.Gastroenterology. 1990; 99: 1396-1400PubMed Google Scholar, 7Sherlock S. Dooley J. Diseases of the liver and biliary system. 10th ed. Blackwell Scientific, London1997Google Scholar Processing biopsy specimens is time consuming and labor intensive. For all of these reasons, frequent repetition of liver biopsies is deemed unacceptable to patients and doctors alike, although monitoring the evolution of disease or response to treatment may require repeated assessment. Due to the small size of a needle biopsy and the patchy nature of many liver diseases, biopsies may not be representative.8Regev A. Berho M. Jeffers L.J. Milikowski C. Molina E.G. Pyrsopoulos N.T. Feng Z.Z. Reddy K.R. Schiff E.R. Sampling error and intraobserver variation in liver biopsy in patients with chronic HCV infection.Am J Gastroenterol. 2002; 97: 2614-2618Crossref PubMed Google Scholar, 9Bedossa P. Dagere D. Paradis V. Sampling variability of liver fibrosis in chronic hepatitis C.Hepatology. 2003; 38: 1449-1457Crossref PubMed Scopus (1993) Google Scholar, 10Colloredo G. Guido M. Sonozogni A. Leandro G. Impact of liver biopsy size on histological evaluation of chronic viral hepatitis the smaller the sample, the milder the disease.J Hepatol. 2003; 39: 239-244Abstract Full Text Full Text PDF PubMed Scopus (721) Google Scholar The histologic analysis of biopsy specimens requires experience and skill but remains subjective and prone to both intraobserver and interobserver variation.11Theodossi A. Skene A.M. Portmann B. Knill-Jones R.P. Patrick R.S. Tate R.A. Kealey W. Jarvis K.J. O’Brian D.J. Williams R. Observer variation in assessment of liver biopsies including analysis by kappa statistics.Gastroenterology. 1980; 79: 232-241PubMed Scopus (105) Google Scholar, 12Scheuer P.J. Lefkowitch J.H. Liver biopsy interpretation. 6th ed. WB Saunders, London2002Google Scholar There is considerable clinical need to identify surrogate markers of liver fibrosis. Such tests might be used to estimate the extent of fibrosis in place of a biopsy or, alternatively, might be used in conjunction with a single liver biopsy to follow up progression or regression of fibrosis and response to changes in lifestyle and antifibrotic, antiviral, or other therapies. Ideally, these markers would be based on accurate and reproducible tests that could be automated and performed repeatedly with little disruption to patients. Serum assays for products of matrix synthesis or degradation and the enzymes involved in these processes have been investigated as surrogate markers of liver fibrosis in a number of studies.13Yamauchi M. Mizuhara Y. Maezawa Y. Toda G. Serum tenascin levels in chronic liver disease.Liver. 1994; 14: 148-153Crossref PubMed Scopus (24) Google Scholar, 14McHutchison J.G. Blatt L.M. de Medina M. Craig J.R. Conrad A. Schiff E.R. Tong M.J. Measurement of serum hyaluronic acid in patients with chronic hepatitis C and its relationship to liver histology. Consensus Interferon Study Group.J Gastroenterol Hepatol. 2000; 15: 945-951Crossref PubMed Scopus (220) Google Scholar, 15Hayasaka A. Schuppan D. Ohnishi K. Okuda K. Hahn E.G. Serum concentrations of the carboxyterminal cross-linking domain of procollagen type IV (NC1) and the aminoterminal propeptide of procollagen type III (PIIIP) in chronic liver disease.J Hepatol. 1990; 10: 17-22Abstract Full Text PDF PubMed Scopus (37) Google Scholar, 16Schuppan D. Cantaluppi M.C. Becker J. Veit A. Bunte T. Troyer D. Schuppan F. Schmid M. Ackermann R. Hahn E.G. Undulin, an extracellular matrix glycoprotein associated with collagen fibrils.J Biol Chem. 1990; 265: 8823-8832Abstract Full Text PDF PubMed Google Scholar, 17Murawaki Y. Ikuta Y. Koda M. Kawasaki H. Serum type III procollagen peptide, type IV collagen 7S domain, central triple-helix of type IV collagen and tissue inhibitor of metalloproteinases in patients with chronic viral liver disease relationship to liver histology.Hepatology. 