Development of a simple noninvasive index to predict significant fibrosis in patients with HIV/HCV coinfection


  • Potential conflict of interest: Nothing to report.

  • Presented at the 12th Conference on Retroviral Infections, February 23–25, 2005, Boston, MA.


Liver biopsy remains the gold standard in the assessment of severity of liver disease. Noninvasive tests have gained popularity to predict histology in view of the associated risks of biopsy. However, many models include tests not readily available, and there are limited data from patients with HIV/hepatitis C virus (HCV) coinfection. We aimed to develop a model using routine tests to predict liver fibrosis in patients with HIV/HCV coinfection. A retrospective analysis of liver histology was performed in 832 patients. Liver fibrosis was assessed via Ishak score; patients were categorized as 0–1, 2–3, or 4–6 and were randomly assigned to training (n = 555) or validation (n = 277) sets. Multivariate logistic regression analysis revealed that platelet count (PLT), age, AST, and INR were significantly associated with fibrosis. Additional analysis revealed PLT, age, AST, and ALT as an alternative model. Based on this, a simple index (FIB-4) was developed: age ([yr] × AST [U/L]) / ((PLT [109/L]) × (ALT [U/L])1/2). The AUROC of the index was 0.765 for differentiation between Ishak stage 0–3 and 4–6. At a cutoff of <1.45 in the validation set, the negative predictive value to exclude advanced fibrosis (stage 4–6) was 90% with a sensitivity of 70%. A cutoff of >3.25 had a positive predictive value of 65% and a specificity of 97%. Using these cutoffs, 87% of the 198 patients with FIB-4 values outside 1.45–3.25 would be correctly classified, and liver biopsy could be avoided in 71% of the validation group. In conclusion, noninvasive tests can accurately predict hepatic fibrosis and may reduce the need for liver biopsy in the majority of HIV/HCV-coinfected patients. (HEPATOLOGY 2006;43:1317–1325.)

Chronic hepatitis C virus (HCV) coinfection is common in patients with HIV infection.1 Until recently, the clinical course of HCV infection in coinfected individuals was overshadowed by the high morbidity and mortality of HIV. With the introduction of highly active antiretroviral therapy (HAART) and its associated improvements in survival,2 HCV has now emerged as a significant comorbid disease in coinfected patients.3, 4 Although the exact mechanism is not completely understood and may5 or may not be associated with specific HAART use,6, 7 HCV appears to have a more progressive course in coinfected patients compared with those with HCV monoinfection.6, 8–11 With recent improvements in HCV therapy in patients with coinfection,12–15 accurate assessment of liver disease severity is now even more important in the evaluation of HCV.16

The gold reference standard for determining hepatic histology is liver biopsy.17 However, liver biopsy is an invasive procedure that has been associated with complications,18, 19 and there are concerns regarding sampling error and inter- and intraobserver variation in interpretation.20–22 Consequently, there has been considerable interest in noninvasive markers to accurately asses the extent of hepatic involvement.23, 24 Several unique markers have been evaluated either alone or in combination in noninvasive models to predict hepatic fibrosis.25–29 These tests, however, are not widely available and can be costly. Although there are little comparative data,30, 31 simple models using readily available tests have gained popularity.32–35 Unfortunately, little information exists on the applicability of these models in coinfected patients.36–39 The aims of the current study were to develop a simple model of readily available tests to accurately predict hepatic fibrosis in a large cohort of HIV/HCV-coinfected patients and to differentiate those with mild to moderate fibrosis from those with advanced disease.


HCV, hepatitis C virus; PLT, platelet; AST, aspartate aminotransferase; ALT, alanine aminotransferase; HAART, highly active antiretroviral therapy; APRICOT, AIDS Pegasys Ribavirin International Coinfection Trial; AUROC, area under the receiver operating characteristic.

