Meta-analysis: Natural history of non-alcoholic fatty liver disease (NAFLD) and diagnostic accuracy of non-invasive tests for liver disease severity

Abstract

Background. NAFLD ranges from simple steatosis (SS) to non-alcoholic steatohepatitis (NASH). The natural history of NAFLD and the optimal strategy to identify subjects with progressive liver disease are unclear.

Objectives. To assess the evidence in: (1) natural history of NAFLD; and (2) non-invasive methods to differentiate NAFLD histological subtypes.

Design and setting. Among 4185 articles published on MEDLINE, Cochrane Library, EMBASE, Pubmed, national and International meeting abstracts through July 2010, 40 articles assessing the natural history of NAFLD and 32 articles evaluating the diagnostic accuracy of non-invasive tests against liver biopsy (LB) were included.

Measurements. Two reviewers retrieved articles and evaluated study quality by appropriate scores. Main outcomes were pooled using random- or fixed-effects models.

Results. NAFLD has an increased overall mortality (OR: 1.57, 95% CI: 1.18–2.10), deriving from liver-related and cardiovascular disease, and a 2-fold risk of diabetes. Compared to SS, NASH has a higher liver-related (OR for NASH: 5.71, 2.31–14.13; OR for NASH with advanced fibrosis: 10.06, 4.35–23.25), but not cardiovascular mortality (OR: 0.91, 0.42–1.98). Three non-invasive methods received independent validation: pooled AUROC, sensitivity and specificity of cytokeratin-18 for NASH are 0.82 (0.78–0.88), 0.78 (0.64–0.92), 0.87 (0.77–0.98). For NASH with advanced fibrosis, pooled AUROC, sensitivity and specificity of NAFLD fibrosis score and Fibroscan are 0.85 (0.80–0.93), 0.90 (0.82–0.99), 0.97 (0.94–0.99) and 0.94 (0.90–0.99), 0.94 (0.88–0.99) and 0.95 (0.89–0.99).

Conclusions. NAFLD warrants screening for cardio-metabolic risk and for progressive liver disease. The combination of three noninvasive tests with LB may optimally individuate patients with NASH, with or without advanced fibrosis.

Key words::

• Fatty liver
• liver disease severity
• management
• NAFLD
• NASH
• non-invasive
• steatosis

Abbreviations
AASLD	=	American Association for the Study of Liver Disease
ADA	=	American Diabetes Association
AGA	=	American Gastroenterological Association
ALT	=	alanine aminotransferase
AST	=	aspartate aminotransferase
AUROC	=	area under the receiver operator curve
BMI	=	body mass index
CK-18	=	cytokeratin-18 fragments
CVD	=	cardiovascular disease
DDW	=	Digestive Disease Week
EASL	=	European Association for the Study of Liver Disease
ELF	=	Enhanced Liver Fibrosis panel
GGT	=	gamma-glutamyltyranspeptidase
IMT	=	carotid intima-media thickening
LB	=	liver biopsy
LR	=	likelihood ratio
MOOSE	=	Meta-Analyses of Observational Studies in Epidemiology
MRS	=	magnetic resonance spectroscopy
NAFLD	=	non-alcoholic fatty liver disease
NASH	=	non-alcoholic steatohepatitis
NHANES-III	=	National Institutes of Health Survey-III
OGTT	=	oral glucose tolerance test
OR	=	odds ratio
RCT	=	Randomized Controlled Trial
SHIP	=	Study of Health in Pomerania
SS	=	simple steatosis
STARD	=	Standards for Reporting of Diagnostic Accuracy
STROBE	=	Strengthening the Reporting of Observational Studies in Epidemiology

Key messages

• NAFLD is associated with an increased mortality than in the general population, deriving from liver-related and cardiovascular disease, and carries also a 2-fold increased risk of type 2 diabetes.

• Different histological subtypes have different prognoses: while liver-related mortality is confined to NASH (and particularly to NASH with advanced fibrosis), cardiovascular mortality does not differ between simple steatosis and NASH.

• Liver biopsy, the gold standard for staging and monitoring NAFLD, is not feasible in all patients; three non-invasive methods received independent validation for individuating NASH (plasma cytokeratin-18 fragments) and advanced fibrosis (NAFLD fibrosis score and Fibroscan). While biopsy remains the gold standard for staging and monitoring NAFLD, a combination of these non-invasive tests with liver biopsy may be used to individuate patients with progressive liver disease, who should be referred to a gastroenterologist for experimental therapies and tight follow-up.

Introduction

Non-alcoholic fatty liver disease (NAFLD) is defined by the presence of liver fat accumulation exceeding 5% of hepatocytes, in the absence of significant alcohol intake (20 grams/day for men and 10 g/d for women), viral infection, or any other specific etiology of liver disease (1). NAFLD encompasses a histological spectrum ranging from simple steatosis (SS) to non-alcoholic steatohepatitis (NASH), the latter characterized by steatosis plus necroinflammation; NASH can have different stages of fibrosis ranging from absent (stage F0) to cirrhosis (stage F4). SS, NASH, and different fibrosis stages can only be differentiated by liver biopsy (LB).

The prevalence of NAFLD in the general adult population depends heavily on the sensitivity of the method used, ranging from 33% (by magnetic resonance spectroscopy (MRS)), to 25% (by ultrasonography), or 3%–12% (by liver enzymes) (2,3). The prevalence of ultrasonographic NAFLD reaches 59% in subjects with the metabolic syndrome, 70% in diabetic and obese, and 90% in morbidly obese patients (4,5), while NASH affects 3% of the general adult population, 25%–30% of obese and diabetic, and 35% of morbidly obese individuals (6,7).

NAFLD carries a significant burden for the public health: in a 5-year population-based follow-up, the presence of NAFLD increased by 26% the overall health care costs, after controlling for co-morbidities. Furthermore, NASH is projected to be the leading cause of liver transplantation by 2020 (8,9).

There is no agreement on how to manage NAFLD patients, because the prognosis of NAFLD and its histological subtypes (SS and NASH with different fibrosis severity) is unclear. Furthermore, liver biopsy, the gold standard for staging liver disease, is not feasible in all NAFLD patients, and an alternative strategy to screen subjects with progressive liver disease remains to be defined.

We reviewed the evidence regarding two issues:

1. What is the prognosis (in terms of overall and cause-specific mortality and morbidity) of NAFLD and its histological subtypes (simple steatosis, NASH with different fibrosis stages)? To this aim, fibrosis was defined as significant (histological stage ≥ F2) or advanced (histological stage ≥ F3) in accordance with current guidelines (1).

2. What is the diagnostic accuracy of non-invasive tests for differentiating NAFLD histological subtypes?

Methods

Data sources and searches

We searched English and non-English language publications on MEDLINE, Ovid MEDLINE In-Process, Cochrane Library, EMBASE, PubMed, and abstracts from annual AASLD (American Association for the Study of Liver Disease), AGA (American Gastroenterological Association), EASL (European Association for the Study of Liver Disease), and DDW (Digestive Disease Week) meetings through July 2010. Relevant meta-analyses of observational studies were also included if following MOOSE (Meta-Analyses of Observational Studies in Epidemiology) Guidelines (10).

We also contacted authors to get information about published studies (see Acknowledgment). Search terms were: NASH, NAFLD, non-alcoholic steatohepatitis, non-alcoholic fatty liver disease, fatty liver, liver fat, steatosis, liver enzymes, transaminase, alanine aminotransferase (ALT), aspartate aminotransferase (AST), gamma-glutamyltyranspeptidase (GGT), diagnosis, prognosis, natural history, non-invasive methods, severity of liver disease, fibrosis.

Study selection

Inclusion criteria: Participant adult population of any sex or ethnicity with NAFLD, including NASH-related cirrhosis, diagnosed on the basis of biochemical, radiological, or histological evidence of fatty liver and exclusion of other competing causes of steatosis (11). For issue 2, studies reporting at least sensitivity, specificity, and area under the receiver operator curve (AUROC) of non-invasive methods against liver biopsy (LB) were included.

Exclusion criteria: non-human studies, letters/case reports, studies enrolling < 10 subjects or subjects aged < 12 years, articles not reporting outcomes of interest or primary data (editorials, reviews), or using inadequate case definition. Non-invasive methods that were not internally (i.e. in the study presenting the method for the first time) or externally validated (i.e. by other studies) against LB were excluded.

Outcome measures

Issue 1: overall and cause-specific morbidity and mortality, including liver-related disease, cardiovascular disease (CVD) (defined as cardiac or cerebrovascular fatal and non-fatal events), and malignancy. Additionally, the risk of developing type 2 diabetes was evaluated.

Issue 2: sensitivity, specificity, negative and positive likelihood ratios (LR+, LR-), and AUROC curves of non-invasive methods against liver biopsy (LB).

Data extraction and quality assessment

Data were extracted from each study independently and in duplicate by two authors (G.M., G.P.); discrepancies were resolved by mutual discussion. The agreement between the two reviewers for selection and validity assessment of trials was decided by kappa coefficient. For issue 1, the analysis was carried out in concordance with the MOOSE Guidelines (10).

Methodological quality of observational and diagnostic accuracy studies was assessed by the 22-item STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) and the 25-item STARD (Standards for Reporting of Diagnostic Accuracy) check-lists, respectively (12,13). The following items were specifically incorporated into the STARD check-list: time elapsed between LB and non-invasive test (≤ 6 months versus > 6 months), sample representative of the typical NAFLD population, blinding of LB-reading pathologist and of non-invasive test researcher, LB processed and scored according to standard criteria, adequate biopsy specimen (fragment length ≥ 1.5 cm, with > 6 portal tracts), compliance with pre-analytical/analytical procedures, interquartile range/median ratio measures < 30% (for Fibroscan).

Data synthesis and analysis

We used WinBUGS 1.4 (WinBUGS 1996–2003, Imperial College of Science & MRC, UK). The analysis was carried out in concordance with the Cochrane Handbook of Systematic Reviews.

Issue 1: We presented dichotomous variables as odds ratios (OR) with 95% confidence interval (CI). Data were stratified by sex whenever possible. The fixed-effect model was used, with significance set at P = 0.05. Statistical heterogeneity was assessed using the I2 statistic: when I2 values ≥ 50%, we used a random effects model and planned to explore characteristics of individual studies and those of subgroups of the main body of evidence. Sensitivity analyses were performed by removing one study at a time, and the meta-analysis was repeated to assess whether any one study significantly affected pooled estimates. The effect of duration of follow-up on study results was assessed by meta-regression. Publication bias was examined using funnel plots. Additionally, for issue 1 subgroup analysis was planned to assess the impact of metabolic syndrome on the association between NAFLD and cardio-metabolic outcomes (studies adjusting versus studies not adjusting for the presence of metabolic syndrome, overall or each component).

Issue 2: The diagnostic accuracy of non-invasive tests was analyzed by a bivariate random effects model to provide mean (95% CI) summary estimate of AUROC, sensitivity, specificity, and positive and negative likelihood ratio (LR+, LR-) (14). The AUROCs of different non-invasive tests were compared with the non-parametric method by DeLong et al. (15). Subgroup analysis was planned to assess the effect of each quality item of the STARD check-list, of different ethnic origin of study population (Caucasian versus non-Caucasian), of different degrees of obesity (non-obese versus obese versus morbidly obese subjects), on pooled estimates of diagnostic accuracy.

Strength of recommendation was rated according to the GRADE system: for each outcome four key domains (risk of bias, consistency, directness, and precision) were assessed, and the level of evidence was ranked as high, moderate, low, insufficient (16). On this basis recommendations were rated as Level I (strong) or Level II (weak).