1994; 20: 780-787Crossref PubMed Scopus (130) Google Scholar, 18Kasahara A. Hayashi N. Mochizuki K. Oshita M. Katayama K. Kato M. Masuzawa M. Yoshihara H. Naito M. Miyamoto T. Inoue A. Asai A. Hijioka T. Fusamoto H. Kamada T. Circulating matrix metalloproteinase-2 and tissue inhibitor of metalloproteinase-1 as serum markers of fibrosis in patients with chronic hepatitis C. Relationship to interferon response.J Hepatol. 1997; 26: 574-583Abstract Full Text PDF PubMed Scopus (109) Google Scholar, 19Murawaki Y. Ikuta Y. Okamoto K. Koda M. Kawasaki H. Serum matrix metalloproteinase-3 (stromelysin-1) concentration in patients with chronic liver disease.J Hepatol. 1999; 31: 474-481Abstract Full Text Full Text PDF PubMed Scopus (28) Google Scholar, 20Trinchet J.C. Clinical use of serum markers of fibrosis in chronic hepatitis.J Hepatol. 1995; 22: 89-95PubMed Google Scholar, 21Imbert-Bismut F. Ratziu V. Pieroni L. Charlotte F. Benhamou Y. Poynard T. Biochemical markers of liver fibrosis in patients with hepatitis C virus infection a prospective study.Lancet. 2001; 357: 1069-1075Abstract Full Text Full Text PDF PubMed Scopus (1291) Google Scholar, 22Castera L. Hartmann D.J. Chapel F. Guettier C. Mall F. Lons T. Richardet J.P. Grimbert S. Morassi O. Beaugrand M. Trinchet J.C. Serum laminin and type IV collagen are accurate markers of histologically severe alcoholic hepatitis in patients with cirrhosis.J Hepatol. 2000; 32: 412-418Abstract Full Text Full Text PDF PubMed Scopus (54) Google Scholar Generally, the diagnostic performance of these markers has been disappointing; however, some individual assays have shown promise in detecting cirrhosis,23Pilette C. Rousselet M.C. Bedossa P. Chappard D. Oberti F. Rifflet H. Maiga M.Y. Gallois Y. Cales P. Histopathological evaluation of liver fibrosis: quantitative image analysis vs semi-quantitative scores. Comparison with serum markers.J Hepatol. 1998; 28: 439-446Abstract Full Text PDF PubMed Scopus (192) Google Scholar, 24Guechot J. Laudat A. Loria A. Serfaty L. Poupon R. Giboudeau J. Diagnostic accuracy of hyaluronan and type III procollagen amino-terminal peptide serum assays as markers of liver fibrosis in chronic viral hepatitis C evaluated by ROC curve analysis.Clin Chem. 1996; 42: 558-563PubMed Google Scholar including hyaluronic acid (HA) in alcoholic liver disease (ALD),25Murawaki Y. Ikuta Y. Koda M. Nishimura Y. Kawasaki H. Clinical significance of serum hyaluronan in patients with chronic viral liver disease.J Gastroenterol Hepatol. 1996; 11: 459-465Crossref PubMed Scopus (71) Google Scholar or milder fibrosis in alcoholic liver disease (ALD), or milder fibrosis, such as YKL-40,26Johansen J.S. Christoffersen P. Moller S. Price P.A. Henriksen J.H. Garbarsch C. Bendtsen F. Serum YKL-40 is increased in patients with hepatic fibrosis.J Hepatol. 2000; 32: 911-920Abstract Full Text Full Text PDF PubMed Scopus (211) Google Scholar in nonalcoholic fatty liver disease (NAFLD), and YKL-40 and amino-terminal propeptide of type III collagen (PIIINP) in ALD.27Nojgaard C. Johansen J.S. Christensen E. Skovgaard L.T. Price P.A. Becker U. The EMALD GroupSerum levels of YKL-40 and PIIINP as prognostic markers in patients with alcoholic liver disease.J Hepatol. 2003; 39: 179-186Abstract Full Text Full Text PDF PubMed Scopus (127) Google Scholar Other markers have been reported to reflect changes in liver histology attributable to antiviral therapy.28Nojgaard C. Johansen J.S. Krarup H.B. Holten-Andersen M. Moller A. Bendtsen F. Danish Viral Hepatitis Study GroupEffect of antiviral therapy on markers of fibrogenesis in patients with chronic hepatitis C.Scand J Gastroenterol. 2003; 38: 659-665Crossref PubMed Scopus (22) Google Scholar, 29Patel K. Lajoie A. Heaton S. Pianko S. Behling C.A. Bylund D. Pockros P.J. Blatt L.M. Conrad A. McHutchison J.G. Clinical use of hyaluronic acid as a predictor of fibrosis change in hepatitis C.J Gastroenterol Hepatol. 2003; 18: 253-257Crossref PubMed Scopus (74) Google Scholar An alternative approach is to combine a number of serum markers to generate an algorithm capable of evaluating fibrosis over a range of severity. In chronic hepatitis C21Imbert-Bismut F. Ratziu V. Pieroni L. Charlotte F. Benhamou Y. Poynard T. Biochemical markers of liver fibrosis in patients with hepatitis C virus infection a prospective study.Lancet. 