Patients and Methods


We performed a retrospective analysis of baseline histology and clinical parameters in HIV/HCV-coinfected patients entering the AIDS Pegasys Ribavirin International Coinfection Trial (APRICOT). The details of this trial have recently been published.12 Briefly, eligibility criteria included age greater than 18 years, infection with both HIV and HCV, elevated serum alanine aminotransferase (ALT) levels on 2 or more occasions within the previous 12 months with compensated liver disease, and a liver biopsy with liver histology consistent with chronic HCV. All patients were positive for antibodies to HIV (Amplicor HIV-1 Monitor Test, version 1.5; Roche Diagnostics), had detectable HCV RNA, and were HCV treatment–naïve. Patients with CD4 counts >200 cells/mm3 were eligible regardless of HIV RNA level, whereas those with CD4 counts of 100–199 cells/mm3 were eligible if the HIV RNA load was <5,000 copies/mL. All patients were required to be on a stable HAART regimen for at least 6 weeks prior to study entry with no expected changes for the first 8 weeks of therapy or not to be on any antiretroviral treatment at least 8 weeks prior to randomization. Patients were excluded if they had the following: active HIV-related opportunistic infection or cancer; an absolute neutrophil count below 1,500 cells/mm3; a platelet count below 70,000/mm3; a hemoglobin count below 11 g/dL for women and 12 g/dL for men; a serum creatinine level >1.5 times the upper limit of normal; concurrent hepatitis A or B infection; evidence of decompensated liver disease; severe psychiatric disease; clinically significant coexisting medical conditions that would preclude HCV therapy; or previous treatment with interferon or ribavirin. All patients were required to abstain from excessive alcohol consumption and/or drug or substance abuse within 6 months prior to entry. However, specific history of alcohol consumption was not collected. All patients gave written informed consent prior to entering the study.


Of the 868 patients enrolled in APRICOT, 832 were included in the analysis. Thirty-six patients were excluded because of lack of interpretable histology. Liver histology was determined via Ishak score40 as assessed by local pathologists (from 95 centers in 19 countries). Total histological activity index and fibrosis scores (0–6) were recorded. For analysis, patients were categorized into 3 fibrosis strata: mild (Ishak fibrosis score 0–1), moderate (Ishak fibrosis score 2–3), and advanced (Ishak fibrosis score 4–6). Significant fibrosis was defined as an Ishak fibrosis score of 4 or greater and cirrhosis as a score of 5 or 6. Patients were randomly divided and assigned to a training set (n = 555) or a validation set (n = 277). All laboratory information was obtained before the start of study treatment. Baseline liver biopsy was performed within 15 months of the start of study treatment and correlated to laboratory tests performed at the start of study treatment. To determine if central pathology reading and the variable time between laboratory studies used for study entry and liver biopsy affected our results, we also tested our model in both a validation set of the 283 individuals who underwent a follow-up liver biopsy 6 months after completion of therapy (laboratory tests were performed within 1 month of the biopsy with central pathology reading) and in 92 of the 277 patients in the validation set that had central pathology reading at baseline.

Statistical Analysis.

Data are expressed as mean ± SD unless otherwise noted. Statistical analysis was performed with SAS software (Statistical Analysis Software, version 8.2; SAS Institute, Cary, NC). Three possible outcomes were considered for the primary end point: little or mild fibrosis (score 0–1), moderate fibrosis (score 2–3), and advanced fibrosis (score 4–6). Logistic regression analysis was performed to develop models for prediction of fibrosis using the training data set and considering the following variables at baseline: age, sex, race (Caucasian vs. other), weight, body mass index, aspartate aminotransferase (AST), ALT, alkaline phosphatase, albumin, bilirubin, triglyceride, total cholesterol, glucose, total white blood cell count, platelet count, CD4 cell count, CD8 cell count, INR, partial thromboplastin time (PTT), HIV RNA titer, HCV RNA titer, HCV genotype, and HAART use including protease inhibitor (PI), nucleoside reverse-transcriptase inhibitor (NRTI), and nonnucleoside reverse-transcriptase inhibitor (NNRTI) use. Logarithmic transformation was considered for continuous variables to improve the normality of distribution. A two-sided P value of <.05 was considered statistically significant. Factors found significant on univariate analysis (P < .05) were then included in the stepwise multiple logistic regression analysis to identify independent factors associated with fibrosis. A simplified formula (fibrosis index) was then derived using the independent factors of the final logistic regression model and the proportions of the corresponding regression coefficients. The diagnostic value of the logistic regression model was assessed via the C-index (a generalization of the area under the receiver operating characteristic [AUROC] curves), where 1.0 indicates perfect discrimination and a score of 0.5 indicates a random prediction. The distribution of predicted probability of the 3 outcome levels as a function of the derived fibrosis index were then visualized. Because ROC curves are only able to handle binary outcomes, we used that method to assess the index's ability in differentiating advanced fibrosis (Ishak 4–6) from mild to moderate fibrosis (Ishak 0–3) and mild fibrosis (Ishak 0–1) from moderate to severe fibrosis (Ishak 2–6). The sensitivity, specificity, and positive and negative predictive values for the derived fibrosis index were calculated to determine the optimal cutoff values that would predict or exclude significant fibrosis. The fibrosis index derived from the training set was then applied to the validation set to test the predictive power of the selected model.