Results

The agreement between two reviewers for study selection was 0.89 and for quality assessment of trials was 0.90. The flow of study selection is reported in Figure 1.

Figure 1. Evidence acquisition flow diagram. Quality scores of included studies are provided as median (range).

For issue 1, 45 cohort studies and 2 meta-analyses were retrieved: 5 studies were excluded for inadequate definition of cases and/or outcomes; eventually, 40 cohort studies (median, range STROBE score: 21, 18–22) and 2 meta-analyses were included.

For issue 2, 43 studies assessing 21 different non-invasive methods for assessing liver disease severity were retrieved: 8 studies presented methods that did not receive internal or external validation and were excluded, 2 studies overlapped for population with previously published studies, and 2 studies provided insufficient data on outcomes of interest. At the end of selection, 31 studies were included in the analysis for issue 2 (median STARD score 24, range 19–25).

Issue 1: what is the prognosis of NAFLD and its histological subtypes?

Overall and cause-specific mortality in NAFLD. We retrieved three population-based studies (the Third National Health and Nutrition Examination Survey (NHANES-III), the Danish National Registry of Patients, and the Study of Health in Pomerania (SHIP)) and four community-based studies, with median follow-up ranging from 7.3 to 24 years. The NHANES-III used biochemical definition of NAFLD, the others using ultrasonographic or histological definitions. Pooled overall mortality was higher in NAFLD compared to the general population: OR 1.57; 95% CI 1.18–2.10; I2 = 88%; P = 0.002 (Figure 2) (17–24). Heterogeneity was high: after excluding the Danish study (23), which less extensively adjusted for confounding variables, heterogeneity dropped down while pooled mortality ratio remained similar in magnitude and direction to overall effect: OR 1.40; 95% CI 1.23–1.60; I2 = 0%; P < 0.00001.

Figure 2. Mortality in NAFLD compared with the general population.

In population-based studies, liver disease was the third leading cause of death (13% of all deaths) in NAFLD after malignancy (28% of all deaths) and ischemic heart disease (25% of cases), compared to the eleventh in the general population (21,23). In the NHANES-III the burden of NAFLD was particularly relevant among the 45–54-years age group, with a standardized mortality ratio of 4.40 (95% CI 1.27–13.23) for all-causes and of 8.15 (95% CI 2.00–33.20) for cardiovascular disease, after adjusting for conventional risk factors (22).

In synthesis, the presence of NAFLD as assessed by different methods carried an increased overall mortality.

NAFLD and incident CVD. A total of 14 population-based and 9 community-based prospective studies (STROBE score ranging 19–22), using a biochemical or ultrasonographic/histological definition, assessed NAFLD as a predictor of CVD (mean duration of follow-up ranging 3.3–24 years) (Table I) (4,17,25–60).

Table I. Prospective studies assessing the natural history of NAFLD, grouped according to definition of NAFLD (biochemical, radiological, or histological).

