Vibration-Controlled Transient Elastography Scores to Predict Liver-Related Events in Steatotic Liver Disease

Key Points

Question

What are the clinical implications of single or serial measurements of vibration–controlled transient elastography (VCTE)–based scores in metabolic dysfunction-associated steatotic liver disease (MASLD)?

Findings

In this cohort study of 16 603 patients with MASLD who underwent VCTE examination, the VCTE-based scores outperformed most noninvasive tests and were at least similar if not better than histologic fibrosis staging in predicting liver-related events. On repeated testing, the scores were largely stable, and patients with improvement in the scores had substantial reduction in the risk of liver-related events.

Meaning

Findings of this study suggest that the VCTE-based scores are generally accurate for predicting liver-related events, making them suitable alternatives to liver biopsy in routine clinical practice and in phase 2b and 3 clinical trials for steatohepatitis.

Abstract

Importance

Metabolic dysfunction–associated steatotic liver disease (MASLD) is currently the most common chronic liver disease worldwide. It is important to develop noninvasive tests to assess the disease severity and prognosis.

Objective

To study the prognostic implications of baseline levels and dynamic changes of the vibration-controlled transient elastography (VCTE)–based scores developed for the diagnosis of advanced fibrosis (Agile 3+) and cirrhosis (Agile 4) in patients with MASLD.

Design, Setting, and Participants

This cohort study included data from a natural history cohort of patients with MASLD who underwent VCTE examination at 16 tertiary referral centers in the US, Europe, and Asia from February 2004 to January 2023, of which the data were collected prospectively at 14 centers. Eligible patients were adults aged at least 18 years with hepatic steatosis diagnosed by histologic methods (steatosis in ≥5% of hepatocytes) or imaging studies (ultrasonography, computed tomography or magnetic resonance imaging, or controlled attenuation parameter ≥248 dB/m by VCTE).

Main Outcomes and Measures

The primary outcome was liver-related events (LREs), defined as hepatocellular carcinoma or hepatic decompensation (ascites, variceal hemorrhage, hepatic encephalopathy, or hepatorenal syndrome), liver transplant, and liver-related deaths. The Agile scores were compared with histologic and 8 other noninvasive tests.

Results

A total of 16 603 patients underwent VCTE examination at baseline (mean [SD] age, 52.5 [13.7] years; 9600 [57.8%] were male). At a median follow-up of 51.7 (IQR, 25.2-85.2) months, 316 patients (1.9%) developed LREs. Both Agile 3+ and Agile 4 scores classified fewer patients between the low and high cutoffs than most fibrosis scores and achieved the highest discriminatory power in predicting LREs (integrated area under the time-dependent receiver-operating characteristic curve, 0.89). A total of 10 920 patients (65.8%) had repeated VCTE examination at a median interval of 15 (IQR, 11.3-27.7) months and were included in the serial analysis. A total of 81.9% of patients (7208 of 8810) had stable Agile 3+ scores and 92.6% of patients (8163 of 8810) had stable Agile 4 scores (same risk categories at both assessments). The incidence of LREs was 0.6 per 1000 person-years in patients with persistently low Agile 3+ scores and 30.1 per 1000 person-years in patients with persistently high Agile 3+ scores. In patients with high Agile 3+ score at baseline, a decrease in the score by more than 20% was associated with substantial reduction in the risk of LREs. A similar trend was observed for the Agile 4 score, although it missed more LREs in the low-risk group.

Conclusions and Relevance

Findings of this study suggest that single or serial Agile scores are highly accurate in predicting LREs in patients with MASLD, making them suitable alternatives to liver biopsy in routine clinical practice and in phase 2b and 3 clinical trials for steatohepatitis.

This cohort study evaluates the prognostic implications of baseline and repeated vibration-controlled transient elastography scores and liver stiffness measurements in a large cohort of patients with metabolic dysfunction–associated steatotic liver disease.

