Abstract

Assessment of liver fibrosis is critical for successful individualized disease management in

persons with chronic hepatitis B (CHB) or chronic hepatitis C (CHC). We expanded and
validated serum marker indices to provide accurate, reproducible and easily applied methods of
fibrosis assessment. Liver biopsy results from over 284 CHB and 2304 CHC patients in the
Chronic Hepatitis Cohort Study ('CHeCS') were mapped to a F0F4 equivalent scale. APRI and
FIB-4 scores within a 6-month window of biopsy were mapped to the same scale. A novel
algorithm was applied to derive and validate optimal cut-offs for differentiating fibrosis levels.
For the prediction of advanced fibrosis and cirrhosis, the FIB-4 score outperformed the other
serum marker indices in the CHC cohort and was similar to APRI in the CHB cohort. The area
under the receiver operating characteristic curves (AUROC) for FIB-4 in differentiating F3F4
from F0F2 was 0.86 (95% CI: 0.800.92) for CHB and 0.83 (95% CI: 0.810.85) for CHC. The
suggested cut-offs based on FIB-4 model produced high positive predictive values [CHB: 90.0%
for F0F2, 100.0% for cirrhosis (F4); CHC: 89.7% for F0F2; 82.9% for cirrhosis (F4)]. In this
large observational cohort, FIB-4 predicted the upper and lower end of liver fibrosis stage
(cirrhosis and F0F2, respectively) with a high degree of accuracy in both CHB and CHC
patients.
Introduction
Accurate assessment of liver fibrosis is essential for successful individualized disease
management for persons with chronic hepatitis B (CHB) or chronic hepatitis C (CHC). However,
the lack of accurate, reproducible and easily applied methods for fibrosis assessment remains a
major limitation in clinical practice.[1] Liver biopsy is the gold standard for staging liver fibrosis
in these patients; however, liver biopsy is invasive, expensive and can be subject to sampling
error and variability in pathological interpretation. Transient elastography (FibroScan) has
recently been approved in the United States and has been shown to be superior to FIB-4 in
identifying later-stage (F3F4) hepatic fibrosis and cirrhosis.[2,3] However, compared with serum
markers, it is expensive, less useful in diagnosing low-to-moderate fibrosis and less readily
available. In addition, there are concerns that the applicability and reliability of elastography are
not as good as that for serum fibrosis markers.[4] Various serum markers have been proposed,
although they have shown wide variability and have not been widely applied in CHB.[5,6] Both
the aminotransferase-to-platelet ratio index (APRI)[711] and the FIB-4,[5,1214] which consists of
ALT level, AST level, platelet count and age, have been studied in several cohorts with area
under the receiver operating characteristic curves (AUROC) in the 7080% range for predicting
cirrhosis or advanced fibrosis. However, the studies were conducted in relatively small, racially
homogenous or single-centre patient cohorts under somewhat controlled conditions (e.g. data
collected through clinical trials, laboratory data collected at same time as biopsy), making the
results difficult to generalize to broader patient populations in real-world clinical settings.[6,7]
Recently, Holmberg et al.[4] found that FIB-4 was superior to APRI and much superior to a
simple AST/ALT ratio at distinguishing advanced fibrosis from mild-to-moderate fibrosis in
CHC patients. The study was part of the Chronic Hepatitis Cohort Study (CHeCS), a
comparative effectiveness research[15,16] study established to assess the implications and impact of

chronic hepatitis B and C on patients in the United States. All the data collected for CHeCS are
derived from routine care in large healthcare systems.
The present study is an extension and validation of their recent work. To validate for both CHB
and CHC, we used a training data set to build a multiple logistic regression model and a
validation data set to assess the model's predictive ability. In addition, while the previous work
focused on differentiating between mildmoderate and advanced fibrosis, we have extended the
research to determine optimal cut-off values for distinguishing between F0F2, F3 (advanced
fibrosis) and F4 (cirrhosis).
Materials and Methods

