Skip to content

Non-invasive assessment

TL;DR — Fibrosis stage is what matters (natural history), and it can now be estimated well enough without a biopsy that biopsy is no longer the default. In individual-patient-data meta-analysis of 5,735 patients from 37 studies, AUROCs for advanced fibrosis were 0.85 for liver stiffness by VCTE, 0.76 for FIB-4 and 0.73 for the NAFLD fibrosis score (Mózes 2022, PMID 34001645). The workhorse design is sequential: FIB-4 <1.3 rules out, FIB-4 >2.67 or ≥3.48 triggers a second test, elastography resolves the middle — a strategy that leaves 19–33% needing biopsy depending on cut-offs (PMID 34001645), and that all major guidelines now recommend. The main failure modes are well characterised and specific: FIB-4 loses discrimination in people under 35 and over 65 (AUC 0.51 in the young in one general-population comparison, van Kleef 2024, PMID 38513745), the intermediate zone is large, positive predictive value collapses when prevalence is low, and diagnostic accuracy varies by region — FIB-4 pooled AUC 0.80 globally but 0.75 in Latin America and 0.84 in MENA across 17,792 patients from 41 countries (Younossi 2026, PMID 41100867). The best-performing composites in that global sample were Agile 3+ (AUC 0.87 for advanced fibrosis) and Agile 4 (0.90 for cirrhosis).

What each test measures

Test Inputs Target Notes
FIB-4 age, AST, ALT, platelets advanced fibrosis (rule-out) free; poor at age extremes
NAFLD fibrosis score (NFS) age, BMI, hyperglycaemia, platelets, albumin, AST/ALT advanced fibrosis superseded in most pathways
ELF hyaluronic acid, PIIINP, TIMP-1 fibrosis severity proprietary; fewer false positives than FIB-4/NFS
PRO-C3 / ADAPT PRO-C3, age, diabetes, platelets advanced fibrosis collagen-formation neo-epitope
VCTE (FibroScan) liver stiffness shear-wave velocity fibrosis probe choice does not bias LSM; obesity/ascites limit
CAP ultrasound attenuation steatosis quantifies fat, not fibrosis
MRE MR shear-wave fibrosis highest accuracy; cost and access limit
MRI-PDFF proton density fat fraction steatosis (quantitative) the trial endpoint for fat
FAST LSM + CAP + AST "at-risk" MASH (MASH + NAS≥4 + F≥2) trial-enrichment tool
Agile 3+ / Agile 4 LSM, AST/ALT, platelets, sex, diabetes (+age for 3+) advanced fibrosis / cirrhosis designed to shrink the indeterminate zone
SAFE age, BMI, diabetes, platelets, AST, ALT, globulins ≥F2 (rule-out) built for primary care
MAF-5 waist, BMI, diabetes, AST, platelets LSM ≥8 kPa age-independent by design
LiverPRO age + 3–9 routine blood analytes clinically significant fibrosis CE-marked IVDR class b, 2024
NIS4 proprietary panel at-risk NASH NIMBLE-evaluated

Head-to-head accuracy

Elastography and imaging (82 studies, 14,609 patients; summary AUCs against histology) (Selvaraj 2021, PMID 33991635):

Modality Significant fibrosis Advanced fibrosis Cirrhosis
VCTE 0.83 0.85 0.89
MRE 0.91 0.92 0.90
Point SWE 0.86 0.89 0.90
2D-SWE 0.75 0.72 0.88

MRE additionally gave AUC 0.83 for NASH itself. The authors' own caveat is important and rarely quoted: only 3% of studies reported intention-to-diagnose analysis and only 18% validated pre-specified thresholds, so real-world impact is not established by these numbers.