NAFLD and diabetes
Author year (ref) STROBE score (failing items)	Follow-up (yrs); Population, age, sex (% M)	NAFLD definition	Outcome	Result
Vozarova 2002 (45) 21(g)	6.9 yrs; Community-based cohort of 451 Pima Indians, aged 18–50 yrs, 62% M	ALT	Diabetes incidence	OR for ALT elevation: 1.9 (95% CI 1.1–3.3) after adjustment for age, sex, body fat, insulin resistance and secretion, alcohol intake
Lee 2003 (26) 18(i,u,z,y)	4 yrs; Community-based cohort of 4088 healthy men working in a steel manufacturing company, aged 25–55 yrs	GGT	Diabetes incidence	OR for top versus lowest GGT quintile in drinkers ≤90 g/wk: 3.6 (95% CI 1.1–11.4). Adjusted for age, BMI, smoking, exercise, family history of diabetes mellitus, fasting blood glucose, alcohol intake
Sattar 2004 (27) 21(y)	4.9 yrs; Population-based cohort of 5974 men, mean age 55 yrs (West of Scotland Coronary Prevention Study)^a	ALT	Diabetes incidence	OR for top versus lowest ALT quartile: 2.04 (95% CI 1.16–3.58). Adjusted for age, BMI, smoking, systolic BP, total/HDL cholesterol ratio, pravastatin treatment, triglycerides, alcohol intake, fasting glucose, C-reactive protein, metabolic syndrome
Hanley 2004 (43) 21(u)	5.2 yrs; Population-based cohort of 906 subjects aged 40–69 yrs, 44% M (Insulin Resistance Atherosclerosis Study)^a	AST, ALT	Diabetes incidence	OR for top versus other ALT quartiles: 2.00 (1.22–3.28); for top versus other AST quartiles: 1.98 (1.23–3.17). Adjusted for age, sex, ethnicity, alcohol intake, waist, insulin sensitivity and secretion, plasma lipids
Nakanishi 2004 (28) 21(u)	7 yrs; Population-based cohort of 3260 Japanese male office workers, aged 35–59 yrs	GGT	Diabetes incidence	OR for top versus lowest GGT quintile: 2.44 (1.34–4.46). Adjusted for age, family history of diabetes, BMI, alcohol intake, smoking habits, physical activity fasting plasma glucose, white blood cell (WBC) count, other liver enzymes
Wannamethee 2005 (31) 22	5 yrs; Population-based cohort of 3500 men, aged 60–79 yrs (British Regional Heart Study)^a	ALT, GGT	Diabetes incidence	OR for highest versus lowest ALT quartile: 1.91 (1.01–3.60); for highest versus lowest GGT quartile: 2.69 (1.21–5.97). Adjusted for BMI, insulin resistance, C-reactive protein, alcohol intake, age, physical activity, smoking status
Nannipieri 2005 (44) 22	7 yrs; Population-based cohort of 1441 subjects, aged 35–64 yrs, 40% M (Mexico City Diabetes Study)	GGT	Diabetes incidence	OR for elevated GGT: 2.05 (1.39–3.03). Adjusted for age, sex, BMI, waist circumference, alcohol intake, fasting plasma insulin
Andre 2005 (36) 22	3 yrs; Population-based cohort of 4201 subjects, aged 30–65 yrs, 49% M (DESIR study)	GGT	Diabetes incidence	OR for top versus lowest GGT quartile: 5.0 (1.5–15.8) in men and 3.5 (0.9–14.2) in women. Adjusted for age,WHR, insulin resistance, alcohol intake, other liver enzymes, physical activity, smoking habits
Schindhelm 2005 (29) 21(y)	6 yrs; Population-based cohort of 1289 subjects, aged 50–74 yrs, 46% M (Hoorn Study)	ALT	Diabetes incidence	OR for upper versus lowest ALT tertile: 1.18 (0.66–2.11). Adjusted for age, sex, follow-up duration, BMI, waist circumference, alcohol intake, fasting plasma insulin, 2-h OGTT glucose levels
Meisinger 2005 (56) 21(g)	14.7 yrs; Population-based cohort of 3687 subjects (MONICA study), age range 25–64 yrs, 50% M	GGT	Diabetes incidence	OR for top versus lowest GGT quintile: 4.24 (2.11–8.54) in men and 2.41 (1.05–5.52) in women. Adjusted for age, hypertension, BMI, dyslipidemia, parental history of diabetes, regular smoking, alcohol intake, physical activity, education
Doi 2007 (35) 22	9 yrs; Population-based cohort of 1804 Japanese subjects, age 40–79 yrs, 40% M (Hisayama Study)	ALT, GGT	Diabetes incidence	OR for top versus lowest ALT quartile: 2.32 (0.91–5.92) for men, 4.40 (1.38–14.06) for women. OR for top versus lowest GGT quartile: 2.54 (1.03–6.26) for men, 5.73 (1.62–20.19) for women. Adjusted for age, total and abdominal adiposity, C-reactive protein, alcohol intake
Cho 2007 (39) 20 (u,z)	2 yrs; Cohort of 10038 Koreans from an urban and a rural community, age 51 yrs, 47% M	ALT	Diabetes incidence	OR for top versus lowest ALT quartiles: 2.20 (1.28–3.73) in males, 1.97 (1.03–3.77) in females. Adjusted for age, BMI, systolic blood pressure, family history of diabetes, C-reactive protein, smoking/alcohol/exercise status, fasting plasma glucose/triglyceride/HDL-C
Sato 2008 (46) 22	4 yrs; Community-based cohort of 8576 Japanese men, aged 40–55 yrs (Kansai Healthcare Study)	ALT, GGT	Diabetes incidence	In non-drinkers, highest AST, ALT, and GGT tertile had an OR of 2.02 (1.30–3.14), 2.37 (1.55–3.62), and 3.18 (1.75–5.76) for incident diabetes compared with moderate drinkers with the lowest GGT tertile. Adjusted for age, BMI, smoking habit, family history of diabetes, physical activity, fasting plasma glucose
Monami 2008 (42) 22	39.6 months; Population-based cohort of 2662 Italian subjects, aged 40–75 yrs, 43% M (FIBAR study)	GGT	Diabetes incidence	OR: 2.54 (1.26–5.11) for GGT > 40 IU/L. Adjusted for age, sex, metabolic syndrome
Jiamjarasrangsi 2008 (41) 21(u)	2.6 yrs; Community-based cohort of 2370 employees of a university hospital in Bangkok, mean age 43 yrs	ALT	Diabetes incidence	OR for top versus lowest ALT quartile: 6.14 (1.54–24.45). Adjusted for age, gender, education, alcohol consumption, smoking, family history of diabetes, BMI, systolic and diastolic blood pressure, fasting plasma glucose and uric acid, blood urea, follow-up duration, AST levels
Ford 2008 (40) 22	7 yrs; Population-based cohort of 2298 subjects, mean age 49.5 yrs, 38% M (from the EPIC-Potsdam Study)	ALT, GGT	Diabetes incidence	OR of 1.93 (1.27–2.92) for highest versus lowest ALT quintile and of 2.67 (1.63–4.27) for highest versus lowest GGT quintile. Adjusted for age, sex, education, smoking status, alcohol use, BMI, waist circumference, occupational activity, sports activity, systolic blood pressure, total and HDL-cholesterol, C-reactive protein, glucose
Goessling 2008 (58) 22	20 yrs; Community-based cohort of 2812 participants in the Framingham Offspring Heart Study, age 44 yrs, 44% M	AST, ALT	Diabetes incidence	OR of 1.33 (95% CI 1.17–1.53) per 1 gender-specific SD increase in logAST; OR of 1.48 (95% CI 1.30–1.69) per 1 gender-specific SD increase in logALT Adjusted for age, gender, smoking, alcohol intake, BMI, menopausal status, base-line plasma glucose, interim weight change
Fraser 2009 (47) 19(u,v,z)	7.3 yrs; Population-based cohort of 3041 women, aged 60–79 yrs (British Women's Heart and Health Study)	GGT	Diabetes incidence	OR 1.24 (1.04–1.48) per 1 SD change in serum GGT. Adjusted for age, alcohol intake, social class, physical activity, smoking, insulin resistance, waist-to-hip ratio, triglycerides, HDL-C, systolic blood pressure
Kim 2009 (37) 21(y)	5 yrs; Population-based cohort of 3556 subjects, mean age (range) 46 (20–79 yrs), 62% M	ALT, GGT	Diabetes incidence	OR of 4.31 (1.56–11.88) for highest versus lowest ALT quartile and of 3.07 (1.21–7.76) for highest versus lowest GGT quartile. Adjusted for sex, age, family history of diabetes, BMI, systolic blood pressure, alcohol consumption, smoking, exercise, fasting glucose, serum triglyceride, HDL-cholesterol, CRP
Adams 2009 (60) 22	11 yrs; Community-based cohort of 352 subjects, mean age 60 yrs, 68% M	ALT	Diabetes incidence	HR of 0.60 (0.21–1.20) after adjustment for age, BMI, waist, HOMA, fasting insulin, fasting glucose, triglycerides, alcohol intake
Okamoto 2003 (33) 19 (g,u,y)	10 yrs; Community-based cohort of 840 subjects, mean age 43 yrs, 56% M	US	Diabetes incidence	OR for NAFLD: 1.83 (0.95–3.51). Adjusted for age, sex, BMI, change in BMI, fasting plasma glucose, HbA1c, frequency of medical check-ups, alcohol intake, family history of diabetes
Shibata 2007 (34) 22	4 yrs; Community-based cohort of 3189 Japanese men, aged ≥ 40 yrs consuming < 20 g alcohol/day	US	Diabetes incidence	OR for NAFLD: 4.6 (3.0–6.9) after adjusting for age, BMI, metabolic syndrome, smoking and physical activity status
Kim 2008 (32) 21(u)	5 yrs; Community-based cohort of 5372 Koreans, mean age (range) 47 (20–79) yrs, 68% M	US	Diabetes incidence	OR for NAFLD: 3.92 (2.89–5.31) after adjusting for age, sex, alcohol intake, smoking, plasma lipids. The severity of ultrasonographic steatosis was associated with the risk of diabetes, even after excluding frequent drinkers.
Ekstedt 2006 (20) 20(g,u)	13.7 yrs; Community-based cohort of 129 patients with NAFLD, mean age 51 yrs, 67% M	Histology	Diabetes incidence	Diabetes occurred in 71% of patients with NASH and 46% of patients with SS (OR 2.98; 95% CI 1.23–7.22; P = 0.01)
NAFLD and incident fatal and non-fatal CVD events
Author	Follow-up (yrs); Population, age, sex (% M)	NAFLD definition	Outcome	Result/covariates in fully-adjusted models
Wannamethee 1995 (25) 21(g)	11.5 yrs; Population-based cohort of 7613 British middle-aged men	GGT	Fatal CHD	OR of highest quintile of 1.42 (95% CI 1.12–1.80). Adjusted for alcohol intake, BMI, smoking, pre-existing CHD, diabetes, antihypertensive medication, systolic and diastolic blood pressure, total and HDL-cholesterol, heart rate, blood glucose, physical activity, lung function (forced expiratory volume in 1 second (FEV1))
Ruttmann 2005 (11) 22	11.1 yrs (M), 12 yrs (F); Population-based cohort of 163944 Austrian subjects, aged 19–95 yrs, 46% M	GGT	Fatal CVD events	OR in 5th versus 1st GGT quintile of 1.57 (95% CI 1.30–1.91) in men and 1.68 (95% CI 1.26–2.25) in women for total mortality from CVD. Adjusted for age, body mass index, systolic blood pressure, cholesterol, triglycerides, glucose, smoking, work status, and year of examination
Lee DH 2006 (53) 21(u)	11.9 yrs; Population-based cohort of 28838 Finnish subjects, age 25–74 yrs, 48% M	GGT	Fatal and non-fatal CHD	Risk of CHD incrementally increased across GGT quintiles: OR for highest versus lowest GGT quintile: 1.57 (1.22–2.01) in men and 1.44 (1.03–2.02) in women. Adjusted for age, sex, BMI, diabetes, smoking habits, hypertension, alcohol intake, physical activity, total and HDL cholesterol
Meisinger 2006 (57) 21(g)	15.7 yrs; Population-based cohort of 1878 Austrian adults, age 25–64 yrs, 100% M (from MONICA study)	GGT	Fatal and non-fatal CHD	OR for the highest versus lowest GGT quartile: 2.34 (95% CI 1.23–4.44). Adjusted for age, diabetes, hypertension, smoking, education, TC/HDL-C ratio, physical activity, BMI, alcohol intake
Hozawa 2007 (30) 22	9.6 yrs; Population-based Japanese cohort of 2724 men and 4122 women, aged ≥ 30 yrs	GGT	Fatal CVD events	OR in highest versus lowest quartile: 2.88 (95% CI 1.14–7.28) in women. In the never-drinker women OR 1.62 (95% CI 1.11–2.37). No significant relationships observed in men. Adjusted for age, alcohol consumption, smoking, total and HDL-cholesterol, triglyceride, GOT, GPT, BMI, exercise, systolic BP, diabetes, antihypertensive medication
Lee DS 2007 (50) 22	19.1 yrs; Population-based cohort of 3451 subjects, mean age 44 yrs, 48% M (Framingham Heart Study)	GGT	Fatal and non-fatal CVD	OR in highest GGT quartile: 1.67 (1.25–2.22). Adjusted for age, sex, BMI, alcohol intake, smoking habits, physical activity, total and HDL cholesterol, diabetes, hypertension fasting plasma insulin, C-reactive protein
Schindhelm 2007 (49) 21(y)	10 yrs; Population-based cohort of 1439 subjects, aged 61 yrs, 45% M (Hoorn Study)	ALT	Fatal and non-fatal CHD events	OR for upper tertile versus lowest ALT tertile: 2.04 (1.35–3.10). Adjusted for age, sex, BMI, waist circumference, smoking, alcohol intake, blood pressure, LDL-cholesterol, triglycerides, glucose, insulin resistance
Goessling 2008 (58) 22	20 yrs; Community-based cohort of 2812 participants in the Framingham Offspring Heart Study, age 44 yrs, 44% M	AST, ALT	Fatal and non-fatal CVD events	OR of 0.97 (95% CI 0.88–1.07) per 1 gender-specific SD increase in logAST; OR of 1.05 (95% CI 0.96–1.16) per 1 gender-specific SD increase in logALT Adjusted for age, gender, systolic blood pressure, hypertension treatment, smoking, alcohol intake, BMI, diabetes, total/ HDL-cholesterol, lipid treatment
Monami 2008 (42) 22	39.8 months; Population-based cohort of 2662 Italian subjects, aged 54 yrs, 43% M (FIBAR study)	AST, GGT	Fatal and non-fatal CVD	OR 2.21 (0.98–5.43) for GGT > 40 IU/L; OR 6.5 (1.5–28.1) for AST > 40 IU/L. Adjusted for age, sex, metabolic syndrome
Ong2008 (21) 22	8.7 yrs; Population-based cohort of 11285 subjects, 47% M (NHANES-III and NHANES-III-Linked Mortality File)	ALT	Mortality from all causes, including liver disease and CVD	NAFLD had higher overall mortality (OR: 1.038, 95% CI 1.036–1.041) and liver-related mortality (OR: 9.32, 95% CI 9.21–9.43). Adjusted for age, gender, race, education, income, BMI, diabetes, hypertension
Dunn 2008 (22) 22	8.7 yrs; Population-based NHANES-III cohort of 7574 US subjects (980 NAFLD), aged 35–84 yrs, 41% M	ALT	Mortality from all causes, including CVD	NAFLD had higher age-adjusted all-cause mortality (OR: 1.37; 95% CI 0.98–1.91). In the 45–54 age group, NAFLD had significantly higher all-cause (OR: 4.40; 95% CI 1.27–13.23) and CVD mortality (HR: 8.15; 95% CI 2.00–33.20). Adjusted for age, gender, race, blood pressure, waist circumference, triglyceride, total and HDL-cholesterol, smoking, C-reactive protein, alcohol intake, physical activity, diabetes, statin use
Olynyk 2009 (59) 22	10 yrs; Population-based cohort of 3719 Australian subjects, age 53 yrs, 42% M (Busselton Health survey)	ALT	Fatal and non-fatal CVD events	HR 0.78 (0.54–1.13) for men and 1.09 (0.78–1.52) for women for higher versus lowest ALT tertile after adjustment for age, smoking, alcohol use, systolic blood pressure, fasting glucose, total and HDL-cholesterol, triglycerides, waist circumference
Fraser 2009 (51) 19(u,v,z)	7.3 yrs; Population-based cohort of 2961 women, aged 60–79 yrs (British Women's Heart and Health Study)	ALT, GGT	Fatal and non-fatal CVD events	OR 0.94 (0.65–1.37) per 1 SD change in serum ALT. OR 1.17 (0.93–1.48) per 1 SD change in serum GGT. Adjusted for age, alcohol intake, social class, physical activity, smoking, diabetes/insulin resistance, components of metabolic syndrome, BMI, triglycerides, HDL-C, systolic blood pressure
Jepsen 2003 (23) 20(g,h)	16 yrs; Population-based cohort (Danish National Registry of Patients) 1804 subjects with NAFLD (hospital diagnosis), 53% M	US	Mortality for all causes, including liver disease and CVD	NAFLD had an age-adjusted standardized mortality ratio (SMR) of 2.3 (95% CI 2.1–2.6) for all causes, of 19.7 (95% CI 15.3–25.0) for hepatobiliary disease, and of 2.1 (95% CI 1.8–2.5) for CVD. Adjusted for sex, diabetes, cirrhosis at base-line
Targher 2005 (52) 21(y)	5 yrs; Nested case-control study of 744 type 2 diabetic patients, age 65 yrs, 62% M	US	Fatal and non-fatal CVD	OR for NAFLD of 1.53, 95% CI 1.1–1.7. Adjusted for age, sex, smoking history, diabetes duration, HbAlc, metabolic syndrome, LDL cholesterol, liver enzymes, use of medications
Targher 2007 (48) 21(y)	6.5 yrs; Cohort of 2392 diabetic subjects aged 64 yrs, 56% M (Valpolicella Heart Study)	US	Fatal and non-fatal CVD	OR for NAFLD: 1.87 (1.2–2.6). Adjusted for age, sex, smoking habits, diabetes duration, HbAlc, metabolic syndrome, LDL-cholesterol, medications
Hamaguchi 2007 (55) 22	5 yrs; Community-based cohort of 1637 apparently healthy Japanese, aged 48 yrs, 60% M	US	Non-fatal CVD events	OR for NAFLD: 4.12(1.58–10.75) independently of age, sex, BMI, alcohol intake, smoking habits, LDL-cholesterol, each component of metabolic syndrome and metabolic syndrome
Haring 2009 (24) 22	7.2 yrs; Population-based cohort of 4160 German subjects, aged 50 yrs, 49% M (SHIP Study)	US + GGT	All-cause and CVD mortality	OR for all-cause mortality and CV mortality of ultrasonographic steatosis and highest GGT quintile in men: 1.98 (1.21–3.27) and 2.41 (1.05–5.55). The risk was not increased in women. Adjusted for age, waist circumference, alcohol consumption, physical activity, educational level, civil status, equalized income, functional co-morbidity index
Matteoni 1999 (19) 20(g,u)	18 yrs; Community-based cohort of 132 NAFLD patients, aged 53 yrs, 48% M	Histology	Overall and cause-specific deaths	Higher liver-related death in NASH (10%) compared to simple steatosis (2%). CVD death non-significantly higher in NASH compared to simple steatosis
Dam-Larsen 2003 (63) 20(g,u)	16.7 yrs; Population-based cohort (Danish National Registry of Patients) 109 subjects with non-alcoholic simple steatosis, age 19–80 (mean 39), 30% M	Histology	Overall and cause-specific deaths	No excess of overall (25%) and cause-specific mortality in non-alcoholic simple steatosis compared with the general Danish population
Adams 2005 (18) 21(g)	7.6 yrs; Community-based cohort of 420 patients with NAFLD aged 49 yrs, 49% M	US + histology	Overall and cause-specific mortality	Higher age- and sex-adjusted all-cause mortality (OR: 1.34, 95% CI 1.003–1.76). Liver disease was the 3rd leading cause of death (13% of all deaths) as compared with the 13th in the general population. Other causes of death were malignancy (28% of all deaths) and CVD (25% of all deaths)
Sanyal 2006 (64) 19(u,v,z)	10 yrs; Community-based cohort of 152 NASH-related and 150 age-, sex-, and race-matched HCV-related cirrhotic patients, aged 55 yrs, 49% M	Histology	All-cause deaths, including liver-related and CVD-related	Patients with NASH had a lower overall mortality (19% versus 29%; P = 0.04) but a higher CVD mortality rate (5%, 2nd cause of death, versus 1%, 5th cause of death; P = 0.03). Adjusted for age, BMI, diabetes, Child-Pugh score
Author year (ref) STROBE score (failing items)	Follow-up (yrs); Population, age, sex (% M)	NAFLD definition	Outcome	Result
Ekstedt 2006 (20) 20(g,u)	13.7 yrs; Community-based cohort of 129 patients with NAFLD, aged 51 yrs, 67% M	Histology	Fatal and non-fatal CVD events	CVD incidence and mortality of 29% and 16% in NASH, 3-fold and 2-fold higher, respectively, than in the SS and in general population
Rafiq 2009 (54) 21(g)	10.5–13 yrs; Community-based cohort of 173 patients with NAFLD, aged 50 yrs, 40% M	Histology	Fatal CVD events	CVD was the top cause of death (12.7% of cases) in the entire NAFLD cohort, with no difference between NASH (12.2%) and non-NASH (20%; P = 0.23)
Soderberg 2010 (17) 20(g,u)	24 yrs; Community-based cohort of 118 NAFLD patients (43% NASH), aged 47 yrs, 63% M	Histology	Overall and cause-specific deaths	Age, sex, and calendar period-adjusted mortality ratio of 1.69 (1.24–2.25) for NAFLD compared to the general population (even after excluding cirrhotic patients at base-line). Excess mortality of NAFLD was accounted for by NASH cohort, while steatosis survival rate was comparable to the general population. Cardiovascular disease, malignancy and liver disease were the top 3 causes of death