Patient Characteristics.

The characteristics of patients randomized to the training and validation sets are shown in Table 1. The groups were well matched. The mean values for the entire cohort were: age (40 ± 7 years), sex (80% male), race (78% Caucasian), body weight (73 ± 14 kg) and body mass index (25 ± 4), AST (56 ± 40 U/L), ALT (84 ± 65 U/L), alkaline phosphatase (80 ± 34 U/L), bilirubin (12.1 ± 7.6 μmol/L), albumin (42 ± 5 g/L), cholesterol (4.4 ± 1.1 mmol/L), triglycerides (2.0 ± 1.6 mmol/L), glucose (5.3 ± 2.0 mmol/L), platelet count (194 ± 66 × 109/L), CD4 (535 ± 272 cells/μL), CD4% (26 ± 10), CD8 (969 ± 444 cells/μL), HCV RNA (6.4 ± 0.8 log10 copies/mL), HCV genotypes 1 and 3 (62% and 27%), HIV RNA (9080 ± 45,773 copies/mL), proportion with undetectable HIV RNA (59%), and HAART use (nucleoside reverse-transcriptase inhibitor 84%, nonnucleoside reverse-transcriptase inhibitor 35%, and protease inhibitor 44%). Total histological activity index and distribution of fibrosis were also similar between the training and validation sets (Table 1). The median time between pretreatment liver biopsy and laboratory test at baseline was 98 days (interquartile range 45–220), and in 15% the difference was less than 1 month.

Table 1. Characteristics of Patients in the Cohort
 Training setValidation set
  • Abbreviations: BMI, body mass index; AP, alkaline phosphatase.

  • *

    Data presented as mean ± SD or proportions.

Age (yrs)*40 ± 740 ± 7
Gender (% male)8082
Race (% Caucasian)7782
Weight (kg)73 ± 1473 ± 13
BMI25 ± 424 ± 4
AST (U/L)56 ± 3955 ± 42
ALT (U/L)83 ± 6285 ± 71
AST/ALT ratio0.75 ± 0.310.74 ± 0.34
AP (U/L)80 ± 3481 ± 34
Albumin (g/L)42 ± 542 ± 4
Glucose (mmol/L)5.3 ± 1.65.3 ± 2.6
HCV RNA (log IU/mL)6.4 ± 0.86.3 ± 0.8
HCV genotype 1 (%)6163
 Training setValidation set
PLT (×109)194 ± 66194 ± 64
INR1.04 ± 0.151.04 ± 0.13
PTT (sec)33 ± 732 ± 6
CD4 (cells/μL)529 ± 255547 ± 303
CD8 (cells/μL)970 ± 425967 ± 482
HIV RNA (copies/mL)9031 ± 445909178 ± 48129
NRTI use (%)8386
NNRTI use (%)3437
PI use (%)4445
HAI8.1 ± 3.87.8 ± 3.7
% mild fibrosis3638
% moderate fibrosis4440
% advanced fibrosis2122

Factors Associated With Fibrosis.