The CheCS investigation follows the guidelines of the US Department of Health and Human
Services regarding the protection of human subjects. The protocol was approved and is renewed
annually by the institutional review board at each participating site. The CheCS project's
methods have been previously summarized.[17]
Patients

The Chronic Hepatitis Cohort Study (CHeCS) included adults aged 18 years from the four
participating healthcare organizations (Geisinger Health System, Danville, PA; Henry Ford
Health System, Detroit MI; Kaiser Permanente Northwest, Portland, OR; Kaiser Permanente
Honolulu, HI, USA) and has rolling admissions and follow-up from patients' routine care with
their clinicians.[17] Patients who met electronic identification criteria for CHC or CHB infection
were included. For this study, we selected patients who had available laboratory data within a 6month interval of a liver biopsy.
Laboratory Data

All patients' laboratory data (alanine aminotransferase [ALT], aspartate aminotransferase [AST],
platelet count) were collected through electronic medical records. If multiple laboratory values
were available, the results closest to the time of biopsy were used. APRI, FIB-4 and, for purposes
of comparison, AST/ALT ratio, were calculated when the laboratory assessments were within 7
days of each other and within 6 months of the biopsy. The FIB-4 and other indices were
calculated using the following formulas:

(* where ULN = upper limit of normal for that laboratory)

Histological Liver Assessment

Fibrosis stage was collected by abstraction from liver biopsy reports. Fibrosis scores from
different scoring systems (IASL, Batts Ludwig, Metavir, Ishak, Knodell, Scheuer) were mapped
to a F0F4 equivalency scale: F0, no fibrosis; F1, portal fibrosis without septa; F2, portal fibrosis
with few septa; F3, numerous septa without cirrhosis; and F4, cirrhosis (Table 1). If a patient had
a liver transplant, the laboratory results and biopsies after the transplant were excluded from this
analysis. If the patient had more than one biopsy, the earliest biopsy with the highest fibrosis
stage and available laboratory results was used for this analysis.
Statistical Analysis

We first validated predefined serum markers, then developed and validated cut-offs of the serum
markers for classification of cirrhosis (F4) or advanced fibrosis (F3F4). The serum markers of
interest were as follows: APRI, FIB-4 and ALT/AST ratio, as defined in the previous section. The
endpoints of interest were the presence of advanced fibrosis (F34 vs F02) and the presence of
cirrhosis (F4 vs F03).
Data were randomly divided into a training data set and a validation data set using a 2:1 ratio.
The training data set was used to build a multiple logistic regression model. The validation data
set was used to assess the model's predictive ability, measured by the ROC curves and AUROC.
The predictive abilities (AUROCs) of the three serum markers were compared using a
nonparametric approach proposed by DeLong et al..[18] The final serum-based marker was
selected with the highest AUROC based on the validation set. The marker was considered to be a
valid marker if the AUROC was >0.80. To make a useful model for clinical practice, we were
interested in determining cut points for the non-invasive markers to classify patients into three
classes: nonadvanced fibrosis (F0F2, as Class 1), advanced fibrosis (F3, as Class 2) and
cirrhosis (F4, Class 3). Inspired by Nakas et al.,[19] we extended the cut point selection approach
from two classes to three classes. That is, similar to the generalized Youden index (J3) that has
been studied by Nakas et al., we selected the two optimal cut points, cut1 and cut2, by balancing
between positive predictive values (PPV) and Youden index. The PPV was defined as
PPV1+PPV3, where PPV1 stands for the probability that a patient belongs to Class 1 (F0F2) if
his/her risk score was less than cut1, and PPV3 stands for the probability that a patient belongs to
Class 3 (F4) if the patient's risk score was greater than cut2. Cut1 and cut2 were derived from the
training data. The performance of cut1 and cut2 was further tested on the validation data
Characteristics of CHB and CHC Patients in the Training Set