FibroScan thresholds from a prospective 7-centre UK study (450 biopsied, 404 with valid examinations) (Eddowes 2019, PMID 30689971):

Target AUROC (95% CI) Youden cut-off
Steatosis ≥S1 0.87 (0.82–0.92) CAP 302 dB/m
Steatosis ≥S2 0.77 (0.71–0.82) CAP 331 dB/m
Steatosis S3 0.70 (0.64–0.75) CAP 337 dB/m
Fibrosis ≥F2 0.77 (0.72–0.82) LSM 8.2 kPa
Fibrosis ≥F3 0.80 (0.75–0.84) LSM 9.7 kPa
Cirrhosis F4 0.89 (0.84–0.93) LSM 13.6 kPa

Multivariable analysis found fibrosis stage was the only parameter significantly affecting LSM (p<10⁻¹⁶) — neither steatosis nor probe type biased it, which contradicts a common clinical assumption. Note that CAP performs worse as steatosis becomes more severe, the opposite of the intuitive direction.

Blood panels against a hard reference. The FNIH-NIMBLE project evaluated five panels in the NASH CRN DB2 observational cohort (n=1,073), with a prespecified bar of AUROC ≥0.7 plus superiority over ALT (activity) or FIB-4 (fibrosis). NIS4 achieved AUROC 0.81 (95% CI 0.78–0.84) for at-risk NASH; ELF, PRO-C3 and FibroMeter VCTE all achieved ≥0.8 for their respective fibrosis endpoints, and ELF and FibroMeter VCTE outperformed FIB-4 across all fibrosis endpoints (Sanyal 2023, PMID 37679433). ADAPT (PRO-C3 + age + diabetes + platelets) gave AUROC 0.86 (0.79–0.91) in derivation and 0.87 (0.83–0.91) in validation for advanced fibrosis, superior to APRI, FIB-4 and NFS in most comparisons (Daniels 2019, PMID 30014517).

Sequential pathways — the operative design

Single tests are not used alone in any current guideline. The evidence for sequencing:

Strategy Sensitivity Specificity Biopsies avoided Source
FIB-4 (<1.3; ≥2.67) → VCTE (<8.0; ≥10.0 kPa) 66% (63–68) 86% (84–87) 67% (33% need biopsy) Mózes 2022, PMID 34001645
FIB-4 (<1.3; ≥3.48) → VCTE (<8.0; ≥20.0 kPa) 38% (37–39) 90% (89–91) 81% (19% need biopsy) PMID 34001645
FIB-4 → ELF in indeterminate cases false negatives 4% false positives 8% 88% correctly classified Kjaergaard 2023, PMID 37088311
FIB-4 → VCTE, type 2 diabetes/obesity clinic — — "excellent" performance for referral Caussy 2025, PMID 39887699
FIB-4 → ELF (threshold 9.8), same cohort NPV 88–89% PPV 39–46% — PMID 39887699
FIB-4 → 2D-SWE, same cohort NPV 91% PPV 58–62% — PMID 39887699

The trade-off is explicit: raising the rule-in threshold from FIB-4 2.67 to 3.48 and LSM 10 to 20 kPa nearly halves the number of biopsies but cuts sensitivity from 66% to 38%. Neither is "correct"; the choice depends on whether the pathway is trying to find everyone with advanced fibrosis or to identify a high-certainty group for treatment.

In a Danish prospective screening study of 3,378 people (general population plus at-risk-of-ALD and at-risk-of-NAFLD groups), 3.4% of the general population screened positive at TE ≥8 kPa versus 12% and 14% of the at-risk groups. ELF alone generated far fewer false positives (11%) than FIB-4 (35%) or NFS (45%) while keeping false negatives <8%; among 155 biopsied screen-positives, 35% had ≥F3, and ELF diagnosed advanced fibrosis better than FIB-4 or NFS (AUROC 0.85, 95% CI 0.79–0.92 vs 0.73, 0.64–0.81 and 0.66, 0.57–0.76) (Kjaergaard 2023, PMID 37088311). The correlation between ELF and transient elastography was weak (Spearman ρ = 0.207) despite the concordance in classification — the tests are measuring related but distinct things.