In the training set (n = 555), increasing histological activity index correlated with increasing fibrosis strata (Table 2). The distribution of fibrosis scores among the 3 strata are shown in Table 2. Stages 0 and 1 were observed in 27% and 73%, respectively, of patients with mild disease; stages 2 and 3 were observed in 45% and 55%, respectively, of patients with moderate disease; and stages 4, 5, and 6 were observed in 33%, 35%, and 32%, respectively, of patients with advanced fibrosis. When comparing the 3 strata of fibrosis, univariate analysis revealed significant association with age, AST, ALT, AST/ALT ratio, alkaline phosphatase, albumin, INR, partial thromboplastin time, platelet count, CD4, bilirubin, cholesterol and AST/platelet ratio index (Table 3). No significant associations were observed with HCV RNA, HCV genotype, HIV RNA or proportion with undetectable HIV RNA, CD8 counts, HAART use, triglycerides, or glucose. Multiple logistic regression analysis identified 4 variables as independent predictors of fibrosis: age, AST, INR, and platelet count with a C-index of 0.708 (Table 4). However, only 505 of the 555 patients in the training set had all 4 variables available. INR values were missing in most of the remaining 50 patients. ALT was the next variable in the selection process, which was not included in the first model, because it did not quite reach significance (P = .0522). Therefore, a second model was investigated applicable to 553 patients that considered age, AST, platelet count, and ALT instead of INR. This model had a C-index of 0.704, which was very close to the original model (Table 4).

Table 2. Histology and Fibrosis Strata in the Training Set
 Mild (0–1)Moderate (2–3)Advanced (4–6)
  • *

    Data presented as mean ± SD or proportions.

HAI*4.7 ± 1.98.6 ± 2.512.4 ± 3.5
Fibrosis stage (%): 02700
Table 3. Comparison of Patients by Fibrosis in the Training Set (n = 555)
 Mild (0–1)Moderate (2–3)Advanced (4–6)P**
  • Abbreviation: AP, alkaline phosphatase.

  • *

    Data presented as mean ± SD or proportions.

  • **

    Univariate logistic regression analysis for association between severity of fibrosis (3 levels) and the factors listed.

  • #

    Based on natural logarithmic transformed values.

Age (yrs)*39 ± 840 ± 743 ± 8<.0001#
Gender (% male)797982.6802
Race (% Caucasian)797477.5919
Weight (kg)*72 ± 1374 ± 1573 ± 14.3462
BMI*24 ± 425 ± 525 ± 4.4235
AST (U/L)*44 ± 2961 ± 4368 ± 42<.0001#
ALT (U/L)*74 ± 5990 ± 6784 ± 52.0039#
AST/ALT ratio*0.68 ± 0.280.74 ± 0.280.88 ± 0.36<.0001
AP (U/L)*74 ± 2881 ± 3687 ± 37.0010#
Albumin (g/L)*42 ± 442 ± 440 ± 6.0001#
INR*1.02 ± 0.101.03 ± 0.101.12 ± 0.25<.0001
PTT (sec)*32 ± 832 ± 635 ± 8.0005#
PLT (×109)*216 ± 61194 ± 63158 ± 66<.0001#
Bilirubin (μmol/L)10.4 ± 5.911.8 ± 6.114.3 ± 10.2<.0001#
CD4 (cells/μL)562 ± 269513 ± 232505 ± 275.021#
Cholesterol (mmol/L)4.67 ± 1.134.34 ± 1.064.34 ± 1.08.0047#
APRI*0.66 ± 0.880.99 ± 0.891.46 ± 1.32<.0001
Table 4. Multiple Logistic Regression Model Identifying Independent Factors Associated With Fibrosis Staging (3 Levels) in the Training Set Including INR (n = 505, top) or the Final model replacing INR by ALT (n = 553, bottom)
Exploratory Factor (N = 505)Odds Ratio*Lower 95% Limit of Odds RatioUpper 95% Limit of Odds RatioSignificance (P)
  • Abbreviation: Ln, natural logarithm.

  • *

    The odds ratio is the multiplicative factor by which the odds are changed for a unit increase.

Ln Platelets (109/L)0.2330.1380.396<.0001
Ln AST (U/L)1.8391.3632.481<.0001
Ln Age (yrs)5.5442.20313.957.0003
Ln INR (ratio)16.6573.35282.770.0006
Exploratory Factor (N = 553)Odds Ratio*Lower 95% Limit of Odds RatioUpper 95% Limit of Odds RatioSignificance (P)
Ln Platelets (109/L)0.1890.1150.310<.0001
Ln AST (U/L)3.3292.0515.403<.0001
Ln Age (yrs)4.1181.66410.195.0022
Ln ALT (U/L)0.4930.3160.770.0019

Development of a Novel Model to Predict Fibrosis.

