IDDF2019-ABS-0131 Algorithms using noninvasive tests can accurately identify patients with advanced fibrosis due to NASH: data from the STELLAR clinical trials

Background There is a major unmet need for accurate, readily available, noninvasive tests (NITs) to identify patients with advanced fibrosis due to NASH. Our goal was to evaluate sequential NIT algorithms to minimize the requirement for biopsy and improve accuracy overuse of single tests. Methods The STELLAR studies (NCT03053050, NCT03053063) enrolled NASH patients with bridging fibrosis (F3) or compensated cirrhosis (F4). Baseline liver biopsies were read using the NASH CRN fibrosis classification and noninvasive fibrosis markers: FIB-4 index, ELF test, and FibroScan® (FS). The performance of these tests to discriminate advanced fibrosis was evaluated using AUROCs with 5-fold cross-validation repeated 100x. Thresholds were obtained by maximizing specificity given ≥85% sensitivity. The cohort was divided (80%/20%) into evaluation/validation sets. The evaluation set was further stratified 250x into training and test sets (66%/33%). Optimal thresholds were derived as the average across training sets, and applied sequentially (FIB-4 followed by ELF and/or FS) to the validation set. Data are from an interim analysis on 26 July 2018. Results All patients with available liver histology (N=3202, 71% F3-F4) and NIT results were included. While single tests were able to discriminate advanced fibrosis (AUROCs of 0.78, 0.80, and 0.80 for FIB-4, ELF, and FS in validation cohort), up to 32% of patients had an indeterminate result. Using thresholds derived from STELLAR data, FIB-4 followed by FS or ELF in those with indeterminate FIB-4 values (1.23 to 2.1) reduced indeterminate results to as low as 13% (table 1). Published NIT thresholds yielded similar results (data not shown). Adding a third test (FIB-4 then ELF then FS) reduced the rate of indeterminate results to 8%. Misclassification occurred at rates similar to biopsy (15–21%). The majority of misclassifications (63–81%) were false negatives; among false positive cases (19–27% of misclassifications) up to 70% had F2 fibrosis. Conclusions FIB-4 followed by ELF and/or FS nearly eliminated the need for liver biopsy and accurately identified patients with advanced fibrosis due to NASH with misclassification rates similar to liver biopsy.