The author name is followed in the same box by the year of publications and by the 22-item STROBE score, with the item(s) not satisfied by the study indicated in parentheses:

(a) Title and abstract informative and balanced.

(b) Background/rationale stated in the introduction.

(c) Objective(s) specified in the introduction.

(d) Study design correctly presented early in the paper.

(e) Setting, locations, and relevant dates described.

(f) Eligibility criteria, methods of selection, and follow-up described.

(g) Diagnostic criteria, outcomes, exposures, predictors, potential confounders, and effect modifiers for all variables clearly defined.

(h) Sources of data and details of methods of measurement given for each variable of interest.

(i) Any efforts to address potential sources of bias described.

(l) How the study size was arrived at clearly explained.

(m) How quantitative variables were handled in the analyses clearly explained.

(n) All statistical methods, how missing data and loss to follow-up were addressed, any sensitivity analyses clearly described.

(o) Numbers of individuals at each stage of study reported.

(p) Characteristics of study participants, number of participants with missing data, average and total follow-up time clearly described.

(q) Outcome events or summary measures over time reported.

(r) Unadjusted and confounder-adjusted estimates and their precision (e.g. 95% CI) reported.

(s) Analyses of subgroups and interactions, and sensitivity analyses reported.

(t) Key results with reference to study objectives summarized.

(u) Limitations of the study discussed.

(v) Cautious overall interpretation of results given.

(z) Generalizability (external validity) of the study results discussed.

(y) Source of funding and role of the funders described.

^aProspective studies on diabetes incidence: some population-based prospective studies were not specifically designed to address the risk factors for developing diabetes, rather the relationship between insulin resistance, cardiovascular disease, and their risk factors (25,40,42); in these studies, the relationship between elevations in biochemical markers of liver injury and incident diabetes was analyzed, after adjusting for alcohol intake.

CT = computed tomography; CAD = coronary artery disease; CHD = coronary heart disease; HCV = Hepatitis C virus; IGT = impaired glucose tolerance; LV = left ventricular; MRS = magnetic resonance spectroscopy; OR = odds ratio; IMT = intima-media thickening; TC = Total cholesterol; US = ultrasonography; WHR = Waist-on-hips ratio.

Pooled OR for incident CVD of highest versus lowest ALT and GGT quantiles and of ultrasonographic/histological NAFLD were 1.10 (95% CI 0.85–1.41; I2 = 52%; n comparisons = 6), 1.57 (95% CI 1.42–1.74; I2 = 12%; n comparisons = 10), and 2.05 (95% CI 1.81–2.31; I2 = 31%; n comparisons = 8), respectively (Figure 3). Heterogeneity for ALT was high: after excluding the Hoorn study (49), assessing exclusively coronary heart disease events, heterogeneity disappeared (I2 = 0%) while the magnitude and direction of the effect were unchanged (not shown).

Figure 3. NAFLD as a risk factor for incident CVD events. NAFLD was defined biochemically (increase in ALT and GGT levels, panel A) or by ultrasound or histology (panel B). Forest plot of comparison: Meta-analysis of multiple-adjusted results (OR of top versus bottom quantile of ALT and GGT, panel A, or ultrasound-histologically diagnosed NAFLD, panel B) as determinants of incident CVD, outcome: incident CVD. NAFLD was defined by biochemical criteria (ALT or GGT elevation, panel A) or by ultrasonographic-histological criteria (panel B). OR adjusted for multiple variables from different community-based or population-based prospective studies (Table I) were pooled and analyzed by random or fixed effect models. *Studies assessing fatal CVD events.

Restricting the analysis to studies adjusting for metabolic syndrome did not affect the magnitude and direction of overall effect: in studies with biochemical definition of NAFLD: in studies using GGT elevation: OR 1.30 (95% CI 1.03–1.83; I2 = 34%; n = 7); in ultrasonographic/histological studies: OR 1.75 (95% CI 1.38–2.23; I2 = 0%; n = 6).

Considering only studies assessing fatal CVD events (CVD mortality), the pooled OR of GGT and ultrasonographic/histological NAFLD were 1.59 (95% CI 1.42–1.78; I2 = 29%; P < 0.00001; n comparisons = 5), and 2.16 (95% CI 1.88–2.49; I2 = 0%; P < 0.00001; n comparisons = 5), respectively.

Funnel plots analysis found no strong publication bias (see online Supplementary Figures). There was also no evidence of an association between duration of follow-up and study results (all P values > 0.51).

In synthesis, the presence of NAFLD as assessed by GGT elevation or by ultrasonographic/histological methods, but not by ALT elevation, carried an increased CVD mortality.

NAFLD and incident type 2 diabetes. A total of 15 population-based and 9 community-based prospective studies assessed biochemical (defined as ALT or GGT elevation) or ultrasonographic-histological NAFLD as a predictor of incident diabetes (Table I). Follow-up ranged 2–20 years, and STROBE score ranged 18–22.

The multiple-adjusted OR for incident diabetes of highest versus lowest ALT and GGT quantiles and of ultrasonographic/histological NAFLD were 1.95 (95% CI 1.63–2.33; I2 = 48%; n comparisons = 17), 2.71 (95% CI 2.30–3.20; I2 = 23%; n comparisons = 13), and 3.51 (95% CI 2.28–5.41; I2 = 70%; n comparisons = 3), respectively (Figure 4). The heterogeneity of studies using ALT and ultrasonographic/histological definition was entirely due to the studies by Jiamjarasrangsi and by Okamoto, respectively (33,41): the first enrolled a selected community of employees of a University Hospital in Bangkok, and its findings may not be generalizable; the second assessed hyperglycemia (defined by fasting plasma glucose ≥ 110 mg/dL or HbA1c > 6.4%) as outcome, rather than diabetes. The exclusion of these two studies yielded the following results: OR for incident diabetes of ALT elevation: 1.92 (1.72–2.12; I2 = 9%; n comparisons = 16); OR for incident diabetes of ultrasonographic NAFLD: 4.25 (3.39–5.32; I2 = 0%; n comparisons = 2).

Figure 4. NAFLD as a risk factor for incident type 2 diabetes. NAFLD was defined biochemically (increase in ALT and GGT levels, panel A) or by ultrasound or histology (panel B). Forest plot of comparison: Meta-analysis of multiple-adjusted results (OR of top versus bottom quantile of ALT and GGT, panel A, or ultrasonographic NAFLD, panel B) as determinants of incident type 2 diabetes, outcome: incident type 2 diabetes. NAFLD was defined by biochemical criteria (ALT or GGT elevation, panel A) or by ultrasonographic criteria (panel B). OR adjusted for multiple variables from different community-based or population-based prospective studies (Table I) were pooled and analyzed by random or fixed effect models. *Studies assessing fatal CVD events.

Restricting the analysis to studies adjusting for metabolic syndrome did not affect the magnitude and direction of overall effect: OR 2.08 (95% CI 1.76–2.45; I2 = 17%; n comparisons = 9) in studies with biochemical definition; among ultrasonographic/histological studies only that by Shibata adjusted for metabolic syndrome (34).

No strong evidence was found for the presence of publication bias, or for an association between duration of follow-up and study results (all P values > 0.34).

In synthesis, the presence of NAFLD as assessed by biochemical or by ultrasonographic/histological methods carried an increased risk of developing type 2 diabetes mellitus.

Meta-analyses assessing CVD risk in NAFLD. A meta-analysis of five hospital-based and two population-based case-control studies (total 3,497 subjects) found a 3.13-fold (95% CI 1.75–5.58; P < 0.0002) increased risk of ultrasonography-detected carotid plaques in NAFLD (61).

Other causes of death. Mortality from extrahepatic malignancy was not increased in NAFLD compared to the general population: OR 0.97 (95% CI 0.66–1.26; I2 = 43%; n comparisons = 8; not shown in detail). However, in a cohort of 337 diabetic patients the presence of NAFLD carried a 2.3-fold increased risk (95% CI 0.9–5.9) of dying from malignancy, after adjusting for age, smoking habits, obesity, diabetes duration, hyperlipidemia, earlier malignancy, and cardiovascular disease (62). There were insufficient prospective data to assess the effect of NAFLD on mortality from hepatic malignancy.

Prognosis of NAFLD histological subtypes (SS, NASH with different fibrosis stages). Five community-based studies (total 559 participants) assessed the prognosis of different histological subtypes of NAFLD (median follow-up ranging 7.6–24 years, STROBE score ranging 19–21) (Table I). While the survival of patients with SS approached that of the general population, NASH patients had a higher overall mortality than those with SS (pooled OR 1.81; 95% CI 1.24–2.66; P = 0.002; I2 = 0%).

Liver disease was the main cause of death excess in NASH, with a liver-related mortality rate of 11%–17.5% compared to 1.7%–2.7% of SS (pooled OR of NASH compared to SS: 5.71; 95% CI 2.31–14.13; P = 0.0002; I2 = 0%) (Figure 5).

Figure 5. Outcomes of NASH compared to simple steatosis for overall mortality (panel A), liver-related mortality (panel B), or cardiovascular disease (CVD) mortality (panel C). Forest plot of comparison of NASH versus simple steatosis, outcomes: overall mortality (panel A), liver-related mortality (panel B), CVD mortality (panel C). OR adjusted for variables from different community-based or population-based prospective studies (Table I) were pooled and analyzed by random or fixed effect models.

The rate of progression of liver disease was slow: NASH developed progressive fibrosis in 25%–30% of cases over 4 years and in 50% of cases over 6 years (19,20,23,63,70,71). In the Swedish cohort, 13% of patients with mild-to-moderate fibrosis (stage F1–2) developed cirrhosis, while 25% of patients with fibrosis stage F3 developed cirrhosis and end-stage liver disease (20). Once cirrhosis developed, median survival approached 6 years; compared to viral cirrhosis, NASH cirrhosis had a similar liver-related mortality but a significantly higher CVD-related death (see below).