Based on the relationship of the 4 regression coefficients of the second model, the following simple index—the FIB-4 index—was derived:

FIB-4 = age (yr) × AST (U/L)

Platelet count (109/L) × (ALT (U/L))1/2

This index gave values 0.2 to 10. The univariate logistic regression model using FIB-4 as predictive factor instead of the 4 variables of the second model had a C-index of 0.703 in the training set, which was almost identical to the second model. The estimated probabilities of each fibrosis strata based on the FIB-4 index model are shown in Fig. 1. For example, a 45-year-old patient with an AST of 110 U/L, platelet count of 99 × 109/L, and an ALT of 100 U/L would have a FIB-4 index of 5. This would give an 80% probability of advanced fibrosis (Ishak 4–6), an 18% probability of moderate fibrosis (Ishak 2–3), and only a 3% probability of mild fibrosis (Ishak 0–1). In comparison, a 35-year-old with an AST of 75 U/L, platelet count of 263 × 109/L, and an ALT of 100 U/L would have a FIB-4 index of 1.0. This would correspond to probabilities of advanced, moderate, and mild fibrosis of 12%, 45%, and 43%, respectively (see Fig. 1 for additional explanation).

Figure 1.

Predictive probabilities for Ishak Score (0–1, 2–3, and 4–6) as a function of the FIB-4 index based on the logistic regression model developed with the training set (n = 555). Values for the FIB-4 index on the x axis are plotted against the probability on the y axis. For example 1, a patient with a FIB-4 index of 5 has an 80% chance of having advanced fibrosis (below the solid line) and an 18% chance of having moderate fibrosis (between the solid line and dashed line), and only a 3% chance of having minimal or no fibrosis (above the dashed line). Prob, probability.

Receiver operating characteristic curves for FIB-4 differentiating advanced fibrosis (Ishak 4–6) from mild to moderate fibrosis (Ishak 0–3) and mild fibrosis (Ishak 0–1) from moderate to severe fibrosis (Ishak 2–6) in the training set are shown in Fig. 2a. The AUROC for FIB-4 in differentiating Ishak 0–3 from 4–6 was higher (0.737) compared with its ability to differentiate Ishak 0–1 from 2–6 (0.711). Applying FIB-4 in the validation set gave slightly higher AUROC results for prediction of fibrosis 4–6 than in the training set (0.765) and a slightly lower result for differentiating Ishak 0–1 versus 2–6 (0.688) (Fig. 2b). Therefore, based on the AUROC for fibrosis 0–3 and 4–6, 2 cutoff points were chosen to predict either the absence of advanced fibrosis (Tables 5A–B). Using cutoff values of <1.45 and >3.25, 90% of patients in the validation set with a FIB-4 <1.45 would not have advanced fibrosis (i.e., negative predictive value = 90%), whereas a FIB-4 >3.25 would have a specificity of 97% and a positive predictive value of 65% for advanced fibrosis (Table 5A). Importantly, in both the training and validation sets, at least 70% of patients fell outside these ranges and could thus avoid biopsy with an overall accuracy of 86%. Therefore, of the 830 patients, only 80 would have been misclassified by this index and only 245 would have required a biopsy. To determine if the variable time between laboratories at study entry and biopsy effected the model, we applied both indices to the cohort of patients (validation set 2) who underwent a follow-up biopsy with central pathology reading (n = 283) at study end and found again a higher AUROC for FIB-4 (0.793) for differentiating Ishak 0–3 from 4–6 with excellent specificity compared with its ability to differentiate Ishak 0–1 from 2–6 (AUROC 0.695) (Tables 5A, 5B; Fig. 2C). Importantly, 224 of the 283 subjects (79%) had FIB-4 values outside <1.45 and >3.25 and in 87% of these cases prediction of Ishak stage 0–3 or 4–6 would have been correct. It is important to note that in these 283 subjects, the laboratory results were obtained close to the time of biopsy (median difference: 12 days; interquartile range: 5–35) (Table 5A–B). The model was also tested in 92 of the 277 patients in the validation set at baseline who had a central pathology reading. In this cohort, the AUROC for FIB-4 was 0.756 for prediction of stage 4–6 fibrosis as diagnosed via central pathology reading. This is very close to the AUROC value of 0.765 that was calculated for the validation set using local readings (n = 277). This corresponded to an AUROC of 0.695 for the FIB-4 index to predict patients with mild fibrosis (Ishak 0–1) versus those with moderate to severe fibrosis (Ishak 2–6) in the second validation set. For comparison, using 0.6 and 1.0 as FIB-4 cutoffs (Table 5B) a prediction regarding Ishak 0–1 versus 2–6 would be possible in 73%, 71%, and 65% of the total cohort, the first validation set, and the second validation set, respectively, with a correct prediction in 72%, 68%, and 77%, respectively, in each of these patient groups.