2001; 357: 1069-1075Abstract Full Text Full Text PDF PubMed Scopus (1291) Google Scholar, 30Myers R.P. Benhamou Y. Imbert-Bismut F. Thibault V. Bochet M. Charlotte F. Ratziu V. Bricaire F. Katlama C. Poynard T. Serum biochemical markers accurately predict liver fibrosis in HIV and hepatitis C virus co-infected patients.AIDS. 2003; 17: 721-725Crossref PubMed Scopus (198) Google Scholar and chronic hepatitis B,31Myers R.P. Tainturier M.H. Ratziu V. Piton A. Thibault V. Imbert-Bismut F. Messous D. Charlotte F. Di Martino V. Benhamou Y. Poynard T. Prediction of liver histological lesions with biochemical markers in patients with chronic hepatitis B.J Hepatol. 2003; 39: 222-230Abstract Full Text Full Text PDF PubMed Scopus (297) Google Scholar 5 parameters have been identified that could detect significant fibrosis with a positive predictive value (PPV) of 80%, but this approach failed to determine the severity of fibrosis in approximately 50% of patients. Recently, algorithms and indices using clinical and biochemical measurements have been advanced as methods for detecting liver fibrosis in chronic hepatitis C.32Forns X. Ampurdanes S. Llovet J.M. Aponte J. Quinto L. Martinez-Bauer E. Bruguera M. Sanchez-Tapias J.M. Rodes J. Identification of chronic hepatitis C patients without hepatic fibrosis by a simple predictive model.Hepatology. 2002; 36: 986-992PubMed Google Scholar, 33Wai C.T. Greenson J.K. Fontana R.J. Kalbfleisch J.D. Marrero J.A. Conjeevaram H.S. Lok A.S. A simple noninvasive index can predict both significant fibrosis and cirrhosis in patients with chronic hepatitis C.Hepatology. 2003; 38: 518-526Crossref PubMed Scopus (3173) Google Scholar Neither these nor other previous studies have investigated the use of a panel of fibrosis markers in a wide range of liver diseases. The search for an alternative to liver histology has led us to develop highly sensitive immunoassays to detect serum levels of a panel of molecules that represent constituents of matrix and enzymes involved in fibrosis and fibrolysis. The European Liver Fibrosis study is an international, multicenter, cross-sectional cohort study comparing the diagnostic performance of these serum markers with liver biopsy with the aim of identifying an optimal panel of serum markers that can be used in an algorithm to estimate the severity of liver fibrosis. The study and study protocols were approved by the UK South and West Multicentre Research Ethics Committee (MREC 98/6/08) and the local research ethics committees (m140/98) at all participating sites. We have investigated the relationship among levels of 9 serum fibrosis markers and liver fibrosis assessed by histologic examination of liver biopsy specimens from 1021 prospectively recruited subjects obtained as part of the investigation of chronic liver disease at 13 centers between 1998 and 2000. The full panel of molecular targets was selected to include surrogate markers of matrix synthesis or degradation, based on knowledge of the basic mechanisms involved in liver fibrosis. Markers assayed in this study are all produced by and property of Bayer Healthcare AG (Leverkusen, Germany) unless otherwise stated. They included collagen IV (Fuji [IV-4H12] paired with a polyclonal antibody, T59106R; Biodesign, Saco, ME), collagen VI, PIIINP (P3P 296/3/27; Dade Behring, Deerfield, IL), matrix metalloproteinase 2, matrix metalloproteinase 9, tissue inhibitor of matrix metalloproteinase 1 (TIMP-1), tenascin, laminin, and HA. All assays are heterogeneous immunoassays using a magnetic particle separation technique. The performance of each of the assays was determined in isolation. The sensitivity and specificity, interassay and intra-assay variation, interferences, linearity, and parallelism were determined for each immunoassay. All assays were shown to meet high clinical chemistry standards. The ranges of results obtained for healthy subjects of both sexes, aged 18–75 years, were determined to