Page 25 - EASL POSTGRADUATE COURSE
P. 25
underlying prevalence of fibrosis may influence the AUROC – the so-called spectrum effect and thus it is
difficult to directly compare the accuracy of tests performed in different populations.
Predictive values
Predictive values are dependent upon the underlying fibrosis prevalence as well as sensitivity or specificity.
Thus a test may be highly specific for the diagnosis of cirrhosis, but have a low positive predictive value
(PPV) if the underlying prevalence (or pre-test probability) is very low. It is therefore important to
realize that non-invasive test performance will vary according to the setting (and underlying fibrosis
prevalence) in which they are used. Current, non-invasive tests that have been developed in tertiary
hospital settings have poor agreement for the prediction of advanced fibrosis in a general population
setting, although appear reliable in excluding advanced fibrosis [1].
Clinical prediction of fibrosis
Cross-sectional studies of NAFLD patients undergoing clinically indicated liver biopsies show increasing
age and diabetes (odds ratio ~4) to be consistently associated with fibrosis [2].The association with obesity
and hypertension is more variable between studies, whereas dyslipidemia appears not to have a strong
independent association with fibrosis. Individually these clinical factors lack sufficient accuracy to reliably
predict advanced fibrosis or cirrhosis; however they may be useful to exclude advanced fibrosis, with 0/144
patients in one study having stage 3-4 fibrosis in the absence of diabetes, obesity and age ≥45 years [3].
Serum based tests
Simple, easily calculated serum-based tests include the AST/ALT ratio and the BARD algorithm. Notably,
aminotransaminase levels alone are poorly predictive of fibrosis stage and tend to decline over time as
fibrosis progresses. An AST/ALT ≥1.0 has modest accuracy (AUROC 0.66-0.83), reasonable specificity
(84-92%), is weakly predictive of advanced fibrosis (stage 3-4) with PPVs of 26-55%, and has reported
negative predictive value’s (NPV) of 81-95%. The BARD score is calculated from the cumulative total of
BMI ≥28 kg/m2 (1 point), AST/ALT ratio ≥0.8 (2 points) and diabetes (1 point). Its simplicity is attractive
and its strength is excluding advanced fibrosis, with a score of 0 or 1 having reasonably high NPVs (81-
97%), though poor specificity (44-79%) and low PPVs (22-46%).The AST/ALT ratio (and subsequently
BARD score) appear to be less accurate in patients with diabetes; the odds ratio of having advanced
fibrosis with an AST/ALT ≥0.8 is 16 and 4 in non-diabetics and diabetics, respectively [4].
More complicated algorithms that combine multiple routinely-available biochemical and clinical variables
have greater accuracy for predicting advanced fibrosis (Table 1). Algorithms that have undergone the
greatest external validation, include the NAFLD Fibrosis Score (NFS; www.nafldscore.com) and FIB-4,
which were developed to predict advanced hepatic fibrosis (defined as bridging fibrosis or cirrhosis).
Comparative studies have shown these scores to be more accurate than less complicated algorithms
such as BARD and the APRI [5, 6]. Both scores predict liver-related and overall mortality in NAFLD
patients, although the NFS may be more discriminatory [7]. The NFS and FIB-4 have similar accuracy
with AUROC for advanced fibrosis of 0.82-0.88 and 0.8, respectively. Both scores also have upper and
lower cut-off points in order to maximize their diagnostic accuracy; however, this means indeterminate
results will occur in one quarter to a third of patients. Both tests exclude advanced fibrosis with scores
below the lower cut-off having NPVs of 88-90%. Scores above the upper cut-off are highly specific (96-
98%), translating to PPVs for advanced fibrosis of 80-82%. Notably however, scores that incorporate
AST and ALT become less accurate as aminotransaminase values increase.
Other tests include NAFLD Fibrometer and Fibrotest, which are proprietary. Independent validation
of the Fibrotest has shown AUROC values of 0.64-0.82 for the prediction of portal based fibrosis
(F2+) and AUROC of 0.86-0.89 for cirrhosis [5, 8]. Among morbidly obese individuals (BMI >35 kg/
m2), Fibrotest at a cut-off of 0.48 is poorly sensitive (8%) for significant fibrosis (METAVIR F2-4),
but has excellent specificity (99.6%), translating to a NPV and a PPV of 91% and 67%, respectively,
when the prevalence of significant fibrosis is 10% [9]. Fibrotest is also predictive of overall, liver and
cardiovascular mortality in patients at risk of NAFLD, namely those with diabetes and dyslipidemia.