Figure 2.

(A) ROC plot for FIB-4 in differentiating advanced fibrosis (Ishak 4–6) from mild to moderate fibrosis (Ishak 0–3) and moderate to advanced fibrosis (Ishak 2–6) from mild fibrosis (Ishak 0–1) in the training set (n = 555). FIB-4 had an AUROC of 0.737 for the first prediction (solid line) and an AUROC of 0.711 (dashed line) for the second prediction. (B) ROC plot for FIB-4 in differentiating advanced fibrosis (Ishak 4–6) from mild to moderate fibrosis (Ishak 0–3) and moderate to advanced fibrosis (Ishak 2–6) from mild fibrosis (Ishak 0–1) in the validation set (n = 277). FIB-4 had an AUROC of 0.765 (solid line) for the first prediction and an AUROC of 0.688 (dashed line) for the second prediction. (C) ROC plot for FIB-4 in differentiating advanced fibrosis (Ishak 4–6) from mild to moderate fibrosis (Ishak 0–3) and moderate to advanced fibrosis (Ishak 2–6) from mild fibrosis (Ishak 0–1) in patients who underwent a follow-up biopsy (n = 283) with central pathology reading and laboratory tests within 1 month of biopsy. FIB-4 had an AUROC of 0.793 (solid line) for the first prediction and an AUROC of 0.695 (dashed line) for the second prediction.

Table 5A. Accuracy of the FIB-4 Index in Predicting Advanced Fibrosis in the Total Cohort and Validation Sets
Group (N)FIB-4Stage 0–3Stage 4–6SensitivitySpecificityPPVNPV
Total1 (830)≤ 1.454675866.7%71.2%38.0%89.0%
 > 1.45189116    
 < 3.2563413423.0%96.6%64.5%82.6%
 ≥ 3.252240    
Validation2 (277)≤ 1.451601870.0%73.7%42.4%89.9%
 > 1.455742    
 < 3.252104721.7%96.8%65.0%81.7%
 ≥ 3.25713    
Validation3 (283)≤ 1.451872854.8%84.6%50.0%87.0%
 > 1.453434    
 < 3.252205412.9%99.5%88.9%80.3%
 ≥ 3.2518    
Table 5B. Accuracy of the FIB-4 Index in Predicting Moderate/Advanced Fibrosis in the Total Cohort and Validation Sets
Group (N)FIB-4Stage 0–1Stage 2–6SensitivitySpecificityPPVNPV
  • Abbreviations: PPV, positive predictive value; NPV, negative predictive value.

  • 1

    Total cohort includes both the baseline training (n = 555) and the validation (n = 277) sets.

  • 2

    Validation set 1 included 277 patients at baseline.

  • 3

    Validation set 2 included 283 patients who underwent a follow-up biopsy.

Total1 (830)≤ 0.6714391.8%23.4%67.6%62.3%
 > 0.6232484    
 < 1.017716169.4%58.4%74.4%52.4%
 ≥ 1.0126366    
Validation2 (277)≤ 0.6251889.5%23.8%65.8%58.1%
 > 0.680154    
 < 1.0606164.5%57.1%71.2%49.6%
 ≥ 1.045111    
Validation3 (283)≤ 0.6162488.9%23.9%79.0%40.0%
 > 0.651192    
 < 1.0499157.9%73.1%87.4%35.0%
 ≥ 1.018125    