Guidance and care pathways

The AGA NAFLD Clinical Care Pathway sets out screening, diagnosis and treatment steps for primary care, endocrine, obesity-medicine and gastroenterology practices, explicitly aimed at identifying clinically significant fibrosis (F2–F4) (Kanwal 2021, PMID 34602251). The AGA Clinical Practice Update on non-invasive biomarkers gives eight best-practice statements, of which the operative ones are: FIB-4 <1.3 has strong negative predictive value for advanced fibrosis; two or more NITs combining serum and imaging biomarkers are preferred when FIB-4 >1.3; liver biopsy should be considered when NIT results are indeterminate, discordant, or conflict with other findings; serial NITs may be used to monitor progression or treatment response; and patients whose NITs suggest F3–F4 should be considered for surveillance (Wattacheril 2023, PMID 37542503). A sequential design is also the explicit recommendation of a J Hepatol editorial position, "screening for liver fibrosis — sequential non-invasive testing works best" (Tsochatzis 2023, PMID 37295681). See guidelines.

Where the tests fail

Age. FIB-4 embeds age in the numerator, and its performance degrades at both extremes. In 21,797 individuals with metabolic dysfunction across population and hospital cohorts, the MAF-5 score (waist, BMI, diabetes, AST, platelets) achieved AUC 0.81 versus 0.61 for FIB-4 overall, and the gap widened at the extremes: in young people NPV 99% and AUC 0.86 for MAF-5 versus NPV 94% and AUC 0.51 for FIB-4 — no better than chance — and in older adults 0.75 versus 0.55 (van Kleef 2024, PMID 38513745). Applying age-adapted FIB-4 thresholds in a diabetology screening study lowered both NPV and PPV in every algorithm tested (Caussy 2025, PMID 39887699).

The age problem was quantified against biopsy nearly a decade earlier and the finding has not been superseded. In 634 European patients stratified into five age bands (≤35 n=74, 36–45 n=96, 46–55 n=197, 56–64 n=191, ≥65 n=76), AUROCs for advanced fibrosis were 0.77–0.84 for both FIB-4 and the NAFLD fibrosis score in every band above 35 years — but specificity collapsed with age, to 35% for FIB-4 and 20% for NFS in those aged ≥65, and both scores plus the AST/ALT ratio were useless below 35 (AUROCs 0.52, 0.52, 0.60) (McPherson 2017, PMID 27725647). Re-derived and validated cut-offs for ≥65 years — FIB-4 2.0 (sensitivity 77%) and NFS 0.12 (sensitivity 80%) — restored specificity to 70%. The standard 1.30 lower cut-off therefore generates a false-positive rate approaching two in three in older patients, which is the population MASLD screening is most often applied to.

Screening in low-prevalence populations. Pooling 5,129 patients from five population-based cohorts (Spain, Hong Kong, Denmark, England, France; 3,979 general population and 1,150 at-risk through alcohol, diabetes or obesity) with concurrent FIB-4, NFS and VCTE, 552 (11%) had liver stiffness ≥8 kPa — and of those, 239 (43%) had a normal FIB-4 and 171 (31%) a normal NFS. FIB-4 was falsely negative in 11% of participants with diabetes (NFS 2.5%), and 28–29% of elevated FIB-4/NFS results were false positives in both the general population and the at-risk cohorts. Waist circumference outperformed both scores for detecting LSM ≥8 kPa in the general population (Graupera 2022, PMID 34971806). The authors' conclusion is blunt: FIB-4 and NFS are "suboptimal for screening purposes due to a high risk of overdiagnosis and a non-negligible percentage of false-negatives". This is the strongest evidence against the current first-line position of FIB-4 in every published care pathway (guidelines).