Introduction

Metabolic dysfunction–associated steatotic liver disease (MASLD), previously known as nonalcoholic fatty liver disease (NAFLD), is currently the most common chronic liver disease that affects approximately 30% of the worldwide adult population.¹ It has become one of the leading causes of cirrhosis and hepatocellular carcinoma (HCC) in middle-income and high-income countries,² with estimated annual direct medical costs of approximately US $103 billion in the US and €35 billion in Europe.³

In patients with MASLD, there is a dose-response relationship between the severity of liver fibrosis and future risk of liver-related events (LREs).⁴ In the past 2 decades, a number of noninvasive tests of fibrosis have been adopted for clinical use.⁵ In particular, liver stiffness measurement (LSM) by vibration-controlled transient elastography (VCTE) not only reflects the degree of liver fibrosis but also predicts HCC, portal hypertension, and varices.⁶ Recently, by combining LSM and simple clinical parameters (platelet count, aminotransferases, diabetes, age, and sex), several authors of the present study derived and validated the Agile 3+ score for the diagnosis of advanced fibrosis and the Agile 4 score for the diagnosis of cirrhosis (collectively, Agile scores) in patients with MASLD with improved accuracy and reduced indeterminate zone compared with LSM alone.⁷ Emerging data suggest that the Agile scores are also prognostic.⁸ However, previous studies were limited by small sample sizes. In addition, the prognostic meaning of a change in noninvasive tests over time is unclear, especially as the tests are imperfect and may have false-positive and false-negative results.

With this background, we aimed to evaluate the prognostic implications of baseline and repeated Agile score and LSMs in a large cohort of patients with MASLD. We also compared the prognostic performance of the Agile score with that of other various noninvasive tests of hepatic fibrosis.

Methods

Study Design and Participants

This was a cohort study of patients with MASLD who had undergone VCTE examination at 16 centers from the US, Europe, and Asia, of which data were collected prospectively at 14 centers. Eligible patients were adults aged 18 years or older with hepatic steatosis diagnosed by histologic methods (steatosis in ≥5% of hepatocytes) or imaging studies (ultrasonography, computed tomography or magnetic resonance imaging, or controlled attenuation parameter ≥248 dB/m by VCTE). Patients were excluded if they had other liver diseases such as chronic viral hepatitis, HIV infection, excessive alcohol consumption (>30 g/d in men and >20 g/d in women), secondary causes of hepatic steatosis (eg, use of systemic steroids), or a history of HCC, hepatic decompensation, liver resection, liver transplant, or other malignant carcinomas.

The study protocol was approved by the institutional review boards of the participating sites and followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline for cohort studies as well as the relevant portions of the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD) reporting guideline. The study was conducted in accordance with the principles of the Declaration of Helsinki.⁹ The need for informed written consent was waived because of the retrospective nature of this study.

Assessments

At each clinic visit, the medical history of a patient was recorded. Body mass index (BMI) was calculated as weight in kilograms divided by height in meters squared. A venous blood sample was taken after at least 8 hours of fasting for kidney and liver biochemistry and complete blood cell count. Controlled attenuation parameter and liver stiffness were assessed using the VCTE machine (FibroScan; Echosens) by trained operators as previously described, and patients needed to have at least 10 valid acquisitions (eMethods in Supplement 1).¹⁰

From these assessments, we calculated the VCTE-based scores including the Agile 3+, Agile 4, and FibroScan-aspartate aminotransferase (FAST) scores (developed to diagnose at-risk metabolic dysfunction–associated steatohepatitis) (eMethods in Supplement 1).^7,11 For comparison, we also calculated simple fibrosis scores including the Fibrosis-4 index (FIB-4), NAFLD fibrosis score (NFS), aspartate aminotransferase (AST) to platelet ratio index (APRI), BARD (BMI, AST:ALT ratio, and diabetes) score, and AST to alanine aminotransferase ratio. All calculations and cutoffs were based on the existing literature.¹² Only parameters measured within 1 month of each other were used to calculate the scores. Otherwise, the particular noninvasive test was treated as missing.