Beside NASH, the other factor affecting survival was the presence of advanced (stage F3–4) fibrosis. While NASH with milder (stage F0–2) fibrosis still had a higher liver-related mortality than SS (pooled OR 2.11; 95% CI 1.15–6.36; I2 = 0%; P = 0.02), the presence of advanced fibrosis further increased liver-related mortality compared to NASH with fibrosis stage F0–2: OR 10.06 (95% CI 4.35–23.25; I2 = 0%; n comparisons = 5; P = 0.00001; not shown in detail).

Cardiovascular disease accounted for the remaining mortality excess in NAFLD, but the difference between histological subtypes was non-significant: pooled OR for CVD mortality 0.91 (95% CI 0.42–1.98; P = 0.82; I2 = 32%). CVD remained an important cause of death even in advanced liver disease: in a 10-year follow-up of NASH-related cirrhoses CVD accounted for 28% of deaths (second leading cause of death after liver disease) versus the 2% of CVD-related deaths (sixth leading cause) in HCV-related cirrhosis (64).

Only one study assessed the risk of diabetes of different NAFLD subtypes, finding an OR of 2.98 (95% CI 1.23–7.22; P = 0.01) in NASH compared to SS over 13 years (20).

Sensitivity analysis. There was little or no heterogeneity in the meta-analysis of overall, liver-related, and CVD events, suggesting a consistent disease effect. No strong evidence was found for publication bias or for an association between duration of follow-up and study results (all P values > 0.21).

In synthesis NASH confers a higher liver-related mortality than SS, while mortality from CVD did not differ between NASH and SS. Only one study assessed the risk of diabetes of different NAFLD subtypes, finding an OR of 2.98 in NASH compared to SS.

Issue 2: how to assess the severity of liver disease in NAFLD?

The evidence presented above suggests that different NAFLD histological subtypes (SS, NASH, NASH with advanced fibrosis) have different prognoses, and the two NAFLD stages that should be identified early are NASH and NAFLD with advanced (stage ≥ F3) fibrosis. LB remains the gold standard to stage NAFLD, but it does have limitations that make it an unfeasible approach in all NAFLD subjects: it is costly, invasive, and has complications requiring hospitalization in 1%–3% cases and a procedural mortality of 0.01% (65). Furthermore, LB suffers from intra- and inter-observer reading variability, and from sample inadequacy or variability due to uneven disease distribution, which leads to missing a diagnosis of NASH in up to 27% of cases (66): paired liver biopsies showed a single biopsy has an AUROC, sensitivity, specificity, LR+, and LR− of at best 0.81 (95% CI 0.67–0.89), 0.73, 0.92, 8.62, and 0.30 for the diagnosis of NASH, and of 0.87 (95% CI 0.70–0.95), 0.85, 0.89, 7.73, and 0.16 for advanced (stage 3–4) fibrosis, with a discordance of one or more stages of 41% (67,68).

When taken alone, clinical factors (age ≥ 45 years, obesity, diabetes, metabolic syndrome, insulin resistance, triglycerides ≥ 150 mg/dL (1.7 mmol/L)) lack enough predictive value for advanced liver disease to be clinically useful (69–73). Routine imaging techniques can assess liver fat but not necroinflammation and fibrosis. To overcome these limitations, several non-invasive approaches to assess liver disease severity in NAFLD have been proposed in an attempt to fulfill the requisites of the ideal non-invasive method: cost-effectiveness, ease-of-measurement, liver-specificity, reproducibility within patients and among different populations, and prognostic value. These approaches use radiological techniques or combine clinical and biochemical parameters into quantitative panels.

Among imaging techniques, liver stiffness measurement (LSM) by ultrasonography (Fibroscan) or magnetic resonance (MR) have been used to estimate the presence and severity of hepatic fibrosis. While MR elastography is not broadly available, is expensive, and has not been internally or externally validated yet (17,20,54), ultrasonographic LSM by transient elastography (Fibroscan) is becoming increasingly used. The diagnostic accuracy of Fibroscan for significant (stage ≥ F2) or advanced (stage ≥ F3) fibrosis has been assessed in six studies in NAFLD (median, range STARD score: 24, 22–25) (Table II) (74–79). The diagnostic accuracy and reproducibility of Fibroscan for advanced fibrosis were good and were unaffected by the severity of steatosis or inflammation (Table II), but were hampered by the presence of obesity: in all studies obesity was an independent predictor of failure to measure liver stiffness (5%–13% of cases) and, consistently, in all studies mean patients’ BMI was < 30 kg/m².

Table II. Panel A: Studies assessing biomarker panels validated for non-invasive assessment of the presence of NASH in patients with NAFLD. When > 1 study assessed the panel, summary estimates of diagnostic accuracy were calculated; when the panel was assessed in only 1 study, the AUROC is the summary estimate of the training and validation groups.

Study (ref)	n (% NASH)	Age (yrs)	% Females	BMI (kg/m^2)	% Diabetic	AUC (95% CI)	Cut-off kg/m^2	Se %	Sp %	LR+	LR−	STARD check-list (failing items); Comments
ELISA-detected plasma cytokeratin-18 (CK-18) fragments
Wieckowska 2006 (81)	44 (47%)	51	54	32	31	0.93 (0.83–1.00)	380 U/L	91	94	15.00	0.10	25
Yilmaz 2007 (84)	83 (54%)	49	46	30	15	0.78 (0.67–0.87)	121.6 U/L	60	96	5.44	0.15	25
Younossi 2008 (82)	101 (32%)	41	67	46	NR	0.71 (0.59–0.81)	174 U/L	64	87	4.92	0.41	23(k,r)
Diab 2008 (85)	99 (26%)	51	82	48	35	0.88 (0.77–0.99)	252 U/L	82	77	3.57	0.23	25
Tabesh 2008 (93)	70 (64%)	44	51	31	NR	0.83 (0.74–0.89)	275 U/L	47	91	5.88	0.57	23(k,r)
Malik 2009 (86)	95 (63%)	49	36	31	38	0.80 (0.76–0.84)	300 U/L	84	80	4.15	0.21	25
Feldstein 2009 (83)	139 (50%)	48	63	34	19	0.83 (0.75–0.91)	246 U/L	75	81	3.95	0.30	25
Musso 2010 (100)	125 (18%)	48	39	25	0	0.83 (0.80–0.90)	246 U/L	78	88	6.65	0.25	24 (k)
Crisan 2009 (101)	56 (52%)	47	55	30.2	20	0.79 (0.69–0.88)	479	63	84	3.94	0.44	25
Summary estimates	9 studies assessed CK-18 for NASH					0.82 (0.78–0.88)		78 (64–92)	87 (77–98)	6.00 (5.09–7.17)	0.25 (0.23–0.28)	STARD score (median, range): 25 (23–25)
NASHtest (Age, sex, height, weight, triglycerides, cholesterol, α2-macroglobulin, apolipoprotein A1, haptoglobin, AST, ALT, GGT, bilirubin)
Poynard 2006 (68)	257 (25%)	51	40	28.4	37	0.79 (0.67–0.86)	0.34	33	94	5.5	0.82	25; Patented algorithms owned by the FPHO. Intravascular hemolysis (reduced haptoglobin and increased bilirubin), other causes of bilirubin elevation (Gilbert, cholestasis), acute inflammation (increased haptoglobin) and abnormal apolipoprotein A1 levels limit its accuracy
NASH Predictive Index (NPI) Age, BMI, HOMA-IR, log(AST × ALT), female sex
Zein 2007 (75)	122 (69%)	49	66	NR	28	0.86 (0.80–0.93)	6	56	100	56.01	0.44	20(j,k,p,q,w); Unusually high prevalence of disease in this cohort. Patients on insulin were excluded since insulin treatment can affect results
Obesity-related NASH Diagnostics (Serum cleaved CK-18, adiponectin, resistin)
Younossi 2008 (76)	101 (32%)	41	67	46	NR	0.85 (0.77–0.92)	0.27720.34990.2075	9577100	708747	3.215.931.90	0.060.260.00	23(k,r); It measures plasma adipokines adiponectin and resistin and intact and cleaved CK-18, markers of hepatocyte apoptosis, in a cohort of morbidly obese subjects undergoing surgery
MASH Clinical score for morbid obesity (Hypertension, diabetes, AST ≥ 27 IU/L, ALT ≥ 27 IU/L, sleep apnea, non-black race)
Campos 2008 (77)	200 (32%)	43	84	48	26	0.80 (NR)	6	22	99	22.15	0.79	25; Four risk categories for NASH were identified based on six readily available variables. A total score ≥ 6 predicted NASH with very high probability, but its performance in excluding NASH was fair-to-low
MAFIC score (serum ferritin, serum fasting insulin, serum type IV collagen 7s)
Sumida 2010 (105)	619 (55%)	52	49	27.8	41	0.79 (NR)	> 0≥ 2	9058	4490	1.615.80	0.230.47	25; Multicenter Japanese study from 8 hepatology centers

LR+ = Positive likelihood ratio; LR− = Negative likelihood ratio.

Table II. Panel B: Studies assessing biomarker panels validated for non-invasive assessment of fibrosis in patients with NAFLD. When > 1 study assessed the panel, summary estimates of diagnostic accuracy were calculated.

