The diagnostic value of biomarkers (SteatoTest) for the prediction of liver steatosis

A total of 2,272 subjects were analyzed (Figure 1), being 884 subjects included in the biomarker validation study, distributed as follows: 310 patients in the training group; 171 in the validation group 1; 201 in the validation group 2; 62 in the validation group 3; and 140 subjects in the control group. The 1,388 non-included patients were not significantly different from the 884 patients integrated in the validation assay (data not shown).

Patients included in the 4 groups were similar in age with a predominance of male subjects (range 61–76%). The prevalence of steatosis greater than 5% (grades 2 to 4) varied from 11% in hepatitis C virus (HCV) cured patients to 94% in patients with ALD. In all groups, at least one metabolic risk factor was observed in more than 50% of included patients. Patients in group 3 with alcoholic liver disease were more often male, older, had smaller liver biopsies, more metabolic risk factors, more extensive fibrosis and more grades 2–4 steatosis than the three other groups. Validation group 2 with HCV cured patients had quasi-normal characteristics with normal liver tests and only 11% grade 2–4 steatosis.

In the training group the most significant components associated with the presence of grade 2–4 steatosis in univariate analysis were body mass index (BMI), age, ALT, aspartate aminotransferase (AST), GGT, glucose, and triglycerides. The logistic regression defining the ST included 12 components – ALT, α₂-macroglobulin (A2M), apolipoprotein A-I (ApoA1), haptoglobin, total bilirubin, GGT, cholesterol, triglycerides, glucose, age, gender and BMI. In logistic regression analyses, the most significant components were BMI (P = 0.0002), GGT (P = 0.002), ApoA1 (P = 0.01), A2M (P = 0.02), ALT (P = 0.03) and triglycerides (P = 0.04). In the validation group, similar differences were observed, most significantly for BMI, GGT, ALT and triglycerides (Table 2).

The median ST value was 0.13 in fasting controls; 0.18 in non-fasting controls; 0.14 in blood donors; 0.26 in patients without steatosis; 0.43 in grade 1 steatosis; 0.62 in grade 2; 0.70 in grade 3; and 0.75 in grade 4. Because there were not a sufficient number of patients with grade 3 and 4, these two groups were combined (Figure 2).

The values {Area under the ROC curves (AUROCs)} of ST, GGT and ALT for the diagnosis of grades 2–4 steatosis, in the training and validation groups, are given in Table 3. ST had higher AUROCs: {0.79 (SE = 0.03)} in training group; 0.80 (0.04) in validation group 1; 0.86 (0.03) in validation group 2; and 0.72 (0.05) in validation group 3. These were always significantly higher than the AUROCs of GGT and significantly higher than the AUROCs of ALT, for the training group and validation group 1 (Table 3). The distribution of ST, GGT and ALT, according to the severity of steatosis, is illustrated in Figure 2 for the training and validation groups.

The diagnostic values of ST, GGT and ALT according to cutoffs are shown in Table 4. For the diagnosis of grade 2–4 steatosis, the sensitivity of ST at the 0.30 cut-off was 0.91, 0.98, 1.00 and 0.85 and the specificity at the 0.70 cut-off was 0.89, 0.83, 0.92, and 1.00, for the training and validation groups, respectively.

A total of 75 patients were included with biopsy at baseline and at follow-up. Among them, 23 had an improvement of steatosis (one of 3 grades, two of 2 grades and twenty of one grade); 43 had no change in steatosis grade; and 9 had worsening of one grade. ST significantly decreased in 23 patients with steatosis improvement at biopsy from 0.60 (SE = 0.05) to 0.41 (0.05), a significantly greater difference (P = 0.001) than that observed in 52 patients without biopsy improvement: from 0.44 (0.03) to 0.31 (0.03).

A total of 884 subjects were included in the integrated database combining the training group, the three validation groups and the control group. Of these, 75 patients with HCV were investigated twice (once before and then after treatment), and 29 volunteers were investigated twice (while fasting and, then, non-fasting). There was a very significant overall correlation between ST and the steatosis grades from controls to S3 (Figure 3). For ST, there was a significant difference between all histological grades by Tukey-Kramer multiple comparison test for all pairwise differences between means (P < 0.05). For GGT and ALT, there was no significant difference between S0 and S1. For ALT, there was no significant difference between S0 and S2, S1 and S2, and S2 and S3, either. ST has higher AUROC, 0.80 (0.02) than all the isolated components for the diagnosis of steatosis grade 2–4: ALT, GGT (Table 3), triglycerides 0.63 (0.02), BMI 0.61 (0.02), glucose 0.61 (0.02), bilirubin 0.60 (0.02), ApoA1 0.56 (0.02), A2M 0.56 (0.02) and cholesterol 0.53 (0.02) – all P values < 0.03.

