Direct Comparison of the Specialised Blood Fibrosis Tests FibroMeterV2G and Enhanced Liver Fibrosis Score in Patients With Non-alcoholic Fatty Liver Disease From Tertiary Care Centres

Maeva Guillaume; Valerie Moal; Cyrielle Delabaudiere; Floraine Zuberbuhler; Marie-Angèle Robic; Adrien Lannes; Sophie Metivier; Frederic Oberti; Pierre Gourdy; Isabelle Fouchard-Hubert; Janick Selves; Sophie Michalak; Jean-Marie Peron; Paul Cales; Christophe Bureau; Jerome Boursier

Disclosures

Aliment Pharmacol Ther. 2019;50(11):1214-1222. 

In This Article

Results

Patients

The characteristics of the 417 patients included in the study (335 in Angers and 82 in Toulouse) are detailed in Table 3. Mean age was 56.1 ± 12.2 years, 59.2% were male. Mean body mass index was 33.3 ± 6.6 kg/m2 and 48.2% were diabetic. Mean biopsy length was 29 ± 11 mm, 90.1% were ≥15 mm long and 79.5% were ≥20 mm long. The prevalence according to fibrosis stages was: F0: 9.1%, F1: 23.5%, F2: 27.3%, F3: 32.4% and F4: 7.7%. The mean NAFLD Activity Score was 3.7 ± 1.7 and 291 patients (70.3%) had NASH.

Comparison of Direct Markers of Liver Fibrosis and Blood Fibrosis Tests

There was an excellent agreement between the hyaluronic acid test from Siemens and the hyaluronic acid test from FUJIFILM Wako (intraclass correlation coefficient: 0.918, P < .001). The five direct markers of liver fibrosis (alpha2-macroglobulin, hyaluronic acid from FUJIFILM Wako, hyaluronic acid from Siemens, PIINP and TIMP-1) and the four blood fibrosis tests (NFS, FIB4, ELF and FibroMeterV2G) were significantly correlated with histological fibrosis stages with Rs ranging from 0.41 to 0.52 (all P < .001; Table S3 and Figure S1). AUROCs and Obuchowski indexes of the direct markers of liver fibrosis were not significantly different from those of the simple fibrosis tests NFS and FIB4 (Table 4 and Table S4 for pairwise comparisons). The specialized blood fibrosis tests ELF and FibroMeterV2G were significantly more accurate for advanced fibrosis than their composite markers or the simple fibrosis tests NFS and FIB4. In addition, Obuchowski indexes for ELF and FibroMeterV2G were significantly better than those for NFS and FIB4. There was no significant difference between ELF and FibroMeterV2G when AUROCs or Obuchowski indexes were compared. ELF and FibroMeterV2G were significantly correlated with Rs at .62 (P < .001, Figure 1). Considering that they were the two most accurate tests in our population, we then focused the statistical analyses on FibroMeterV2G and ELF.

Figure 1.

Correlation between FibroMeterV2G and Enhanced Liver Fibrosis (ELF) score (Spearman correlation coefficient Rs = 0.62, P < .001)

3.3 FibroMeterV2G versus ELF

Binary Diagnosis of Advanced Fibrosis. The diagnostic cut-off corresponding to the highest Youden index was 0.434 for FibroMeterV2G and 9.3 for ELF. Using these cut-offs, ELF and FibroMeterV2G showed a similar accuracy for the diagnosis of advanced fibrosis with 72% well-classified patients, 70% sensitivity, 75% specificity, 80% negative predictive value and 65% positive predictive value (Table 5). ELF <9.3 with FibroMeterV2G <0.434 provided 86.0% negative predictive value for advanced fibrosis, whereas ELF ≥9.3 with FibroMeterV2G ≥0.434 provided 73% positive predictive value (Table S5). Both tests were negative for only one of the 32 cirrhotic patients, and both tests were positive for 14 of the 136 F0/1 patients. ELF and FibroMeterV2G were in disagreement in one quarter of the cases, nearly half of which were advanced fibrosis.

Intervals of Reliable Diagnosis. Beyond a single diagnostic cut-off, the test profile shows how a non-invasive test globally behaves for the diagnosis of advanced fibrosis. Figure S2 shows that the ELF and FibroMeterV2G test profiles were very similar. As expected, there was a balance between sensitivity/negative predictive value and specificity/positive predictive value. Therefore, we calculated two thresholds to rule-out and rule-in advanced fibrosis. At fixed 90% sensitivity, the negative predictive value was 86% with ELF and 88% with FibroMeterV2G (Table 5). At fixed 90% specificity, the positive predictive value was 75% and 77%, respectively. Forty-five percentage of the patients were included in the grey zone between the two thresholds with ELF versus 39% with FibroMeterV2G (P = .065, Figure 2). Only two of the 32 cirrhotic patients had a result below the 90% sensitivity threshold with ELF, and three with FibroMeterV2G (Table S6). On the other hand, 10 of the 136 F0/1 patients were misclassified as advanced fibrosis with both tests.

Figure 2.

Advanced fibrosis F3/4 as a function of intervals defined by the 90% sensitivity and 90% specificity thresholds of Enhanced Liver Fibrosis and FibroMeterV2G

Diagnostic Algorithms. We first evaluated an algorithm based on the agreement between FibroMeterV2G and ELF, with a liver biopsy performed in case of a discrepancy between the two tests (Table S7). This agreement-based algorithm provided a good performance for the diagnosis of advanced fibrosis with 86% diagnostic accuracy, 85% sensitivity, 86% specificity, 90% negative predictive value and 80% positive predictive value. Because the agreement-based algorithm has the limitation to require both FibroMeterV2G and ELF, we further evaluated algorithms based on single blood fibrosis tests used with their 90% sensitivity and 90% specificity thresholds, followed by liver biopsy in case of results in the grey zone (Table S7). Whether for ELF or FibroMeterV2G, such algorithms provided excellent performance for the diagnosis of advanced fibrosis with 90% diagnostic accuracy, 90% sensitivity, 90% specificity, 93% negative predictive value and 85% positive predictive value. However, the rate of liver biopsy was higher (39.3% with FibroMeterV2G and 45.3% with ELF) compared to the agreement-based algorithm (26.1%, P < .001).

Comments

3090D553-9492-4563-8681-AD288FA52ACE

processing....