To date (March, 2013), a total of 2738 patients had confirmed chronic CHB and 10 476 patients
had confirmed chronic CHC. After excluding 111 patients with CHB/CHC co-infection, there
were 907 CHB and 7767 CHC patients with a biopsy that classified fibrosis stage. A total of 284
CHB patients had both biopsy and available laboratory (FIB-4 and APRI) data within 6 months
of the biopsy and within 7 days of each other. A total of 196 CHB patients were included in the
training data for model building, and the remaining 88 patients were used for the validation data
set. Similarly, 2304 CHC patients had both biopsy and serum marker data. Of these, 1529 and
775 CHC patients were split for training and validation, respectively.
Characteristics and mean serum marker scores of patients in the CHB and CHC training data sets
are presented in Table 2. Of the patients with CHB, 61 (31%) had advanced fibrosis (F3 and
above) by liver biopsy. Among those, 39 (20%) had confirmed cirrhosis through biopsy. patients'
characteristics were similar in the validation data set.
Among the patients with CHC, 1529 were randomly selected for the training data set. Of those,
610 (40%) had advanced fibrosis and 348 (23%) had cirrhosis by liver biopsy. Patient's
characteristics were similar in the validation data set.
FIB4 and Other Indices
We compared AUROCs of the FIB-4 index with those of the other indices (Fig. 1) for the
classification of advanced fibrosis and cirrhosis, respectively. Based on the training data, both
FIB-4 index and APRI showed significantly higher predictive values of advanced fibrosis when
compared with AST/ALT ratio for both CHB and CHC (P < 0.0001 for both). The AUROC for
FIB-4 in differentiating F3F4 from F0F2 was 0.86 (95% CI: 0.800.92) for CHB and 0.83
(95% CI: 0.810.85) for CHC. The AUROC for APRI was 0.86 (95% CI: 0.810.92) for CHB
and 0.81 (95% CI: 0.780.83) for CHC. Applying FIB-4 on the validation data set gave slightly
lower AUROC results for the prediction of advanced fibrosis than in the training sets. The
AUROC of FIB-4 for CHB was 0.71 but with a wider 95% confidence interval of 0.580.85, as a
much smaller size of validation data set was available for CHB. The AUROC of FIB-4 was 0.85
(95% CI: 0.830.88) for CHC. The significant improvement in AUROC from AST/ALT ratio to
both FIB-4 and APRI was also seen in the validation data.

(Enlarge Image)
Figure 1.
Comparison of laboratory markers (FIB-4, APRI and AST/ALT) in predicting advanced fibrosis
(F34) or cirrhosis (F4) in CHB and CHC cohorts. AUROCs are presented in training and