Prevalence. Eddowes showed directly that applying cut-offs derived in a biopsy-referral cohort to lower-prevalence populations raises NPV and drops PPV (PMID 30689971). This is the arithmetic of screening, not a flaw in the tests, but it means the same 10 kPa result means different things in a hepatology clinic and in a health check.

Geography. In the 17,792-patient G-MASLD sample from 41 countries (14% F0, 32% F1, 18% F2, 22% F3, 13% F4), FIB-4's AUC for advanced fibrosis ranged from 0.75 (95% CI 0.71–0.79) in Latin America to 0.84 (0.82–0.87) in MENA; ELF from 0.72 (0.69–0.76) in Europe to 0.80 (0.78–0.82) in North America. LSM was more stable across regions (pooled 0.84) except in North America (0.78). Agile 3+ (0.87) and Agile 4 (0.90 for cirrhosis, ranging 0.85 in North America to 0.96 in MENA) were the most accurate overall (Younossi 2026, PMID 41100867). There is no accepted explanation for the regional variation, and it undermines the practice of importing a single set of thresholds worldwide.

Technical failure. VCTE requires an adequate acoustic window; the Eddowes cohort lost 10% of examinations to invalid measurement. AGA advises using NITs per manufacturer specification (not in ascites or with pacemakers) to avoid discordant results (PMID 37542503).

The indeterminate zone. The problem Agile 3+ and Agile 4 were explicitly built to solve: existing NITs "are highly effective at excluding advanced fibrosis or cirrhosis but only have moderate ability to rule these in", and leave many patients unclassified. Both new scores outperformed FIB-4 and LSM on AUROC, on percentage indeterminate, and on positive predictive value for ruling in (Sanyal 2023, PMID 36375686).

Tools built for primary care

Tool Population Key performance Source
SAFE NASH CRN (n=676) derivation; FLINT (n=280) and MRE (n=130) testing; NHANES III (n=11,953) for mortality AUROC ≥0.80 for F0/1 vs ≥F2, consistently above FIB-4 and NFS; NPV at SAFE=0 was 88% and 92%; survival with SAFE <0 comparable to no steatosis (p=0.34); SAFE >100 adjusted HR 1.53 (p<0.01) Sripongpun 2023, PMID 35477908
MAF-5 21,797 with metabolic dysfunction 60.9% low / 14.1% intermediate / 24.9% high risk, with observed fibrosis prevalence 3.3% / 7.9% / 28.1%; AUC 0.86 (training) and 0.85 (validation) for LSM ≥12 kPa; MAF-5 >1 associated with all-cause mortality aHR 1.59 (1.47–1.73) van Kleef 2024, PMID 38513745
LiverPRO development n=462; DECIDE validation n=6,468; 3 further cohorts n=2,554 AUC 0.80 (0.78–0.82) for TE ≥8 kPa, vs ELF 0.78, LiverRisk 0.81, FIB-4 0.69, NFS 0.74; rule-out (<25%) sensitivity 80.6% and NPV 98.0% vs FIB-4 (<1.3) sensitivity 53.8% and NPV 95.8%; 2-year liver-event C-statistic 0.80 Lindvig 2025, PMID 39674225

The consistent finding across all three is that FIB-4's rule-out sensitivity in unselected populations is much lower than its reputation — 53.8% in DECIDE — even though its negative predictive value stays high because the disease is rare. High NPV in a low-prevalence setting is not evidence of a good rule-out test.

How much better can a blood test get?

Two lines of work bound the answer differently. Benchmarking. A superlearner ensemble trained on 23 demographic and clinical variables in the NASH CRN observational cohort (n=648) and validated in FLINT (n=270) and NHANES participants with NAFLD (n=1,244) reached AUCs of 0.79 (0.73–0.84) and 0.74 (0.68–0.79) for significant fibrosis — and the SAFE score performed similarly, with both outperforming FIB-4, APRI, Forns and BARD (Charu 2024, PMID 38687634). A 12-base-model ensemble matched a 90-model one. In these datasets, increasing model complexity did not materially improve discrimination beyond SAFE; this is a benchmark, not proof of a universal performance ceiling for routine variables.