establish “normal” values. The assays were applied to subjects with a variety of pathologic disorders. Performance of the key components of the final algorithm is presented in Table 1. The recruitment of patients in the study is shown in Figure 1.Table 1Performance Characteristics of Key AssaysAssayRange (ng/mL)Minimum detectable level (ng/mL)Coefficient of variation (%)Parallelism (%)Linearity (%)Reference range (ng/mL),mean ± SD (no.)Interassay variationTotalPIIINP0–151.51.0–5.31.9–5.3.699.5–100.15.84 ± 3.26 (199)TIMP-10–3000<23.0–2.84.7–3.21098.6–101.8619 ± 111.7 (225)HA0–10005.07<1010–121094.7–100.68.16 ± 8.21 to 34.6 ± 22.47 (382)NOTE. The performance of each assay was determined using samples obtained from healthy donors of both sexes aged 18–75 years. The range, minimum detectable level, and coefficient of variation presented for interassay variation as well as total variation are shown. The results for assessment of parallelism are presented as percentage variation from the expected result and for linearity as the range of variation around 100%. For HA, concentrations are known to be age dependent. The lowest and highest mean results obtained for HA among 182 female and 200 male apparently healthy subjects are presented for men aged 20 years and women aged 60–69 years. For TIMP-1, samples were stratified as younger than or older than 40 years and by sex; for PIIINP, samples were stratified by decade and sex. Open table in a new tab NOTE. The performance of each assay was determined using samples obtained from healthy donors of both sexes aged 18–75 years. The range, minimum detectable level, and coefficient of variation presented for interassay variation as well as total variation are shown. The results for assessment of parallelism are presented as percentage variation from the expected result and for linearity as the range of variation around 100%. For HA, concentrations are known to be age dependent. The lowest and highest mean results obtained for HA among 182 female and 200 male apparently healthy subjects are presented for men aged 20 years and women aged 60–69 years. For TIMP-1, samples were stratified as younger than or older than 40 years and by sex; for PIIINP, samples were stratified by decade and sex. Subjects were considered eligible if they were due to undergo liver biopsy for the investigation of chronic liver disease, defined as abnormal biochemical liver function tests persisting for more than 6 months. Additional inclusion criteria were the ability to provide informed consent and age older than 18 years and younger than 75 years. Patients were excluded from the study if their age fell outside of this range; for any disorder associated with extrahepatic fibrosis, including rheumatic, renal, or lung disease; for cardiovascular disease or cancer; for advanced cirrhosis with evidence of decompensation (Child-Pugh class C); for consumption of regular aspirin; or for hepatocellular carcinoma or drug-induced liver disease. Of the 1021 subjects recruited, the numbers in each diagnostic category were as follows: chronic hepatitis C, 496; ALD, 64; fatty liver, 61; hepatitis B, 61; primary biliary cirrhosis or primary sclerosing cholangitis, 53; recurrent disease after liver transplantation, 48; autoimmune hepatitis, 45; hemochromatosis, 32; cryptogenic cirrhosis, 19; hepatitis B and C, 4; and other (including granulomatous disease of unknown etiology, and cases in which no diagnosis was made in the investigation of abnormal liver function tests that represented the majority), 138. Men represented 63% of the sample; the average age was 44.1 years (SD, 12.8 years; range, 19–25 years). The distribution of the whole cohort between Scheuer fibrosis stages was as follows: stage 0, 24.4%; stage 1, 35.5%; stage 2, 13.4%; stage 3, 14.9%; and stage 4, 11.8%. There were no significant differences among the subjects in the total cohort (GA), the test set (GT), or the validation set (GV). Serum samples were obtained and routine blood tests were performed at the time of liver biopsy and processed immediately. The 9 