With improved survival in HIV3 and anti-HCV therapy,12–15 accurate assessment of liver disease severity in patients with HIV/HCV coinfection is important. Until recently, liver biopsy has been the gold standard.17 However, in addition to the risks of an invasive procedure and anxiety to both patient and physician,18, 19 this approach has been associated with sampling error.20–22 To avoid these pitfalls, several noninvasive models to predict fibrosis have been proposed.23, 24 Many of these use biochemical markers of fibrosis synthesis or degradation25–29 that may not be readily available and are not part of routine laboratory testing. Models using routine tests32, 34, 41 seem to perform as well in select populations.30, 31 Unfortunately, few have been applied to patients with HIV/HCV coinfection.36–39

The present study was conducted to develop a simple model using routine laboratory tests to predict liver fibrosis in a large cohort of coinfected patients. We found that age, serum AST, INR, and platelet count were independent predictors of fibrosis and that a model with age, AST, ALT, and platelet count had equivalent predictive power. These 4 variables were used to derive the novel FIB-4 index. Age has been used in the past—in the absence of knowledge of exact time of disease onset—as a surrogate marker of disease duration and has been associated with more advanced fibrosis.42 Platelet count is known to correlate with the amount of portal hypertension and advanced fibrosis.34, 42–44 Although all patients in the cohort had compensated liver disease, it was not unexpected that INR was associated with more advanced fibrosis, because it is directly related to hepatic synthetic function. Elevations in AST more than ALT have also been associated with more advanced fibrosis16, 23 and are in part related to delayed clearance of AST relative to ALT45 or to mitochondrial injury associated with more advanced fibrosis.46 Therefore, our observations that age, AST, and platelet count were associated with advanced fibrosis are consistent with previous studies.32, 34 Clinical factors that were not associated with advanced fibrosis were CD8 counts, HIV RNA, HCV RNA or genotype, and HAART use. The lack of significance of HIV level and HAART use are in agreement with several recent reports.7, 47, 48

Few studies have addressed noninvasive markers of hepatic fibrosis in HIV/HCV coinfected patients.37 Myers et al.36 studied the use of an index incorporating age, sex, α2 macroglobulin, apolipoprotein A1, haptoglobin, bilirubin, and γ-glutamyltranspeptidase in 130 coinfected patients from a single French center. They found an AUROC of this index to detect METAVIR F2–F4 fibrosis (septal fibrosis-cirrhosis) of 0.856. Using a score of 0–1, cutoff values of >0.6 had a positive predictive value of 86%, whereas a score of <0.2 had a negative predictive value of 93%. These thresholds could reduce the need for liver biopsy in this cohort by 55% with an accuracy of 89%. The use of the AST/platelet ratio index34 was assessed in 119 coinfected patients in a single site in the United States and revealed an AUROC of 0.82.37 An AST/platelet ratio index of <0.5 had a sensitivity of 87%, whereas an index of >1.5 had a specificity of 96% for predicting the presence or absence of bridging fibrosis or cirrhosis. The investigators report that using these cutoffs would avoid liver biopsy in 35% of patients. More recently, Kelleher et al.38 examined the role of fibrosis markers, including hyaluronic acid and YKL-40, in a cohort of 95 HIV/HCV-coinfected patients randomly selected from the Johns Hopkins HIV Clinic to differentiate patients with little to mild fibrosis (Ishak 0–2) from those with moderate or severe fibrosis (Ishak 3–6). They reported that patients with moderate to severe fibrosis had higher levels of hyaluronic acid and developed an index including serum hyaluronic acid, AST, and albumin that had an AUROC of 0.878 compared with 0.71 for AST/platelet ratio index. This index performed at the extreme cutoffs (<0.3 and >0.8). However, only 26 patients had Ishak fibrosis scores of >3 and the index performed poorly in those with scores between 0.3 and 0.8. Consequently, only 42% of the patients could be correctly classified, and a liver biopsy would still be required in the majority of patients. Our model, the FIB-4 index, was simple to use and performed well in differentiating Ishak 0–3 from 4–6 fibrosis. In our cohort, which included 21% with advanced fibrosis, 70% of patients had a FIB-4 index outside of the range (<1.45 and >3.25), resulting in a negative predictive value of 89% for excluding advanced fibrosis. Importantly, if used clinically, only 30% of patients fell within the intermediate range and would therefore require biopsy.