Study	n	Age (yrs)	% Females	BMI	% Diabetic	Diagnostic end-point (prevalence)	AUC (95% CI)	Cut-off	Se %	Sp %	LR+	LR−	STARD check-list (failing items); Comments
Fibrotest (α2-macroglobulin, apolipoprotein A1, haptoglobin, GGT, total bilirubin)
Ratziu 2006 (78)	267	51	42%	48% with BMI > 27	34%	Fibrosis ≥ F2 (27% patients)Fibrosis ≥ F3 (13% patients)	0.81 (0.74–0.86)0.88 (0.82–0.92)	0.300.700.300.70	77159225	77987197	3.357.503.178.33	0.290.870.110.77	25; Patented formula. Limitations are the same as for NASHtest. The 2 proposed cut-off values of 0.30 and 0.70 left in-between 33% of patients, in whom significant liver fibrosis could not be ruled in or out
Adams 2008 (79)	119	49	46%	45% obese	32	Fibrosis ≥ F2 (42%)	0.72 (0.62–0.81)	0.300.70	7370	6960	2.411.82	0.390.50	19 (c,e,h,j,k,n)
Summary estimates	2 studies assessed Fibrotest for fibrosis F2					Fibrosis ≥ F2	0.78 (0.72–0.85)	0.300.70	76 (70–84)32 (24–39)	74 (69–81)87 (81–94)	2.922.46	0.320.78	STARD score: 19–25
NAFLD fibrosis score (Age, BMI, hyperglycemia, platelet count, albumin, AST/ALT ratio)
Angulo 2007 (80)	733	48	47	32	30	Fibrosis ≥ F3 (27%)	0.88 (0.85–0.92)	−1.455 + 0.676	8047	7498	3.1024.01	0.270.54	25; Advanced fibrosis was diagnosed or excluded in 75% of cases, leaving 25% of patients with an indeterminate probability
Qureshi 2008 (86)	331	41	83	48	35	Fibrosis ≥ F3 (16%)	0.82 (NR)	−1.455+0.676	9649	3884	1.513.11	0.110.61	24(f); 47% of patients with an indeterminate probability of advanced fibrosis
Wong 2008 (85)	162	46	41	28.5	57	Fibrosis ≥ F3 (11%)	0.64 (0.49–0.79)	−1.455+0.676	390	8199	2.100.00	0.751.01	24 (p); 20% of patients with an indeterminate probability of advanced fibrosis
Guha 2008 (81)	91	49	71	35.6	58	Fibrosis ≥ F3 (23%)	0.89 (0.81–0.97)	−2.382+0.833	9177	5993	2.2410.74	0.150.25	24 (k); A subgroup of the original cohort of Guha study
Guturu 2008 (88)	118	50	48	31	NR	Fibrosis ≥ F3 (25%)	0.67 (0.58–0.75)	−1.455+0.676	4868	7694	212.21	0.680.33	24(f); 58% of subjects indeterminate probability of advanced fibrosis
Harrison 2008 (87)	92	NR	51	33	NR	Fibrosis ≥ F3 (22%)	0.74	−1.455+0.676	90	29	1.30	0.34	23(f,k)
Cales 2009 (83)	235	51	26	29	24	Fibrosis ≥ F2	0.88 (0.83–0.94)	−1.455+0.676	61	96	16.25	0.41	24 (k)
Ruffillo 2009 (103)	138	49	51	31.3	23	Fibrosis ≥ F3 (26%)	0.68 (0.57–0.78)	−1.455+0.676	5414	27100	0.7414.00	1.70.86	22(f,h,k); Indeterminate probability in 30% of cases
Parker 2009 (102)	50	57	38%	33.4	46	Fibrosis ≥ F2 (12%)	0.85 (0.80–0.91)	−1.455+0.676	100100	4493	1.7914.29	0.000.00	22(f,h,k)
						Fibrosis ≥ F3 (10%)	0.87 (0.81–0.93)	−1.455+0.676	100100	4293	1.7214.29	0.000.00
Wong 2010 (77)	247	51	45	28.0	36	Fibrosis ≥ F2 (43%)	0.62 (0.58–0.72)	−1.455+0.676	5512	7096	1.833.00	0.640.92	25; 34% of patients with indeterminate probability
						Fibrosis ≥ F3 (24%)	0.75 (0.67–0.83)	−1.455+0.676	7318	7096	2.434.50	0.390.85
Fujii 2009 (106)	122	59	61	27.0	NR	Fibrosis ≥ F2 (68%)	0.70 (0.65–0.77)	−1.455+0.676	6820	74100	2.6220.00	0.430.80	24(f)
						Fibrosis ≥ F3 (37%)	0.76 (0.70–0.85)	−1.455+ 0.676	8733	6497	2.4211.00	0.200.69
Sumida 2010 (105)	588	52	49	27.8	41	Fibrosis ≥ F3 (11%)	0.85 (NR)	−1.455+ 0.676	9233	6295	2.426.14	0.120.71	25; Multicenter Japanese study from 8 hepatology centers; 36% with indeterminate result
Pimentel 2010 (122)	158	36	77	41	16	Fibrosis ≥ F3 (14%)	0.85 (NR)	−1.455+ 0.676	9547	8095	4.759.4	0.060.55	24(f); 16% with indeterminate results
Summary estimates	4 studies assessed NAFLD fibrosis score for fibrosis ≥ F2					Fibrosis ≥ F2	4	−1.455+0.676	8236	7090	2.733.60	0.260.71	STARD score (median, range): 24 (22–25)
	13 studies (3064 participants) assessed NAFLD fibrosis score for fibrosis ≥ F3					Fibrosis ≥ F3	0.85 (0.81–0.90)	−1.455+ 0.676	90 (86–95)64 (59–70)	60 (56–65)97 (94–99)	2.2520.3	0.170.37
BARD score BMI ≥ 28 (1 point), AST/ALT ratio ≥ 0.8 (2 points), diabetes (1 point)
Harrison 2008 (87)	827	49	51	33	33	Fibrosis ≥ F3 (22%)	0.81 (NR)	2	91	66	2.68	0.13	23(f,k); Combines 3 variables in a weighted sum. The performance was comparable to the NAFLD fibrosis score in a subgroup from the same cohort and was particularly high in non-diabetic subjects, as a result of the increased OR of AST/ALT ratio in these patients: OR 15.6 (95% CI 8.4–28.9) in non-diabetic versus 4.3 (95% CI 2.5–7.4) in diabetic subjects. AST/ALT ratio increases by decreased clearance of AST as sinusoidal fibrosis progresses
Fujii 2009 (106)	122					Fibrosis ≥ F2 (69%)	0.63 (0.58–0.69)	2	50	76	2.08	0.66	24(f)
						Fibrosis ≥ F3 (38%)	0.69 (0.65–0.76)	2	65	72	2.32	0.49
Wong 2010 (77)	265	51	45	28.0	36	Fibrosis ≥ F2 (43%)	0.59 (0.54–0.65)	2	51	67	1.55	0.73	25
						Fibrosis ≥ F3 (23%)	0.64 (0.60–0.69)	2	62	66	1.82	0.58
Parker 2009 (102)	50					Fibrosis ≥ F2 (12%)	0.60 (0.55–0.66)	2	83	36	1.29	0.47	22(f,h,k)
						Fibrosis ≥ F3 (10%)	0.60 (0.54–0.65)	2	80	36	1.25	0.56
Ruffillo 2009 (103)	138	49	51	31.3	23	Fibrosis ≥ F3 (26%)	0.67 (0.56–0.77)	2	51	77	2.22	0.64	22(f,h,k)
Raszeja 2010 (104)	104	50	35	29.8	18	Fibrosis ≥ F3 (14%)	0.82	2	87	73	3.18	0.18	22(f,h,k)
Summary estimates	3 studies assessed BARD score for fibrosis ≥ F2					Fibrosis ≥ F2	0.60 (0.55–0.64)	2	61 (54–68)	60 (53–67)	1.53	0.66	STARD score (median, range): 24 (22–25)
	6 studies assessed BARD score for fibrosis ≥ F3					Fibrosis ≥ F3	0.78 (0.72–0.84)	2	72 (60–84)	64 (56–72)	2.00	0.44	STARD score (median, range): 23 (22–25)
Enhanced Liver Fibrosis (ELF) panel HA, PIIINP, TIMP-1
Guha 2008 (81)	192	49	36	32	31	Any fibrosis (stage 1–4; 59% patients)Fibrosis ≥ F2 (40% patients)Fibrosis ≥ F3 (23% patients)	0.76 (0.69–0.83)0.82 (0.75–0.88)0.90 (0.84–0.96)	−0.2070−0.10680.3576	617080	808090	3.103.508.01	0.490.380.22	24 (k); Combines plasma levels of 3 proteins and proteinases involved in early stages of hepatic collagen deposition. Only 14% of the patients left in the indeterminate gray area for advanced fibrosis, thus reducing the need for liver biopsy. The diagnostic performance for milder fibrosis is poor. This panel has been validated in a cohort of 112 pediatric NAFLD patients
Guha 2008 (combined panel) (81)	91	NR	NR	NR	NR	Any fibrosis (stage 1–4)Fibrosis ≥ F2Fibrosis ≥ F3	0.84 (0.76–0.92)0.93 (0.88–0.99)0.98 (0.96–1)	−5.002−0.995−0.2826	928991	528696	1.906.4022.81	0.150.130.09	24 (k); The combination of NAFLD fibrosis score plus ELF was tested in a subgroup of 91 patients from the original Guha cohort, yielding a higher diagnostic performance than either panel alone
Fibrometer Age, body weight, glucose, AST, ALT, ferritin, platelet count
Cales 2009 (83)	235	51	26%	29	24%	Fibrosis ≥ F2 (28%)Fibrosis ≥ F3 (19%)	0.94 (0.91–0.98)0.94 (NR)	≤ 0.611≥ 0.715NR	79NR	96NR	19.05NR	0.22NR	24 (k); Fibrometer has higher diagnostic accuracy for significant fibrosis than NAFLD fibrosis score (AUC 0.88; 95% CI 0.83–0.94; P = 0.008), while its accuracy for advanced fibrosis was comparable to NAFLD fibrosis score (AUC 0.93; P = 0.294)
Transient elastography (Fibroscan)
Kelleher 2006 (74)	129	49	51	31	31	Fibrosis ≥ F2 (50%)	0.86	8.7	81	78	3.68	0.24	22(e,f,w)
						Fibrosis ≥ F3		NA	NA	NA	NA	NA
Yoneda 2008 (75)	102	52	59	26.6	0	Fibrosis ≥ F2 (53%)	0.87	6.7	88	74	3.39	0.16	25; failure of LSM in 5% patients, all obese. Median
						Fibrosis ≥ F3 (28%)	0.90	9.8	85	81	4.47	0.19	LSM 7.8 kPa, IQR 18%
Nobili 2008 (76)	50	13	38	26	0	Fibrosis ≥ F2 (24%)	0.99 (0.92–0.99)	7.4	100	92	12.7	0.0	25; Success rate 89%; IQR/M 16%
						Fibrosis ≥ F3 (10%)	1 (0.94–1)	10.2	100	100	∞	0.0
Chang 2008 (79)	19	53	58	25.0	0	Fibrosis ≥ F2 (41%)	0.84 (0.79–90)	7.0	79	76	3.3	0.27	25; Success rate 88%; IQR 1.21
						Fibrosis ≥ F3 (23%)	0.93 (0.89–0.96)	8.7	84	83	5.0	0.19
Obara 2008 (78)	17	50	35	27.6	0	Fibrosis ≥ F2 (26%)	0.99 (0.94–1.0)	9.0	100	93	14.29	0.00	25; Success rate > 60% IQR NR
						Fibrosis ≥ F3 (21%)	0.93 (0.82–1.05)	11.0	75	93	10.71	0.27
Wong 2010 (77)	246	51	45	28.0	36	Fibrosis ≥ F2 (47%)	0.88 (0.78–0.97)	8.1	91	76	3.79	0.12	25; Success rate: 87% (failures of LSM mostly in patients
						Fibrosis ≥ F3 (29%)	0.99 (0.96–1.01)	15.8	100	95	20.00	0.00	with BMI > 30); IQR/M 17%
Summary estimates	6 studies assessed Fibroscan for fibrosis ≥ F2					Fibrosis ≥ F2	0.84 (0.79–90)	7.0	79	76	3.3	0.27	STARD score 25 (22–25)
	5 studies assessed Fibroscan for fibrosis ≥ F3					Fibrosis ≥ F3	0.94 (0.86–0.99)	8.7	94 (88–99)	95 (89–99)	18.8	0.06	STARD score 25 (25–25)

In the same row, the name of the score and its components are given.

Fibrosis was assessed on a 5-stage histological scale validated by NASH Clinical Research Network: 0 = absent; F1 = perisinusoidal or periportal; F2 = perisinusoidal and periportal; F3 = bridging; F4 = cirrhosis.

Collectively, fibrosis stage ≥ 2 is named significant; fibrosis stage ≥ 3 is named advanced (see text).

Summary outcome estimates are calculated including the training cohort and internal and external validation cohorts. If the score was assessed in only 1 study,^a then training and internal validation cohorts were included.

NAFLD fibrosis score: −1.675 + 0.037 × age (years) + 0.094 × BMI (kg/m²) + 1.13 × impaired glucose tolerance/diabetes (yes = 1; no = 0) + 0.99 × AST/ALT ratio – 0.013 × platelets (× 10⁹/L) – 0.66 × albumin (g/dL).

ELF panel: −7.412 + (ln(HA) × 0.681) + (ln(PIIINP) × 0.775) + (ln(TIMP-l) × 0.494).

Fibrometer: 0.4184 × glucose (mmol/L) + 0.0701 × AST (IU/L) + 0.0008 × ferritin (μg/L) – 0.0102 × platelet count (× 10⁹/L) – 0.0260 × ALT (IU/L) + 0.0459 × body weight (kg) + 0.0842 × age (yrs) + 11.6226.