validation data sets separately. *The AUC of each marker was compared against AUC of FIB4.
P-values are presented.
Receiver operating curves for differentiating cirrhosis (F4) from noncirrhosis (F0F3) using FIB4, APRI and AST/ALT in the training and validation sets yielded similar conclusions. APRI had
AUROC comparable to FIB-4 score for the CHB training data set, while both APRI and FIB-4
showed significantly higher AUROCs than AST/ALT ratio (P-values <0.0001). The P-values for
comparison of AUROCs between APRI and FIB-4 were 0.3882 and 0.1331 for CHB training and
validation data sets, respectively. Unlike the CHB cohort, FIB-4 index in the CHC cohort
showed significantly higher predictive values when compared with APRI. The significant
difference was observed in both training and validation data with both P-values <0.0001.
Cut-off Values for Predicting Fibrosis and Cirrhosis Using FIB-4 in CHB and CHC
Based on the AUROC analysis in the previous section, FIB-4 had the best overall utility for
prediction of advanced fibrosis and cirrhosis in a chronic viral hepatitis population compared
with the other two markers, as FIB-4 score outperformed the other serum marker indices in the
CHC cohort, and was superior to AST/ALT ratio and similar to APRI in the CHC cohort. We
next derived optimal cut-off values of FIB-4 for distinguishing the lower end of liver stage (F0
F2) and upper end of liver stage (F4, cirrhosis) for both CHB and CHC, which are presented in
Table 3.
CHB
Using cut-off values of 1.58 and 5.17, 120 (61%) CHB patients with FIB-4 1.58 in the training
data were classified as F0F2. Among these, 108 (90.0%) had been staged within F0F2 by
biopsy. In other words, patients whose FIB-4 was <1.58 could be staged without liver biopsy
with accuracy of 90.0% in our data. Fourteen patients had FIB-4 > 5.17, and all had cirrhosis
confirmed by liver biopsy. Thus, these patients could also be staged without liver biopsy with
100% accuracy in our data. Patients with FIB-4 between 1.58 and 5.17 were more difficult to
stage and are shown in the grey zone in Table 3. For these patients, liver stage could not be
determined efficiently by the FIB-4 index only. When the cut points were applied to the
validation data, 52 patients (58.0%) had low FIB-4 (1.58), 84.6% of whom were between F0
and F2. Six patients had FIB-4 > 5.17, 5 (83.3%) of whom actually had cirrhosis confirmed by
liver biopsy. Therefore, a total of 58 patients in the validation data set could be determined by
FIB-4 score with an overall accuracy of 84.5%.
CHC
The cut-offs yielded similar prediction accuracies in both training and validation data sets.
Focusing on the validation data, there were 233 (30.1%) CHC patients whose FIB-4 scores were
1.21. Among the 233 patients, 216 (92.7%) were at the stage of F2 and lower, and 78 (10.1%)
patients had FIB-4 score higher than 5.88, 64 (82.5%) of whom had cirrhosis confirmed by
biopsy. Overall, 463 (59.8%) patients had FIB-4 scores between 1.21 and 5.88, and are shown in
the grey zone (Table 3).

Discussion

In this study, we validated the diagnostic value of FIB-4 and compared FIB-4 with two other
biomarker indices, AST/ALT ratio and APRI. In the training and validation sets, when we
compared advanced fibrosis with mild-to-moderate fibrosis (F34 vs F02) and cirrhosis with
lower fibrosis (F4 vs F03), we found FIB-4 and APRI to be superior to AST/ALT in both CHC
and CHB cohorts. In addition, FIB-4 was a slightly better predictor in the CHC populations but
not in the CHB populations.
The present study demonstrates that FIB-4 can predict the upper and lower end of liver fibrosis
stage (cirrhosis and F0F2, respectively) with a high degree of accuracy in both CHB and CHC
patients. We found that FIB-4 scores 1.58 for CHB and 1.21 for CHC accurately predicted
mild-to-moderate fibrosis (F0F2) in both the training and validation sets. In addition, FIB-4
scores >5.17 for CHB and >5.88 for CHC accurately predicted cirrhosis (F4). Thus, the use of
FIB-4 could obviate the need for liver biopsy in uncomplicated earlier-stage patients.
Holmberg et al.[4] have discussed several advantages of using FIB-4 and other serum fibrosis
markers to initially stage and follow CHC patients. In general, the parameters of these serum
markers are easy, rapid and inexpensive to measure. This can be an important factor where more
complicated and expensive tests, such as liver biopsy and transient elastography, are less
available. More importantly, these markers can be measured in routine clinical care, which
makes it easy to monitor patients' disease status over time. As Holmberg et al.[4,17] pointed out, a
non-invasive serum fibrosis marker could avert the challenges in longitudinal monitoring of
patients with CHC. In addition, it is possible that a fibrosis marker could be used in conjunction
with liver biopsy and transient elastography for successful, individualized and longitudinal
disease management because concurring results could provide a higher confidence of predicted
fibrosis stage.[20]
There are many other advantages to using FIB-4 in a clinical setting. FIB-4 is based on
minimally invasive parameters collected during routine care and thus can be used in primary care
settings for ongoing monitoring of CHB and CHC patients at regular intervals for indication that
liver fibrosis progression requires further clinical investigation or potential treatment. FIB-4 can
also be used when patients are not suitable candidates for FibroScan and cannot or will not
undergo biopsy. Use of this parameter can further the clinical research and assessment of large
cohorts of viral hepatitis patients for which there is otherwise a lack of means by which to assess
liver fibrosis status.
Several studies on prediction of liver disease stage using fibrosis markers among CHB or CHC
patients have been published in the past few years. Our study has several unique features. First,
the training and validation cohorts used in this analysis were derived from the CHeCS study,
which to our knowledge is the first U.S. cohort study to characterize a general population of over
12 000 patients with chronic CHB and CHC infections with both EHR and chart abstraction data