A cut-off of 0.30 had 90% sensibility and a cut-off of 0.70 had 88% specificity permitting to achieve useful predictive values for steatosis grade 2–4, 93% negative predictive value (NPV) and 63% positive predictive value (PPV) for a steatosis prevalence of 30% (Table 4). The 90% specificity was obtained for a 0.72 cut-off with a corresponding 63% PPV. The overall percentage of patients classified with at least 90% sensitivity or 90% specificity was 59% (363+156/884).

Among the 744 patients with biopsy, for the diagnosis of steatosis 3–4, the ST AUROC was 0.79 (0.02), significantly higher than GGT 0.74 (0.02) (P = 0.03), and ALT was 0.71 (0.02) (P = 0.007). The 90% sensitivity was obtained for a 0.32 cut-off; the 90% specificity was obtained for a 0.81 cut-off.

ST is a continuous linear biochemical assessment of steatosis grade. It provides a numerical quantitative estimate of liver steatosis ranging from 0.00 to 1.00, corresponding to a steatosis scoring system of grades S0 to S4. Among the 140 controls, the median ST value (± SE) was 0.08 ± 0.004 (95th percentile, 0.23). Among the 744 patients with liver biopsy, the ST conversion was 0.000 – 0.3000 for S0; 0.3001 – 0.3800 for S0-S1; 0.3801 – 0.4800 for S1; 0.4801 – 0.5700 for S1-S2; 0.5701 – 0.6700 for S2; 0.6701 – 0.6900 for S2-S3S4; and 0.6901 – 1.000 for S3-S4.

Ultrasonography has been preformed together with ST and biopsy in 304 patients. Concordance between steatosis diagnosed, at ultrasonography and at biopsy, was lower (kappa coefficient = 0.32 ± 0.05) than the concordance with ST (at 0.50 cut-off, kappa = 0.44 ± 0.06; P = 0.02), as well as lower AUROC 0.65 ± 0.03 for ultrasonography versus 0.78 ± 0.03 for ST (P = 0.001). The ST values according to the presence of histological and radiological steatosis are given in Table 5.

A total of 635 (85%) patients had a time lapse between biopsy and serum smaller than one month. The AUROC of ST was similar in those patients (0.77, 95% CI 0.73–0.80) than in the 109 (15%) patients with greater lapse (0.82, 95% CI 0.72–0.89; P = 0.36). A total of 670 (78%) patients had a biopsy sample length smaller than 20 mm. The AUROC of ST was slightly smaller in those patients (0.76, 95% CI 0.71–0.79) than in the 161 (15%) patients with greater sample (0.82, 95% CI 0.74–0.88; P = 0.10).

Consecutive patients who were included were those with an available serum sample, a liver biopsy, and a time interval between serum sampling and biopsy of less than three months (Figure 1).

These patients were retrospectively included for this specific analysis, but had been analyzed in previous prospective validation studies of FT between September 2000 and August 2004 [23, 24, 27, 38]. All were patients hospitalized in the of Hepato-Gastroenterology department of Groupe Hospitalier Pitié-Salpêtrière for NAFLD, hepatitis C and B, and ALD.

These patients were retrospectively analyzed from a study of steatosis in patients with chronic hepatitis C [33]. For this purpose, previously non-treated patients of a prospective multicentre randomized trial of pegylated-Interferon and ribavirin were included. The biomarkers and the biopsy results at baseline were used.

These patients were those from the patients of the same randomized trial [33] who had been "cured" – they had a sustained virologic response, with undetectable HCV RNA, at the end of treatment and 24 weeks after the end of treatment. The biomarkers and the biopsy results performed 24 weeks after the end of treatment were used. This group can be considered to be a validation group of non-viral steatosis because possible viral steatosis had been cured by the treatment [33].

These patients were retrospectively included for this specific analysis but had been prospectively included between 1998 and 2000 in a cohort of alcoholic patients for which one primary endpoint was the identification of biochemical markers. The details of this cohort have been recently published in a validation study of FT [26]. All were patients hospitalized in the Hepato-Gastroenterology Department of Hôpital Antoine Béclère, for complications of alcoholic liver disease.

Non-inclusion criteria included non-available serum, non-available biopsies and biopsy and serum samples which had been collected more than 3 months apart (Figure 1). Patient characteristics are given in Table 1.

This included a group of, apparently, healthy volunteers who had been previously included in a validation study of FT, in fasting and non-fasting conditions [39]. A group of non-fasting blood donors were also prospectively included.

