Our results highlight the utility of a new panel of biochemical markers (ST) for the prediction of steatosis of different origins. A cut-off of 0.30 had 90% sensibility and a cut-off of 0.72 had 90% specificity permitting to achieve useful predictive value, 93% NPV and 63% PPV for a steatosis prevalence of 30%. These predictive values are far from perfection, particularly for PPV; however, already predictive and significantly higher than those of previous usual markers GGT, ALT and ultrasonography, as demonstrated by the increase of AUROCs. This benefit was observed for the most frequent chronic liver diseases: chronic viral hepatitis, and alcoholic and non-alcoholic fatty liver diseases.
We have not identified any reports of a single or a combination of biomarkers with accurate value for the diagnosis of steatosis in different chronic liver diseases. Marceau et al observed in 551 severely obese patients with liver biopsy that steatosis was associated with male gender, age, BMI, waist/hip ratio, diabetes, systolic blood pressure, fasting blood sugar, triglycerides, and non-HDL cholesterol, but no diagnostic algorithm was provided . Papadia et al.  observed in 1000 obese patients an association between steatosis and AST, ALT, AST/ALT ratio, body weight, waist/hip ratio, serum glucose, serum triglycerides, BMI, GGT, age, and unconjugated bilirubin using regression analysis . No panel was constructed and they concluded that no reliable biochemical marker could identify patients with severe steatosis with sufficient sensitivity for avoiding liver biopsy. Loguercio et al.  observed that in 305 patients with abnormal GGT or ALT, age, ferritin and tissue 4-hydroxynonenal were associated with steatosis. On multivariate analysis, no single factor was found to be an independent predictor .
In the present study, the predictive value of ST was related to the discriminant values of its different components. The most striking observation was that the combination of 12 parameters allowed a very significant increase in the diagnostic values of isolated GGT or ALT. The diagnostic value of ALT was better than that of GGT, as assessed by AUROCs in all the different groups. This is surprising as an elevated GGT is generally thought to be a serum marker of steatosis and elevated transaminases to be a marker of NASH. A better association between ALT and steatosis versus GGT and steatosis has also been observed using proton magnetic resonance imaging .
The diagnostic values of GGT, ALT, triglycerides, cholesterol, glucose and BMI were expected, because they had been previously associated with steatosis of different origins [3, 29, 31]. Those biomarkers are also associated with insulin resistance and triglyceride deposition in the liver . ApoA1 is highly associated with HDL-cholesterol and a negative association was also expected with steatosis . The advantage of combining biomarkers of steatosis and those more specific for fibrosis such as A2M, haptoglobin and bilirubin is to adjust the predictive values according to the associated stage of fibrosis. In the present study we observed that the grade of steatosis in patients with extensive fibrosis was significantly lower than in patients without extensive fibrosis (data not shown).
Our study has several limitations that must be acknowledged. Firstly, despite the use of prospective cohorts of patients, our study was not a classical prospective study. The validation groups consisted of previously studied groups of patients: groups 1 and 2 were from a prospective randomized trial with a previous publication on steatosis , and group 3 was a prospective cohort of patients with alcoholic liver disease from a study which had been published for validation of fibrosis biomarkers . There were three different pathologists but very skilled in these scoring systems and expert in variability studies. The analyses of histological specimens and biochemical markers were performed blindly, and the recommended pre-analytical and analytical procedures were respected for most of the components. The analytical variability of cholesterol, triglycerides and glucose should be assessed.
A second limitation was the relatively small number of patients with grade 3 and 4 steatosis. We observed a non-significant difference between ST medians, 0.70 for grade 3 versus 0.75 for grade 4. Due to the small sample size of patients with grade 3–4 steatosis in the validation groups, further studies should be performed in order to determine whether ST could discriminate between patients with marked steatosis (between 30 and 66%) and those with severe steatosis (over 66%). Grade 3 and 4 steatosis is more frequent in patients with NAFLD and further studies must be performed in these patients.