Outcomes

The primary outcome was a composite end point of LREs including HCC, hepatic decompensation (ascites, variceal hemorrhage, hepatic encephalopathy or hepatorenal syndrome), liver transplant, and liver-related death. Secondary outcomes included HCC and hepatic decompensation, analyzed separately. The diagnosis of the events was based on prospective follow-up, medical record review, or validated registries with positive predictive values of at least 90%.

Statistical Analysis

In the baseline model, the baseline date was defined as the date of the first noninvasive test. For the Agile scores and FAST scores that included both VCTE and blood tests, the latter date was taken as baseline to avoid immortal time bias. Pairwise comparisons between the Agile scores and the other tests were performed by comparing the area under the receiver-operating characteristic curves (AUROC) using z test for patients in whom the results of both tests were available.¹³ We also calculated the integrated AUROC, area under the time-dependent precision-recall curves (AUPRC),¹⁴ and integrated Brier score over time. The Agile scores and other tests were evaluated for continuous net reclassification improvement (NRI) with reference to LSM using the inverse probability weighting estimator.^15,16 All fibrosis scores classified patients into low-risk, intermediate-risk, and high-risk groups on the published low and high cutoffs. For histology, we stratified the 3 groups as F0 to F2, F3, and F4 according to the Nonalcoholic Steatohepatitis Clinical Research Network scoring system. The cumulative incidence of outcomes with adjustment of competing events was estimated by the Gray method and compared by the Fine and Gray test among different risk categories (eMethods in Supplement 1). For both the primary outcome and HCC, non–liver-related death was treated as a competing event. For hepatic decompensation, both non–liver-related death and HCC were treated as competing events.

In the serial model, we considered patients with 2 or more VCTE examinations. For those with multiple examinations, we selected the first and last examinations, with a maximum 5-year interval and a minimum 6-month separation. We assessed the incidence of the outcomes from the last VCTE examinations onward. Patients developing LREs between these examinations were documented but not included in the serial prediction models. Transition among risk categories based on published cutoffs was depicted using Sankey diagrams. We also evaluated the prognostic utility of serial noninvasive tests based on their relative change between the 2 examinations (eMethods in Supplement 1). The association between changes in noninvasive tests and the risk of LREs was examined using restricted cubic spline analysis.

All statistical analyses were performed using R software version 4.2.2 (R Foundation for Statistical Computing). Statistical significance was defined as 2-sided P < .05.

Results

Participants

From February 2004 to January 2023, we identified 17 949 patients with 1 or more VCTE examinations. After excluding 1346 patients according to the inclusion and exclusion criteria, 16 603 patients were included in the baseline model (Figure 1). Their mean (SD) age was 52.5 (13.7) years; 9600 were men (57.8%); and 7003 were women (42.2%) (Table 1). A total of 5761 patients (34.7%) had diabetes and 5769 (34.8%) hypertension. A total of 3030 patients (18.2%) were from the US or Europe (Western countries [listed in Figure 1]), and 13 573 patients (81.8%) were from Asia (Asian countries [listed in Figure 1]). Among 3532 patients with liver biopsy, 1182 (33.5%) had F3 or F4 fibrosis. The median interval between liver biopsy and VCTE examinations was 28 (IQR, 0-214) days.

Figure 1. Study Participant Flow.

HCC indicates hepatocellular carcinoma; MASLD, metabolic dysfunction–associated steatotic liver disease. VCTE, vibration-controlled transient elastography.

Table 1. Clinical Characteristics of the Cohorts in the Baseline and Serial Models.