STARD check-list:

(a) TITLE/ABSTRACT/KEYWORDS. Identify the article as a study of diagnostic accuracy (recommend MeSH heading ‘sensitivity and specificity’).

(b) INTRODUCTION. State the research questions or study aims, such as estimating diagnostic accuracy or comparing accuracy between tests or across participant groups.

STUDY METHODS

(c) Study population: The inclusion and exclusion criteria, setting and locations where data were collected.

(d) Participant recruitment: Was recruitment based on presenting symptoms, results from previous tests, or the fact that the participants had received the index tests or the reference standard?

(e) Participant sampling: Was the study population a consecutive series of participants defined by the selection criteria in item C and D? If not, specify how participants were further selected.

(f) Data collection: Was data collection planned before the index test and reference standard were performed (prospective study, 1 point) or after (retrospective study, 0 point)? TEST METHODS

(g) The reference standard and its rationale.

(h) Technical specifications of material and methods involved including how and when measurements were taken, and/or cite references for index tests and reference standard. Adequate biopsy fragment length? Liver biopsy processed according to standard pre-analytical/analytical procedures and scored following standard criteria?

(i) Definition of and rationale for the units, cut-offs, and/or categories of the results of the index tests and the reference standard.

(j) The number, training, and expertise of the persons executing and reading the index tests and the reference standard.

(k) Whether or not the readers of the index tests and reference standard were blinded to the results of the other test and describe any other clinical information available to the readers.

STATISTICAL METHODS

(l) Methods for calculating or comparing measures of diagnostic accuracy, and the statistical methods used to quantify uncertainty (e.g. 95% confidence intervals).

(m) Methods for calculating test reproducibility, if done.

RESULTS

(n) When study was performed, including beginning and end dates of recruitment.

(p) Clinical and demographic characteristics of the study population (at least information on age, gender, spectrum of presenting symptoms). Was the study population representative of the typical NAFLD spectrum?

(q) The number of participants satisfying the criteria for inclusion who did or did not undergo the index tests and/or the reference standard; describe why participants failed to undergo either test (a flow diagram is strongly recommended).

(r) Time-interval between the index tests and liver biopsy < 6 months, and any treatment administered in-between.

(s) Test results. Distribution of severity of disease (define criteria) in those with the target condition; other diagnoses in participants without the target condition.

(t) Test results. A cross-tabulation of the results of the index tests (including indeterminate and missing results) by the results of the reference standard; for continuous results, the distribution of the test results by the results of the reference standard.

(u) Test results. Any adverse events from performing the index tests or the reference standard.

(v) Estimates of diagnostic accuracy and measures of statistical uncertainty (e.g. 95% confidence intervals).

(w) How indeterminate results, missing data, and outliers of the index tests were handled.

(x) Estimates of variability of diagnostic accuracy between subgroups of participants, readers, or centers, if done.

(y) Estimates of test reproducibility, if done.

(z) Discussion. Discuss the clinical applicability of the study findings.

^a In the last column the score of the STARD check-list is reported, with the item(s) not satisfied by the study indicated in parentheses.

Sp = specificity; Se = sensitivity; FPHO = French public health organization; NR = not reported; TIMP-1 = tissue inhibitor of matrix metalloproteinase-1; PIIINP = N-terminal propeptide of type III collagen; HA = hyaluronic acid; CK-18 = cytokeratin-18; LSM = liver stiffness measurement; IQR/M = interquartile range/median ratio (reflects the variability of the validated measures; a ratio ≥ 30% is considered to indicate good accuracy and reproducibility of the measures).

The other group of non-invasive methods to assess liver disease severity in NAFLD includes biomarker panels. Biomarker panels combine routinely assessed clinical variables with different biochemical parameters: routine biochemical tests, markers of hepatocyte apoptosis or of hepatic collagen matrix remodeling, and adipose tissue-released cytokines. Multiple regression analysis of these variables yielded equations predictive of the probability of a single patient having the target condition (i.e. NASH, advanced fibrosis) (80–106).

A total of 21 non-invasive panels were identified; 13 were internally and/or externally validated in 32 studies and included in our meta-analysis (median, range STARD score of included studies 24, 19–25). Each study's characteristics and summary estimates of each test's diagnostic performance are reported in Table II.

Five panels identified the presence of advanced fibrosis among patients with NAFLD. The mean AUROC of these panels ranged 0.80–0.90. The Fibrotest, enhanced liver fibrosis (ELF) panel, combined panel, and NAFLD fibrosis score had better diagnostic accuracy than the BARD (BMI, AST/ALT Ratio, Diabetes) score (P < 0.02 for each of the four panels versus BARD score), while the AUROC of these panels did not differ among each other and among each of these panels and Fibroscan.

Fibrotest, ELF, and combined panel received only internal validation in the original study, so their reproducibility across different populations is unknown. The NAFLD fibrosis score has been most extensively validated for detecting advanced fibrosis. This score combines seven routinely measured clinical-biochemical parameters into a score; two cut-offs with distinct (high and low) probability of advanced fibrosis are provided. This panel has been validated in 13 studies, including 2 multicenter trials from America and Japan, enrolling a total of 3,064 patients of different ethnicities (including American Caucasians, blacks, and Hispanics, Europeans, and Asians), ages, and obesity and diabetes status. The median STARD score of these studies is 24 (range 22–25). Summary estimates of AUROC, sensitivity, specificity, LR+, and LR− of NAFLD fibrosis score for advanced fibrosis are reported in Table II and Figure 6.

Figure 6. Summary ROC curve and summary point of plasma CK-18 for diagnosing NASH (panel A), and of NAFLD fibrosis score (low cut-off, panel B), NAFLD (high cut-off, panel C), and Fibroscan (panel D) for advanced fibrosis.

A shortcoming of this panel is that in different studies 20%–58% of all patients fell in-between the two cut-offs and could therefore not be classified as having a high or low probability of advanced fibrosis (indeterminate probability of advanced fibrosis).

Six biomarker panels for detecting NASH met inclusion criteria; their summary AUROC ranged from 0.79 for NASHtest to 0.86 for NASH Predictive Index (P = NS among AUROC of different tests). Five of these panels (NASHtest, NASH Predictive Index, Obesity-related NASH Diagnostics, NASH Clinical Score, NAFIC score) have currently been validated exclusively in the original studies, and two of them (Obesity-related NASH Diagnostics and NASH Clinical Score) were tested exclusively in morbidly obese subject candidates for bariatric surgery: their reproducibility and generalizability remain therefore unknown. Plasma ELISA-detected cytokeratin-18, a marker of hepatocyte apoptosis, is the only marker for NASH that has been externally validated in nine independent studies (median STARD score 25, range 23–25) enrolling a total of 856 NAFLD patients of different ethnicities, BMI, and diabetes status). Summary AUROC, sensitivity, specificity, LR+, and LR− are 0.82 (95% CI 0.76–0.88), 0.78 (0.65–0.91), 0.86 (0.75–0.97), 6.13 (5.09–7.17), and 0.28 (0.26–0.30) (Table II; Figure 6). Even after excluding two studies enrolling morbidly obese subjects, not representative of the general NAFLD population, pooled estimates remained similar in magnitude to overall diagnostic accuracy (not shown). Notably, the multicenter NASH Clinical Research Network (CRN) study, the largest study so far on cytokeratin-18 fragments (CK-18), demonstrated that the addition of routinely available clinical-laboratory parameters to CK-18 measurement did not significantly improve its diagnostic performance.

An important limitation of these non-invasive methods is that they were all validated in a cross-sectional fashion, but their accuracy to monitor disease progression and treatment response remains largely unknown: only serum CK-18 and NASHtest/Fibrotest correlated accurately with post-treatment histological changes in three small trials (93,107,108).

In synthesis, three non-invasive methods were independently validated against LB for detecting NASH (serum CK-18) and advanced fibrosis (Fibroscan and NAFLD fibrosis score).

Other diagnostic issues

Oral glucose tolerance test (OGTT). Some authors suggested an OGTT is indicated in all NAFLD patients without known diabetes, based on the association between NAFLD and postprandial hyperglycemia. Current American Diabetes Association (ADA) guidelines recommend fasting plasma glucose (FPG) as the initial screening test for diabetes and OGTT only in patients with impaired fasting plasma glucose (IFG) (i.e. FPG of 100–125 mg/dL, 5.6–6.9 mmol/L) or when diabetes is strongly suspected on the basis of associated clinical risk factors (109). Five observational studies (total 472 NAFLD patients), all from Asian populations, with a median STROBE score of 19 (range 17–21), and one randomized controlled trial (RCT) (110–115) assessed this issue: 16%–63% of newly diagnosed cases of diabetes were based on isolated 2-h glucose elevation, with non-diabetic FPG, and 29%–52% of patients had impaired glucose tolerance (IGT). Most of these newly diagnosed cases of diabetes had IFG or at least one clinical risk factor for diabetes, indicating the need for OGTT. Further well designed studies are needed to identify subgroups of patients who might benefit most from an OGTT before this test can be routinely recommended to all NAFLD patients.

Carotid ultrasound. A systematic review of cross-sectional studies suggested that routine measurement of carotid intima-media thickening (IMT) might be implemented in all NAFLD patients, as they have a 13% higher risk of carotid IMT and an OR for carotid plaques of 3.13 (95% CI 1.75–5.58) (61). Awaiting further prospective confirmation, recent guidelines suggest a B-mode carotid ultrasonography is mostly cost-effective in patients without diabetes or established cardiovascular disease, at intermediate CVD risk (Framingham risk score 6%–20%) (116).

Discussion

Principal findings

Our analysis found that NAFLD confers an increased overall mortality, deriving from CVD and liver disease, and an increased risk of developing diabetes. These associations hold even when adopting different definitions of hepatic steatosis and after adjusting for the presence of metabolic syndrome, a cluster of cardio-metabolic risk factors of which NAFLD is considered the hepatic manifestation.

An important finding is that CVD mortality does not significantly differ between SS and NASH: however, histologically studies assessed CVD mortality, not incident CVD, and patients with NASH may die from liver-related disease before they can develop cardiovascular complications. Therefore, whether cardiovascular risk differs between SS and NASH requires further investigation. Another important finding of our analysis is the different hepatological prognoses of histological subtypes of NAFLD: compared to SS, NASH has a higher liver-related mortality, further increased by the coexistence of advanced fibrosis.

Included studies were heterogeneous with respect to the definition of NAFLD (biochemical, radiological, histological). We addressed this issue by subgrouping studies according to different definitions. The low sensitivity of liver enzyme elevation (particularly ALT) for NAFLD may have attenuated the strength of the association between fatty liver and assessed outcomes. Consistently, in both community and population-based studies, radiological/histological definitions of NAFLD were more strongly associated with outcomes than biochemical definitions. Among liver enzymes, GGT was associated with increased cardiovascular risk, while ALT was not. This may be explained by the fact that ALT is more liver-specific but less sensitive for steatosis, while GGT is more sensitive but less specific for steatosis, and serum GGT levels may be a marker of increased systemic or vascular oxidative stress rather than simply of increased liver fat content. Whether the combination of imaging and biochemical markers of NAFLD may enhance outcome prediction in NAFLD needs further investigation: in the SHIP study, the addition of serum GGT to ultrasonographic steatosis enhanced prediction of future all-cause and cardiovascular mortality compared to either method alone (24).

The different prognosis of NAFLD histological subtypes prompts screening all NAFLD patients for the presence of NASH, with or without advanced fibrosis. Despite its limitations, LB remains the gold standard for staging and monitoring NAFLD, but it is not feasible in all NAFLD patients.