It is important to recognize that our model was developed on 3 levels of fibrosis (Ishak 0–1, 2–3, and 4–6), which makes it difficult to directly compare it with other models based on 2 levels of fibrosis.34, 36, 38, 41 A 3-level model of fibrosis was chosen over a 2-level model as in other studies because clinically, patients with a fibrosis score of 2 (portal fibrosis) are significantly different than those with scores of 5–6 (cirrhosis). This difference may explain why 70% of our patients had a FIB-4 index outside the cutoff range where it is applicable compared with 45%–65% in other models that were developed on 2 fibrosis levels.36, 38 Furthermore, our model performed better in differentiating Ishak 4–6 patients from Ishak 0–3 patients than it did differentiating Ishak 0–1 patients from Ishak 2–6 patients.

The present study has several strengths. First, the use of an international cohort of well-characterized patients from 95 centers and 19 countries who enrolled in APRICOT12 gave us the unique opportunity to develop the model in a large cohort of 832 patients that included a wide spectrum of liver disease, including 36% with mild fibrosis, 43% with moderate fibrosis, and 21% with advanced fibrosis. Given the size of the training set and its performance in both validation sets, our model appears to be robust and accurate. Our model includes readily available laboratory tests that are routine in the evaluation of HCV. Our findings of increased age, increased AST, and decreased platelet count correlate with increasing fibrosis are consistent with other studies32, 34 and support their inclusion in the model. Its relative simplicity, like AST/platelet ratio index, makes it user-friendly. Because the cutoff values chosen included 70% of the cohort, it could avoid liver biopsy in the majority of patients with high accuracy. Furthermore, the FIB-4 index performed at least as well or better than a more complicated model using tests not readily available in routine clinical practice.36, 38

We acknowledge several limitations to our analysis. First, APRICOT was not designed to develop a model to predict histology. Therefore, the retrospective nature and potential selection bias of patients who were accepted and enrolled in APRICOT cannot be overemphasized, and our results may not be applicable to all HIV/HCV-coinfected patients. We did not include patients with hepatic decompensation. However, diagnosis of cirrhosis in these patients can be made clinically and often does not require liver biopsy. We also recognize that a significant proportion of coinfected patients may have normal ALT levels,47 which was an exclusion criterion for entry into APRICOT. Therefore, our model may not apply to these populations. We also used liver histology as the reference standard, which may complicate the correlation of biochemical tests and fibrosis. However, this is true of all studies in which biochemical tests are compared with histology.

The time interval between liver biopsy and laboratory studies and the lack of central pathology reading for the entire cohort may have influenced our results. However, when we analyzed the cohort of patients (n = 283) who had central pathology reading and laboratory testing performed at the time of the follow-up liver biopsy, the index performed equally well (AUROC = 0.79), minimizing these potential confounding variables. To support our results, the model also performed well in the 92 patients in the baseline validation set that had a central pathology reading (AUROC = 0.756). Although we did not take into account biopsy length and fragmentation, a recent study in patients with advanced fibrosis found that these variables did not affect the performance of their model.41 We also did not accurately assess previous alcohol consumption, which is a known factor in liver disease progression9 and could have potentially influenced the AST values that were included in the model. The large numbers of patients included in the analysis should minimize these potential confounding factors. Importantly, approximately 30% of the patients fell between the 2 cutoff values of the FIB-4 index identified. In these patients with an indeterminate FIB-4 index, liver biopsy would still be needed. Nevertheless, the proportion in our study with an indeterminate value is less than the 45%–65% reported by others.36, 38 Finally, although our index is relatively simple and includes standard laboratory tests, it does require a calculation that includes a square root in the denominator. However, the calculation can easily be performed at the bedside with any standard calculator.

In conclusion, the FIB-4, a simple novel index composed of readily available routine laboratory tests, can accurately differentiate mild to moderate fibrosis from bridging fibrosis and cirrhosis in patients coinfected with HIV/HCV. In addition, its applicability to the majority of coinfected patients can reduce the need for liver biopsy in up to 70% of individuals with an overall accuracy of 87%. If additional studies in patients with chronic HCV support our findings, the FIB-4 index could be used to accurately identify patients with significant fibrosis who might benefit from anti-HCV therapy and—just as importantly—patients with mild disease in whom therapy could be deferred.