different immunoassays were developed to run on the Bayer IMMUNO 1 system. No serum marker scores were deemed indeterminate. All biopsy specimens were analyzed locally and by one central pathologist (A). Clinical details or biochemical samples were incomplete for 45 subjects, and 55 of the remaining 976 biopsy specimens were considered to be inadequate for full histologic analysis due to inadequate length (<12 mm) or <5 portal tracts. Biopsy specimens, serum samples, and clinical details were therefore available for 921 subjects in the final analysis (group GA). GT and GV were derived from this group. The central pathologist assessed all 921 biopsy specimens using the Scheuer34Scheuer P.J. Classification of chronic viral hepatitis a need for reassessment.J Hepatol. 1991; 13: 372-374Abstract Full Text PDF PubMed Scopus (1393) Google Scholar and Ishak35Ishak K. Baptista A. Bianchi L. Callea F. De Groote J. Gudat F. Denk H. Desmet V. Korb G. MacSween R.N. et al.Histological grading and staging of chronic hepatitis.J Hepatol. 1995; 22: 696-699Abstract Full Text PDF PubMed Scopus (4174) Google Scholar staging systems. For conditions other than chronic viral or immune hepatitis, modifications of the criteria statements were made to reflect the distribution of fibrosis (eg, in alcoholic and nonalcoholic steatohepatitis, perivenular and pericellular fibrosis replaced portal and periportal fibrosis). To evaluate further the relationship between the serum marker scores and histologic staging, 2 additional expert hepatopathologists (B and C) independently staged a randomly selected “consensus set” of 620 biopsy specimens using the same descriptors used by pathologist A. To determine intraobserver variation, pathologist A repeated staging of all 921 biopsy specimens, blinded to the results assigned in the first staging. Interobserver variation was evaluated by comparing the results assigned by each pathologist. To derive algorithms combining serum markers, a group of 400 cases (GT) was selected at random from the group of 921 patients with biopsy specimens. Algorithms were developed by including a marker if its addition to the algorithm increased the overall generalized distance between groups. Clinical chemistry and hematology test results were also examined in this way. An optimal algorithm was selected and the performance of this algorithm was then validated in the remaining set of 521 biopsy specimens from GA, designated as GV, using the staging assigned by pathologist A. The ability of the algorithm to detect significant levels of fibrosis was analyzed using 3 outcomes. The first was a conventional approach assuming a linear increase in fibrosis severity in which the upper 3 stages of the Scheuer system were considered to represent significant fibrosis. The second was based on analysis of the distribution of the algorithm scores, and the third was based on the ability to detect cirrhosis. Biopsy specimens are frequently used to differentiate between “significant” and “nonsignificant” fibrosis for the purposes of prognostication and treatment. By convention, the series of stages in systems such as that of Scheuer from 0 to 4 are divided into 3 equal groupings interpreted as representing “mild” (0, 1), “moderate” (2, 3) and “severe” (4) fibrosis, with moderate and severe fibrosis regarded as “significant.” This approach assumes that the increase in severity of fibrosis is linear between stages and that the step up from nonsignificant to significant fibrosis occurs between stages 2 and 3. Taking an alternative approach, the distribution of discriminant scores generated by the algorithm was analyzed and plotted against histologic fibrosis staging in an attempt to derive insight into the progression of fibrosis severity from biological markers. In addition, we assessed the ability of the algorithm to distinguish cirrhosis from less advanced forms of fibrosis. The reproducibility of the performance of the algorithm was evaluated by determining its performance against biopsy staging assigned by pathologists B and C. Applied statistical