The International Liver Congress™ 2015 • Vienna, Austria • April 22–23, 2015 25
difficult to directly compare the accuracy of tests performed in different populations.
Predictive values
Predictive values are dependent upon the underlying fibrosis prevalence as well as sensitivity or specificity.
Thus a test may be highly specific for the diagnosis of cirrhosis, but have a low positive predictive value
(PPV) if the underlying prevalence (or pre-test probability) is very low. It is therefore important to
realize that non-invasive test performance will vary according to the setting (and underlying fibrosis
prevalence) in which they are used. Current, non-invasive tests that have been developed in tertiary
hospital settings have poor agreement for the prediction of advanced fibrosis in a general population
setting, although appear reliable in excluding advanced fibrosis [1].
Clinical prediction of fibrosis
Cross-sectional studies of NAFLD patients undergoing clinically indicated liver biopsies show increasing
age and diabetes (odds ratio ~4) to be consistently associated with fibrosis [2].The association with obesity
and hypertension is more variable between studies, whereas dyslipidemia appears not to have a strong
independent association with fibrosis. Individually these clinical factors lack sufficient accuracy to reliably
predict advanced fibrosis or cirrhosis; however they may be useful to exclude advanced fibrosis, with 0/144
patients in one study having stage 3-4 fibrosis in the absence of diabetes, obesity and age ≥45 years [3].
Serum based tests
Simple, easily calculated serum-based tests include the AST/ALT ratio and the BARD algorithm. Notably,
aminotransaminase levels alone are poorly predictive of fibrosis stage and tend to decline over time as
fibrosis progresses. An AST/ALT ≥1.0 has modest accuracy (AUROC 0.66-0.83), reasonable specificity
(84-92%), is weakly predictive of advanced fibrosis (stage 3-4) with PPVs of 26-55%, and has reported
negative predictive value’s (NPV) of 81-95%. The BARD score is calculated from the cumulative total of
BMI ≥28 kg/m2 (1 point), AST/ALT ratio ≥0.8 (2 points) and diabetes (1 point). Its simplicity is attractive
and its strength is excluding advanced fibrosis, with a score of 0 or 1 having reasonably high NPVs (81-
97%), though poor specificity (44-79%) and low PPVs (22-46%).The AST/ALT ratio (and subsequently
BARD score) appear to be less accurate in patients with diabetes; the odds ratio of having advanced
fibrosis with an AST/ALT ≥0.8 is 16 and 4 in non-diabetics and diabetics, respectively [4].
More complicated algorithms that combine multiple routinely-available biochemical and clinical variables
have greater accuracy for predicting advanced fibrosis (Table 1). Algorithms that have undergone the
greatest external validation, include the NAFLD Fibrosis Score (NFS; www.nafldscore.com) and FIB-4,
which were developed to predict advanced hepatic fibrosis (defined as bridging fibrosis or cirrhosis).
Comparative studies have shown these scores to be more accurate than less complicated algorithms
such as BARD and the APRI [5, 6]. Both scores predict liver-related and overall mortality in NAFLD
patients, although the NFS may be more discriminatory [7]. The NFS and FIB-4 have similar accuracy
with AUROC for advanced fibrosis of 0.82-0.88 and 0.8, respectively. Both scores also have upper and
lower cut-off points in order to maximize their diagnostic accuracy; however, this means indeterminate
results will occur in one quarter to a third of patients. Both tests exclude advanced fibrosis with scores
below the lower cut-off having NPVs of 88-90%. Scores above the upper cut-off are highly specific (96-
98%), translating to PPVs for advanced fibrosis of 80-82%. Notably however, scores that incorporate
AST and ALT become less accurate as aminotransaminase values increase.
Other tests include NAFLD Fibrometer and Fibrotest, which are proprietary. Independent validation
of the Fibrotest has shown AUROC values of 0.64-0.82 for the prediction of portal based fibrosis
(F2+) and AUROC of 0.86-0.89 for cirrhosis [5, 8]. Among morbidly obese individuals (BMI >35 kg/
m2), Fibrotest at a cut-off of 0.48 is poorly sensitive (8%) for significant fibrosis (METAVIR F2-4),
but has excellent specificity (99.6%), translating to a NPV and a PPV of 91% and 67%, respectively,
when the prevalence of significant fibrosis is 10% [9]. Fibrotest is also predictive of overall, liver and
cardiovascular mortality in patients at risk of NAFLD, namely those with diabetes and dyslipidemia.
The International Liver Congress™ 2015 • Vienna, Austria • April 22–23, 2015 25