Common rules were applied to the different groups. Liver biopsy specimens were processed using standard techniques. Patients with viral hepatitis were evaluated for fibrosis and grade of activity according to the METAVIR scoring system, for which reproducibility had previously been established [40]. Patients with ALD and NAFLD were evaluated with modified staging and grading scores [41‐44]. Fibrosis was staged on a scale of 0 to 4: F0 – no fibrosis; F1 – portal fibrosis or perivenular fibrosis without septa; F2 – few septa; F3 – numerous septa without cirrhosis; and F4 – cirrhosis. Activity (the intensity of necroinflammatory activity mostly based on necrosis) was scored as follows: A0 – no histologic activity; A1 – mild activity; A2 – moderate activity; and A3 – severe activity. Steatosis was scored from 0 to 4 with a four grades scoring system from S0 to S4: S0 – no steatosis; S1 – mild 1 to 5% (% of hepatocytes containing visible macrovesicular steatosis); S2 – moderate 6 to 32%; S3 – marked 33 to 66%; and S4 – severe 67 to 100% [33]. The main histological criterion was the presence of steatosis grade 2–4 (between 6 to 100%). A single pathologist per group, unaware of patient characteristics, analyzed the histological features (Frederic Charlotte for the training group, Zack Goodman for validation groups 1 and 2, and Dominique Capron for validation group 3).

A new panel (ST, Biopredictive, Paris, France, patent pending) was constructed in the training group combining the 6 components of the FibroTest-ActiTest (patented artificial intelligence algorithm USPTO 6,631,330) adjusted for age, gender and BMI, plus serum glucose, triglycerides and cholesterol. ST scores of range from zero to 1.00, with higher scores indicating a greater probability of significant lesions. FT and AT (Biopredictive, Paris, France; FibroSURE LabCorp, Burlington, NC, USA) were determined as has been previously published [23, 38, 39]. The published recommended pre-analytical and analytical procedures were used [23, 38, 39, 45, 46]. In the training and control groups, GGT, ALT, serum glucose, triglycerides, cholesterol, and total bilirubin were measured by Hitachi 917 analyzer or Modular DP analyzers (both Roche Diagnostics Mannheim, Germany) using the manufacturer's reagents. A2M, ApoA1, and haptoglobin were measured using an automatic nephelemeter BNII (Dade Behring; Marburg, Germany). In validation groups 1 and 2, GGT, ALT, serum glucose, triglycerides, cholesterol, and total bilirubin were measured using Hitachi 747 or 911 (Roche Diagnostics, Indianapolis, IN, USA) with the manufacturer's reagents. ApoA1, A2M and haptoglobin were determined in serum samples using an automatic nephelometer BNII (Dade Behring, Marburg, Germany). In validation group 3, ALT, GGT, serum glucose, triglycerides, cholesterol, total bilirubin and haptoglobin were measured by autoanalyzer (Olympus AU 640 Automate) using manufacturer's reagents (Olympus, Rungis, France); A2M and ApoA1 were measured using an automatic nephelometer (BNII, Dade Behring, Marburg, Germany). All coefficients of variation assays were lower than 10%.

Ultrasonography reports have been retrospectively analyzed for the presence or absence of radiological steatosis in the validation group, blindly to histological and biochemical data.

The primary outcome was grade 2, 3 or 4 of steatosis (S2S3S4). The cause of discordance between the presence of S2S3S4 steatosis, as predicted by biochemical markers and biopsy was attributed according to respective risk factors of failure, as previously detailed [38]. Significant discordance was defined as discordance in predicting grades S2S3S4 and a 30% or greater difference in steatosis percentage, as predicted by ST or as observed in the biopsy sample. Risk factors of ST failure were hemolysis, Gilbert's disease, acute inflammation and extra-hepatic cholestasis. Risk factors of biopsy failure were biopsy size (less than 25 mm) and fragmentation (more than one fragment). Failure attributable to biopsy (false negative) was suspected when the biopsy length was less than 15 mm and fragmented with the additional presence of, at least, one metabolic risk factor.

Statistical analysis used Fisher's exact test, the chi-square test, Student's t-test and the Mann-Whitney test; variance analysis used the Bonferroni all-pair wise and the Tukey-Kramer multiple-comparison tests to take into account the multiple comparisons, and multiple logistic regression the for multivariate analysis [47]. The diagnostic values of the markers were assessed using sensitivities, specificities, PPVs and NPVs and AUROCs [47]. Corresponding steatosis grades were calculated from median ST scores and 95% confidence intervals observed in 744 patients and 140 controls. AUROCs were calculated using the empirical non-parametric method according to Delong et al. [48] and compared using the method of Zhou et al. [49]. The binomial approach was used only when the sample size population was less than 30 [50]. For all analyses, two-sided statistical tests were used; a P-value of 0.05 or less was considered significant. Number Cruncher Statistical Systems 2003 software (NCSS, Kaysville, Utah, USA) was used for all analyses [47].

A sensitivity analysis was also performed for determining the accuracy of the markers for the primary outcomes according to biopsy sample size (less than 20 mm or more) and to time lapse between serum and biopsy (less than 4 weeks or more).