collection. In addition, we were able to validate the data. The predictability of different models
was consistent between training and validation data set for both CHB and CHC, suggesting that
the models were robust. It greatly improves our confidence in model predictions of liver fibrosis
stage using FIB-4. The CHeCS study is still ongoing; further validation of FIB-4 and suggested
cut-offs on an additional subsequent study cohort are needed.
Secondly, we suggested cut points for classification between F0F2, F3 (advanced fibrosis) and
F4 (cirrhosis). As opposed to previous studies in which the cut points were selected by
dichotomizing the five liver stages (either F3F4 vs. F0F2 or F4 vs. F0F3), our cut points were
optimized by simultaneously distinguishing the three classes. Our suggested cut points yielded
very high prediction accuracies for the lower and upper ends of the liver stages.
Finally, although laboratory results were drawn within 6 months of liver biopsy, our study still
confirmed the high predictability of FIB-4 for liver fibrosis stage. In a sensitivity analysis
conducted to evaluate the robustness of our findings, laboratory data were collected within a
narrower window of liver biopsy (3 months) and FIB-4 scores were constructed based on these
laboratory data. The AUROCs of FIB-4 for predicting advanced fibrosis in the sensitivity
analysis were 0.82 (95% CI, 0.760.89) and 0.84 (95% CI, 0.840.87) in CHB and CHC,
respectively (Table S1
http://onlinelibrary.wiley.com/store/10.1111/jvh.12224/asset/supinfo/jvh12224-sup-0001TableS1-S2.docx?v=1&s=56fb4a8d4d4b43cacc651177ecb3e51d5216cf9f); the AUROCs for
cirrhosis were 0.89 for CHB and 0.87 for CHC. Applying the Table 3 cut points to this data, we
got prediction accuracy of 87.9% for FIB-4 1.58 and 95.0% for FIB-4 > 5.17 for CHB (Table
S2 http://onlinelibrary.wiley.com/store/10.1111/jvh.12224/asset/supinfo/jvh12224-sup-0001TableS1-S2.docx?v=1&s=56fb4a8d4d4b43cacc651177ecb3e51d5216cf9f). The prediction
accuracy was 90.9% for FIB-4 1.21 and 84.2% for FIB-4 > 5.88 for CHC. Thus, our results are
robust using FIB-4 derived from within a 6-month window or 3-month window of liver biopsy.
The results were validated in a random subset of the samples.
Limitations to this analysis include variability in these serum markers at various stages of liver
disease (fibrosis). Even the 'gold standard', liver biopsy, is subject to sampling error and interand intra-observer discrepancies in reading biopsy specimens of 1020%;[4,21,22] thus, cirrhosis
could potentially be underdiagnosed up to 20% of the time. However, presumably, roughly equal
numbers of patients were mis-categorized by FIB-4 to a higher or lower stage. In addition,
although FIB-4 was validated using the internal independent validation set, further evaluation
using an external cohort with similar patient characteristics may be useful.
Because FIB-4 is based on parameters that are readily accessible to non-hepatologist clinicians
who do not perform liver biopsies but who may wish to care for patients with chronic CHB and
CHC infection, this study demonstrates that use of FIB-4 can facilitate monitoring of CHB and
CHC patients. Apart from use in a clinical setting, the markers can be easily obtained in

longitudinal studies for studying disease progression and treatment effects in a racially diverse
large population cohort outside of clinical trials.