Three non-invasive methods were independently validated to detect respectively NASH (ELISA-detected cytokeratin-18 fragments) and advanced fibrosis (NAFLD fibrosis score and Fibroscan).

NAFLD fibrosis score can be easily calculated with routine clinical parameters, has good diagnostic performance, is cheap, easy-to-calculate from routine parameters, and has been independently validated in populations of various ethnicities, BMI, and diabetic status. A shortcoming of this panel is that 20%–58% of all patients fall in-between the two cut-offs (indeterminate probability of advanced fibrosis) and could not therefore be classified.

Transient elastography (Fibroscan), the other externally validated method for detecting advanced fibrosis, is based on the principle than the stiffer the liver tissue, the faster the vibrations propagate through the tissue underlying the probe. Fibroscan measures liver stiffness in a cylindric volume of 1 per 4 cm, between 25 and 65 mm below the skin surface. Fibroscan accurately individuates advanced fibrosis in non-obese subjects, but cannot measure liver stiffness or may yield unreliable measures in obese patients due to the attenuation of ultrasound and elastic wave by the thick subcutaneous thoracic fat belt: in current studies mean BMI was < 30 kg/m², and obesity was the strongest predictor of Fibroscan failure. Consistently, in a 5-year prospective study of 13,369 examinations, a BMI > 30 kg/m² was associated with failed or unreliable LSM in 3.1%–15.8% of cases (117). Therefore, the diagnostic performance of Fibroscan in obesity, which is commonly present in patients with NAFLD, warrants improvement: in this context, a pilot experience with newly developed Fibroscan probes for obese patients showed encouraging results (118).

ELISA-detected serum CK-18 fragment is the only non-invasive marker differentiating NASH from SS that has been currently externally validated: its fair-to-good diagnostic accuracy (AUROC 0.82, sensitivity 0.78, specificity 0.87) precludes its use as the sole screening method, but it can be combined with other non-invasive methods or with LB to enhance prediction of progressive liver disease.

Implications for practice

The results of our analysis have important clinical implications: newly diagnosed NAFLD patients should be thoroughly assessed for their cardiovascular, metabolic, and liver-related risk (Table III). Concerning the last-mentioned, the early individuation of different histological subtypes is mandatory for the management of NAFLD patients: while SS has a benign hepatological course and may be managed with the same measures targeting associated cardio-metabolic risk factors, progressive NASH has an increased liver-related mortality, especially if advanced fibrosis coexists, and requires hepatological referral, with different clinical purposes. NASH without advanced fibrosis requires earlier treatment with experimental therapies to reverse necroinflammatory changes (119), while the coexistence of advanced fibrosis mandates tight monitoring for complications of cirrhosis (hepatocellular carcinoma, esophageal varices, liver failure), pending the discovery of effective anti-fibrotic therapies.

Table III. Suggested assessment of patients with NAFLD for general physicians.

1) Evaluation of metabolic risk
Steps	Comments
Screen for obesity and metabolic syndrome (ATP III criteria) (Level I)	Measure BMI, waist circumference, blood pressure, fasting plasma glucose (FPG), lipid profile
75-g OGTT(Level II)	Perform in NAFLD patients without known diabetes according to standard guidelines (i.e. ADA) to classify glucose tolerance
Assess insulin resistance (Level II)	Measure fasting insulin together with FPG and calculate fasting indexes of insulin resistance (HOMA-IR, QUICKI), which is associated with the severity of liver disease and has prognostic value in NAFLD; OGTT-derived indexes for research purposes only
2) Evaluation of cardiovascular (CV) risk
Assess CV risk (Level I)	Measure total and HDL-C and calculate LDL-C if Tg < 200 mg/dL (2.26 mmol/L) and evaluate CVD risk scores (i.e. Framingham risk score (FRS)) (Level I). Perform B-mode carotid ultrasonography in patients without diabetes or established cardiovascular disease, at intermediate CVD risk (FRS 6%–20%) according to recent guidelines (Level II) (112)
3) Evaluation of liver-related risk
Assess the risk of progressive liver disease (Level I)	Simple steatosis (SS) has a benign hepatological prognosis and may be managed by general physicians with measures aiming at reducing overall cardio-metabolic risk. NASH, with or without advanced fibrosis, has a progressive course and poorer prognosis and mandates referral to hepatologist for tight follow-up and liver-specific experimental treatment
Non-invasively screen for advanced fibrosis (Level I)	Apply NAFLD fibrosis score and/or Fibroscan (non-obese subjects): if high probability of advanced fibrosis refer to gastroenterologist for liver biopsy and assessment of complications of cirrhosis (hepatic failure, portal hypertension, esophageal varices, HCC)
Assess risk of NASH by CK-18 measurement (Level II)	Refer to gastroenterologist for liver biopsy NAFLD patients with a high probability of NASH, with or without advanced fibrosis after non-invasive screening tests. Liver biopsy remains mandatory to stage and monitor the course of liver disease in NASH and if diagnosis of NAFLD is in doubt

ATP = Adult treatment panel; ADA = American Diabetes Association; OGTT = oral glucose tolerance test; C = cholesterol; Tg = triglyceride; HCC = hepatocellular carcinoma.

Currently, no non-invasive panel alone is accurate enough to replace LB (still the gold standard despite its limitations), and a clinically feasible approach could combine non-invasive panels with LB to biopsy only patients with a high probability of progressive disease after non-invasive assessment, thus reducing the need for LB. Our analysis suggests two non-invasive methods can be used to individuate patients with a high probability of advanced fibrosis, while serum CK-18 has fair accuracy to detect NASH.

The optimal time interval for monitoring non-biopsied patients is unknown: mean (± SD) rate of fibrosis progression in untreated subjects has been found to be 0.03 ± 0.40 stages/year in overall NASH and 0.41 ± 0.40 stages/year in those NASH patients who progressed (38% of overall cases of NASH) (119). Assuming these figures can be extrapolated to NASH patients under treatment, non-invasive monitoring for advanced fibrosis may be repeated every 1–2 years.

Implications for research

Since no non-invasive panel alone can replace biopsy, a strategy combining non-invasive methods with LB may optimally individuate patients with NASH, with or without fibrosis, and reduce the need for LB: three sequential diagnostic algorithms are presented in Figure 7. The initial step is the calculation of NAFLD fibrosis score, which is feasible in all patients and, if yielding a high risk of advanced fibrosis, can reliably direct patients to a gastroenterologist for evaluation of cirrhosis. The subsequent step is the assessment of the probability of having NASH by serum CK-18. Fibroscan can be added to NAFLD fibrosis score to further screen for advanced fibrosis in non-obese subjects. Adopting one of these three algorithms, 85%, 88%, and 90%, respectively, of patients with NASH could be individuated, while 22%, 23%, and 26% of patients with SS would (unnecessarily) be biopsied.

Figure 7. Proposed diagnostic algorithms combining non-invasive methods and liver biopsy. Algorithm 1 can be applied to any patient with NAFLD; algorithm 2 and 3 can be applied only to non-obese patients with NAFLD, since Fibroscan has limitations when BMI > 30 kg/m². In the third algorithm Fibroscan is applied to patients with a low and intermediate score for fibrosis, which implies a wider availability of Fibroscan, has the advantage of individuating all patients with advanced fibrosis, but it biopsies more patients with simple steatosis than the first model (13 versus 10 patients).

All models are conservative estimates of diagnostic performance of NAFLD fibrosis score and Fibroscan, i.e. they assume that all biopsied patients without advanced fibrosis in the first 2 steps have simple steatosis, not NASH, since the diagnostic performance of NAFLD fibrosis score and Fibroscan for detecting NASH is currently unknown.

Explanation: Based on our meta-analysis, three sequential diagnostic algorithms are proposed to individuate NASH, with and without advanced fibrosis; such algorithms require further prospective evaluation in large-scale multicenter cohort studies.

If a sample of 100 NAFLD subjects underwent a single LB, NASH would be found in 40 subjects, of which 20 would have advanced fibrosis (LB-to-all approach). In combined approaches, the first step should be the identification of patients with NASH and advanced fibrosis, requiring prompt gastroenterological referral. To this aim, NAFLD fibrosis score and Fibroscan may be combined with LB.

If we apply NAFLD fibrosis score to our NAFLD sample, this would direct 15 patients (13 with advanced fibrosis) to biopsy and leave 50 patients with a low probability and 35 patients with an indeterminate probability. If patients are obese, then the next step would target NASH with CK-18 measurement. At the end of diagnostic flow, 44 patients would be referred for LB (algorithm 1).

In non-obese subjects Fibroscan is reliable and accurate and may be applied after NAFLD fibrosis score to screen for advanced fibrosis patients with an indeterminate probability (algorithm 2) or with both an indeterminate and low probability (algorithm 3), depending on the local availability and experience.

At the end of the diagnostic flow, algorithm 2 would direct 45 patients (25 NASH, of whom 18 with advanced fibrosis) to LB; algorithm 3 would direct 49 patients (35 NASH, of whom 20 with advanced fibrosis) to LB.

Future research should also evaluate non-invasive or combined approaches not only for screening NASH with or without advanced fibrosis, but also for monitoring the course of disease and treatment response. Follow-up studies should also assess the prognostic value of non-invasive methods for predicting clinical outcomes, including overall and liver-specific morbidity and mortality: promising data have currently come from the ELF panel and Fibrotest, which were able to predict long-term prognosis more accurately than LB in chronic hepatitis C and primary biliary cirrhosis (120–122).

Acknowledgements

This work would have never been possible without the help of the following authors, whom we thank for their courtesy in providing details about their published studies. None of these persons received compensation for the work performed.

Peter Jepsen (Department of Clinical Epidemiology, Aarhus University Hospital, Denmark), Zobair Younossi (Center for Liver Diseases, Inova Fairfax Hospital, Falls Church, Virginia 22042, USA), Nila Rafiq (Center for Liver Diseases, Inova Fairfax Hospital Falls Church, VA 22042, USA), Leon Adams (School of Medicine and Pharmacology, University of Western Australia, Nedlands, WA, Australia), Cecilia Soderberg (Department of Medicine, Karolinska Institute, Solna, Stockholm, Sweden), Mattias Ekstedt (Division of Gastroenterology and Hepatology, Department of Molecular and Clinical Medicine, University Hospital, Linköping, Sweden), Stergios Kechagias (Division of Internal Medicine, Department of Medicine and Care, University Hospital, Linköping, Sweden), Praveen Guturu (Department of Internal Medicine, UTMB, Galveston, Texas, USA), Raza Malik (Department of Medicine, Beth Israel Deaconess Medical Center, Harvard Medical School, Boston, Massachusetts, USA), Hideki Fujii (Department of Hepatology, Osaka City University, Osaka, Japan), Dana Crisan (University of Medicine and Pharmacy ‘Iuliu Hatieganu’, Cluj-Napoca, Romania), Richard Parker (University Hospitals of Bristol, NHS Foundation Trust, Bristol, UK), Noriyuki Obara and Yoshiyuki Ueno (Division of Gastroenterology, Tohoku University Graduate School of Medicine, Aoba, Sendai 980-8574, Japan), and Jason Chang (Department of Gastroenterology & Hepatology, Singapore General Hospital, Singapore).

Declaration of interest: No author has any present or past conflict of interest to report. Giovanni Musso is independent of any commercial funder, performed the statistical analysis, wrote the manuscript, had full access to all the data in the study, and takes responsibility for the integrity of the data and the accuracy of the data analysis. This work was supported in part by the Piedmont Region Funds Comitato Interministeriale per la Programmazione Economica 2008, which were employed for data collection.