New analytes. A translational panel derived from a diet-induced MASLD mouse model (LDLr⁻/⁻.Leiden), mapped through human liver-biopsy transcriptomes to serum protein levels, produced a three-marker panel — IGFBP7, SSc5D and Sema4D — with LightGBM AUCs of 0.82 (F0/F1), 0.89 (F2) and 0.87 (F3/F4), replicated in an independent validation cohort and reported to outperform FIB-4, APRI and FibroScan (Verschuren 2024, PMID 38811591). If both results hold, the practical implication is specific: the marginal return on recombining existing routine variables is near zero, and the field's effort belongs on mechanism-derived analytes and on imaging.

Case-finding in the highest-yield population

In 501 prospectively recruited adults aged ≥50 with type 2 diabetes, systematically assessed with MRI-PDFF, MRE, VCTE and CAP, prevalence was NAFLD 65%, advanced fibrosis 14%, cirrhosis 6%; among 29 with cirrhosis, two had HCC and one had gallbladder adenocarcinoma. Obesity (OR 2.50, 95% CI 1.38–4.54, p=0.003) and insulin use (OR 2.71, 1.33–5.50, p=0.006) independently predicted advanced fibrosis (Ajmera 2023, PMID 36410554). In a diabetology screening cohort of 654 (87% T2D, 74% obesity), 17.6% had intermediate/high risk of advanced fibrosis and 9.3% high risk (Caussy 2025, PMID 39887699). See MASLD and type 2 diabetes.

Monitoring, not just diagnosis

The evidence has moved from cross-sectional accuracy to longitudinal risk. Serial VCTE-based Agile scores in 16,603 patients discriminated liver-related events with integrated time-dependent AUC 0.89, and separated persistently-low from persistently-high Agile 3+ by 0.6 versus 30.1 events per 1,000 person-years; a >20% fall from a high baseline was associated with substantially lower risk (Lin 2024, PMID 38512249). NITs matched or beat histology for 5-year outcome prediction (LSM-VCTE tAUC 0.76 vs histology 0.72) (Mózes 2023, PMID 37290471). AGA best-practice advice 7 endorses serial monitoring (PMID 37542503).

Open questions

  • Should FIB-4 remain the first step at all? In 5,129 people across five population cohorts, 43% of those with LSM ≥8 kPa had a normal FIB-4, 28–29% of elevated results were false positives, and waist circumference outperformed it in the general population (PMID 34971806); specificity falls to 35% above age 65 unless the threshold is raised to 2.0 (PMID 27725647). Every published care pathway nonetheless begins with FIB-4 at 1.30. No pathway using an alternative first step has been prospectively compared for outcomes or cost.
  • Have routine-variable scores hit their ceiling? A superlearner ensemble over 23 clinical variables reached AUC 0.74–0.79 and did no better than SAFE (PMID 38687634), which implies the remaining error is in the inputs rather than the models. Whether mechanism-derived analytes such as IGFBP7/SSc5D/Sema4D (PMID 38811591) clear that ceiling in an independent population with biopsy reference has not been tested.
  • Why does non-invasive test accuracy vary by region? FIB-4 AUC differs by 0.09 and Agile 4 by 0.11 between world regions in a single harmonised cohort (PMID 41100867). Candidate explanations — biopsy practice, fibrosis-stage distribution, platelet reference ranges, genetic background, assay standardisation — have not been separated, and no region-specific thresholds have been derived and validated.
  • What replaces FIB-4 under 35 and over 65? MAF-5 (age-independent by design) and LiverPRO both outperform FIB-4 at the extremes (PMIDs: 38513745, 39674225), but neither has been prospectively compared head-to-head in a screening programme with biopsy or outcome adjudication.
  • Should genotype enter the pathway? PNPLA3 genotype predicts progression (see genetics) and has become a drug target, yet no NIT in clinical use contains genetic information, and no prospective study has tested whether adding it improves discrimination or reclassification.
  • Does non-invasive-test-guided care change outcomes? A prospective same-month FIB-4/ELF/MRE/biopsy study of 186 people quantified the diagnostic loss from sequencing: FIB-4 sensitivity was 57%, excluding 43% of fibrotic MASLD from ELF testing; concurrent FIB-4-or-ELF reached 87% sensitivity and 64% specificity (Allen 2026, PMID 42566274). This remains diagnostic pathway evidence. The 2026-09-02 search found no randomised pathway implementation trial with clinical outcomes; cost-effectiveness modelling exists (PMID 41196592) but is not outcome evidence.
  • Is at-risk MASH the right target? FAST and NIS4 are built to identify MASH with NAS≥4 and F≥2 (PMIDs: 32027858, 37679433) because that is the trial-eligibility definition, not because it is the prognostic threshold — which is fibrosis stage. The two targets have been allowed to merge.