In patients with NAFLD, a liver biopsy is more usually obtained for identifying additional features of steatohepatitis (hepatocellular ballooning, lobular inflammation, Mallory's hyaline) which may be associated with and/or predictive for the development of pericellular and/or periportal fibrosis. FT has been already validated for the diagnosis of fibrosis in NAFLD  and ALD . Studies on biomarkers of steatohepatitis (NashTest, AshTest) are also in progress (personal communication of Thierry Poynard). Combination of those non-invasive markers should help the physician in the management of NAFLD and ALD.
A third limitation was not having compared prospectively the serum biomarkers with imaging techniques such as ultrasonography [28, 32, 34] and proton magnetic resonance imaging . In the retrospective analysis of the training population, we observed that ST had a higher diagnostic value than the routine ultrasonography with higher AUROCs. It has been already observed that the sensitivity of ultrasonography is low in obese patients  for the diagnosis of steatosis. Proton magnetic resonance imaging is expensive; nevertheless, a validation of ST versus proton magnetic resonance imaging would be quite interesting.
In contrast with the above mentioned limitations, one advantage of the present design was the inclusion of heterogeneous patients in the training group with different causes of chronic liver disease as well as the validation of the diagnostic values in more homogeneous groups. Validation groups 1 and 3 included very homogeneous patients, with chronic hepatitis C and ALD, respectively. The advantage of validation group 2 was the inclusion of a group of patients clinically and biologically close to a "normal" population, as these patients are sustained virologic responders and had quasi-normal liver function tests. This population offered the unique opportunity of having liver biopsies in subjects with normal profiles – not possible, for example, in blood donors. The intra and inter-laboratory variability has been studied for the 6 FT components and those studies should also be performed for cholesterol, triglycerides and glucose. We did not find any significant differences in ST AUROCs according to ethnicity (data not showed) .
As discussed for liver fibrosis, it is also possible that the limitations of liver biopsy (sampling error and pathologist concordance) did not allow a perfect area under the curve to be reached . In hepatitis C the ideal gold standard would be at least a 40 mm length biopsy sample. Bedossa et al.  recommend, at least, 25 mm; but the coefficient of variation decreases up to 40 mm. In chronic hepatitis C, 18 % of discordance in fibrosis staging has been attributed to liver biopsy failures (mainly due to small sample size) and 2% to FT (due to hemolysis, inflammation and Gilbert's syndrome) . For liver steatosis, there is also a sampling variability with discordance in 22% of patients . In the present study, we observed discordance between steatosis assessed by ST and that assessed by biopsy, in 10% to 21% according to patient's group. Several discordant cases seem to be attributable to biopsy (false negatives of biopsy) as the quality was poor and, at least, one metabolic risk factor was present. Significant discordance was more often observed in patients with extensive fibrosis. We previously suspected a risk of greater variability in assessing fibrosis when steatosis was present but the inverse could be also true: a greater variability in assessing steatosis in case of cirrhotic or pre-cirrhotic stages .
ST is not a perfect diagnostic tool, but has several advantages over other proposed strategies for steatosis management. The 12 components of ST are readily available. FibroTest-ActiTest is now available in several different countries, including the USA (FibroSure™), with a quality charter for laboratories for reducing inter-laboratory variability [23, 30, 38, 39]. As demonstrated in the present study, ST allowed the assessment of steatosis in patients with paired biopsy. This could be very useful for the follow-up of patients. This has been validated in HCV patients before and after treatment and should be validated in patients with ALD and NAFLD with paired biopsies.
There is no specific approved treatment for steatosis. Recommendations depend on the cause. There is wide agreement for the cessation of alcohol consumption in heavy drinkers, weight reduction in obese patients, and the treatment of diabetes and hyperlipidemia [1–4]. In patients with chronic hepatitis C and genotype 3, 50% of the patients treated and who have a sustained virologic response have a disappearance of liver steatosis at the second biopsy . Bellentani et al.  recommended that subjects with elevated ALT or GGT should be screened for steatosis using hepatic ultrasonography. They suggested that the demonstration of hepatic steatosis should prompt a reduction of caloric and alcohol intake and follow-up with both ultrasonography and biochemical tests. When clinically indicated, a liver biopsy for assessing the degree of fibrosis and inflammation could be performed.