methods included analysis of variance, discriminant analysis, and logistic regression for binary grouped biopsy stage. κ statistics were calculated to determine agreement between pathologists. Sensitivities, specificities, PPVs, negative predictive values (NPVs), and prevalences for the binary outcomes were assessed using receiver operator characteristic (ROC) curve analysis. All analyses were performed using the SPSS software package (SPSS, Inc, Chicago, IL). The primary aim of the study was to investigate the ability of serum markers to identify significant histologic fibrosis. The mean, median, and SEM for each marker in GT and GV were determined. A multivariate analysis of variance indicated that there were no differences between groups for all markers taken together (Hotelling’s T = .01, F = 1.14, df1 = 9, df2 = 911, P = .33). An examination of the associated individual t tests showed no significant differences between the groups on any individual marker. χ2 analysis indicated that there were no differences in the etiologic breakdown for each group (likelihood ratio χ2 = 6.34, df = 6, P = .38) (data not shown). Algorithms combining the serum markers were evaluated for their ability to discriminate between the biopsy stages in GT. Similar performance characteristics were found with algorithms that incorporate patient age, HA, collagen IV, collagen VI, laminin, PIIINP, TIMP-1, and matrix metalloproteinase 2 in varying combinations. The addition of other serum markers, the results of clinical chemistry tests including liver function tests, or hematologic indices including platelet count and prothrombin time did not improve the performance of the algorithms. For all reported analyses, the algorithm was tested for both Scheuer and Ishak staging systems. Results are presented here only for Scheuer stages; however, in all cases, similar results were obtained using the Ishak system. The results for the algorithm that resulted in the maximum separation of the biopsy groups over the full range of fibrosis stages for the validation set GV are shown in Figure 2A. The performance of the 4 components of the algorithm is presented in Figure 2B–E. Examination of the distribution of the discriminant scores plotted against histologic fibrosis stage showed clustering of the results into 3 groups with steps up in algorithm scores between stages 2 and 3 and stages 3 and 4. This was substantiated by examination of the generalized distances between stages (data not shown; see legend to Figure 3), generating 2 alternative categories of “no/mild” and “moderate/severe” fibrosis. These corresponded to Scheuer 0–2 and 3–4 rather than the bifurcation of 0–1 and 2–4 suggested by convention (Figure 3). The performance of the algorithm was assessed using logistic regression to fit the bifurcated staging to all 3 models described previously. The analyses were performed (1) using the conventional bifurcation of the linear progression of stages from 0 to 4 into a lower third (Scheuer stages 0 and 1) and upper two thirds (Scheuer stages 2–4), (2) by using the bifurcation based on the Euclidean distances between groups of algorithm scores (Scheuer stages 0–2 vs 3 and 4), and (3) for fibrosis versus cirrhosis (0–3 vs 4). Logistic scores were obtained for patients in the GT and GV groups. The results are presented in Table 2, demonstrating the area under the curve (AUC) for the bifurcated stages. For the GT series, the AUC was .863 (SE, .0212; P < .0001; 95% confidence interval, .822–.905). For the GV series, the AUC was .804 (SE, .0236; P < .0001; 95% CI, .757–.850).Table 2Algorithm PerformanceScheuer scoreAUCSEP95% CI of areaDSTSensitivity (%)Specificity (%)X = 0,1 vs 2,3,4.782.0213<.0001.740–.823−.86490.329.6Y = 0,1,2 vs 3,4.804.0236<.0001.757–.850.10290.541.0Z = 0,1,2,3 vs 4.887.0256<.0001.837–.937.02590.769.2NOTE. The data represent the performance of the algorithm in detecting bifurcated outcomes (X = 0,1 vs 2–4; Y = 0,1,2 vs 3,4; and Z = 0,1,2,3 vs 4) for the 400 GT and 521 GV samples from the whole cohort of patients with diverse chronic