References

  1. Mózes FE, Lee JA, Selvaraj EA, et al. Diagnostic accuracy of non-invasive tests for advanced fibrosis in patients with NAFLD: an individual patient data meta-analysis. Gut. 2022;71(5):1006-1019. PMID 34001645
  2. Selvaraj EA, Mózes FE, Jayaswal ANA, et al. Diagnostic accuracy of elastography and magnetic resonance imaging in patients with NAFLD: A systematic review and meta-analysis. J Hepatol. 2021;75(4):770-785. PMID 33991635
  3. Eddowes PJ, Sasso M, Allison M, et al. Accuracy of FibroScan Controlled Attenuation Parameter and Liver Stiffness Measurement in Assessing Steatosis and Fibrosis in Patients With Nonalcoholic Fatty Liver Disease. Gastroenterology. 2019;156(6):1717-1730. PMID 30689971
  4. Newsome PN, Sasso M, Deeks JJ, et al. FibroScan-AST (FAST) score for the non-invasive identification of patients with non-alcoholic steatohepatitis with significant activity and fibrosis: a prospective derivation and global validation study. Lancet Gastroenterol Hepatol. 2020;5(4):362-373. PMID 32027858
  5. Sanyal AJ, Foucquier J, Younossi ZM, et al. Enhanced diagnosis of advanced fibrosis and cirrhosis in individuals with NAFLD using FibroScan-based Agile scores. J Hepatol. 2023;78(2):247-259. PMID 36375686
  6. Daniels SJ, Leeming DJ, Eslam M, et al. ADAPT: An Algorithm Incorporating PRO-C3 Accurately Identifies Patients With NAFLD and Advanced Fibrosis. Hepatology. 2019;69(3):1075-1086. PMID 30014517
  7. Sanyal AJ, Shankar SS, Yates KP, et al. Diagnostic performance of circulating biomarkers for non-alcoholic steatohepatitis. Nat Med. 2023;29(10):2656-2664. PMID 37679433
  8. Kjaergaard M, Lindvig KP, Thorhauge KH, et al. Using the ELF test, FIB-4 and NAFLD fibrosis score to screen the population for liver disease. J Hepatol. 2023;79(2):277-286. PMID 37088311
  9. Sripongpun P, Kim WR, Mannalithara A, et al. The steatosis-associated fibrosis estimator (SAFE) score: A tool to detect low-risk NAFLD in primary care. Hepatology. 2023;77(1):256-267. PMID 35477908
  10. van Kleef LA, Francque SM, Prieto-Ortiz JE, et al. Metabolic Dysfunction-Associated Fibrosis 5 (MAF-5) Score Predicts Liver Fibrosis Risk and Outcome in the General Population With Metabolic Dysfunction. Gastroenterology. 2024;167(2):357-367.e9. PMID 38513745
  11. Lindvig KP, Thorhauge KH, Hansen JK, et al. Development, validation, and prognostic evaluation of LiverPRO for the prediction of significant liver fibrosis in primary care: a prospective cohort study. Lancet Gastroenterol Hepatol. 2025;10(1):55-67. PMID 39674225
  12. Caussy C, Vergès B, Leleu D, et al. Screening for Metabolic Dysfunction-Associated Steatotic Liver Disease-Related Advanced Fibrosis in Diabetology: A Prospective Multicenter Study. Diabetes Care. 2025;48(6):877-886. PMID 39887699
  13. Ajmera V, Cepin S, Tesfai K, et al. A prospective study on the prevalence of NAFLD, advanced fibrosis, cirrhosis and hepatocellular carcinoma in people with type 2 diabetes. J Hepatol. 2023;78(3):471-478. PMID 36410554
  14. Younossi ZM, de Avila L, Petta S, et al. Global performance of non-invasive tests in MASLD: Insights from the G-MASLD study. Hepatology. 2026;84(1):161-174. PMID 41100867
  15. Kanwal F, Shubrook JH, Adams LA, et al. Clinical Care Pathway for the Risk Stratification and Management of Patients With Nonalcoholic Fatty Liver Disease. Gastroenterology. 2021;161(5):1657-1669. PMID 34602251
  16. Wattacheril JJ, Abdelmalek MF, Lim JK, et al. AGA Clinical Practice Update on the Role of Noninvasive Biomarkers in the Evaluation and Management of Nonalcoholic Fatty Liver Disease: Expert Review. Gastroenterology. 2023;165(4):1080-1088. PMID 37542503
  17. Tsochatzis EA. Screening for liver fibrosis - sequential non-invasive testing works best. J Hepatol. 2023;79(2):263-265. PMID 37295681
  18. Mózes FE, Lee JA, Vali Y, et al. Performance of non-invasive tests and histology for the prediction of clinical outcomes in patients with non-alcoholic fatty liver disease: an individual participant data meta-analysis. Lancet Gastroenterol Hepatol. 2023;8(8):704-713. PMID 37290471
  19. Lin H, Lee HW, Yip TC, et al. Vibration-Controlled Transient Elastography Scores to Predict Liver-Related Events in Steatotic Liver Disease. JAMA. 2024;331(15):1287-1297. PMID 38512249
  20. Lazarus JV, Ivancovsky Wajcman D, Mark HE, et al. Cost-Effectiveness of MASH Diagnosis and Management Approaches Among Those With Type 2 Diabetes. JAMA Netw Open. 2025;8(11):e2542750. PMID 41196592
  21. Allen AM, et al. Sequential FIB-4-based testing limits detection of fibrotic MASLD despite optimized ELF thresholds. Hepatol Commun. 2026;10(9):e1012. PMID 42566274
  22. McPherson S, Hardy T, Dufour JF, et al. Age as a Confounding Factor for the Accurate Non-Invasive Diagnosis of Advanced NAFLD Fibrosis. Am J Gastroenterol. 2017;112(5):740-751. PMID 27725647
  23. Graupera I, Thiele M, Serra-Burriel M, et al. Low Accuracy of FIB-4 and NAFLD Fibrosis Scores for Screening for Liver Fibrosis in the Population. Clin Gastroenterol Hepatol. 2022;20(11):2567-2576.e6. PMID 34971806
  24. Charu V, Liang JW, Mannalithara A, et al. Benchmarking clinical risk prediction algorithms with ensemble machine learning for the noninvasive diagnosis of liver fibrosis in NAFLD. Hepatology. 2024;80(5):1184-1195. PMID 38687634
  25. Verschuren L, Mak AL, van Koppen A, et al. Development of a novel non-invasive biomarker panel for hepatic fibrosis in MASLD. Nat Commun. 2024;15(1):4564. PMID 38811591