Characteristic	Baseline model (N = 16 603)	Serial model (n = 10 920)
		First test	Last test	P valuea
Age, y	52.5 (13.7)	52.3 (13.5)	54.4 (13.6)	<.001
Sex, No. (%)
Female	7003 (42.2)	4629 (42.4)	4629 (42.4)	NA
Male	9600 (57.8)	6291 (57.6)	6291 (57.6)	NA
BMI	27.0 (24.5 to 30.0)	27.0 (24.5 to 30.0)	27.0 (24.6 to 30.1)	.09
Diabetes, No. (%)	5761 (34.7)	3944 (36.1)	5311 (48.6)	<.001
Hypertension, No. (%)	5769 (34.8)	3925 (35.9)	5291 (48.5)	<.001
ALT, IU/L	37 (23 to 62)	36 (23 to 61)	30 (20 to 48)	<.001
AST, IU/L	31 (23 to 47)	31 (22 to 46)	27 (21 to 38)	<.001
GGT, IU/L	44 (27 to 79)	43 (26 to 76)	36 (23 to 63)	<.001
Albumin, g/L	44.4 (3.9)	44.7 (3.5)	44.8 (3.6)	.02
Total bilirubin, μmol/L	12.0 (8.6 to 15.4)	12.0 (8.6 to 15.4)	12.0 (10.0 to 17.1)	<.001
Platelet, ×109/L	237 (198 to 280)	238 (199 to 281)	235 (196 to 279)	<.001
Creatinine, μmol/L	72 (60 to 83)	72 (60 to 83)	72 (61 to 84)	<.001
FibroScan
LSM,b kPa	6.0 (4.7 to 8.5)	6.0 (4.6 to 8.3)	5.5 (4.5 to 7.7)	<.001
Controlled attenuation parameter, dB/m	303 (273 to 334)	302 (273 to 334)	295 (262 to 328)	<.001
Noninvasive testc
Agile 3+	0.16 (0.06 to 0.44)	0.17 (0.06 to 0.43)	0.21 (0.08 to 0.48)	<.001
Agile 4	0.01 (0.00 to 0.06)	0.01 (0.00 to 0.05)	0.01 (0.00 to 0.05)	.23
FAST	0.28 (0.12 to 0.52)	0.27 (0.12 to 0.51)	0.19 (0.09 to 0.41)	<.001
FIB-4	1.11 (0.74 to 1.71)	1.13 (0.76 to 1.71)	1.18 (0.81 to 1.75)	<.001
NFS	−1.99 (−3.03 to −0.78)	−1.98 (−3.00 to −0.83)	−1.62 (−2.67 to –0.49)	<.001
APRI	0.33 (0.23 to 0.54)	0.33 (0.23 to 0.52)	0.30 (0.22 to 0.45)	<.001
AST:ALT	0.83 (0.62 to 1.12)	0.84 (0.64 to 1.14)	0.90 (0.69 to 1.20)	<.001
BARD	2 (1 to 3)	2 (1 to 3)	2 (1 to 3)	<.001
Fibrosis stage (n = 3532)d
F0	576 (16.3)	NA	NA	NA
F1	1189 (33.7)	NA	NA	NA
F2	585 (16.6)	NA	NA	NA
F3	744 (21.1)	NA	NA	NA
F4	438 (12.4)	NA	NA	NA
Follow-up duration, median (IQR), mo	51.7 (25.2 to 85.2)	NA	34.0 (12.4 to 55.9)	NA

Abbreviations: ALT, alanine aminotransferase; APRI, AST to platelets ratio index; AST, aspartate aminotransferase; BARD, BMI, AST:ALT ratio, and diabetes; BMI, body mass index (calculated as weight in kilograms divided by height in meters squared); FAST, FibroScan-AST; FIB-4, Fibrosis-4 index; GGT, gamma-glutamyl transpeptidase; LSM, liver stiffness measurement; NA, not applicable; NAFLD, nonalcoholic fatty liver disease; NFS, NAFLD fibrosis score.

SI and conventional conversion factors: To convert ALT, AST, and GGT to microkatals per liter, multiply by 0.0167; serum creatinine from micromoles per liter to milligrams per deciliter, divide by 88.4; albumin from grams per liter to grams per deciliter, divide by 10; total bilirubin from micromoles per liter to milligrams per deciliter, divide by 17.104; platelets from ×10⁹/L to ×10³/μL, divide by 1.

^a Paired samples tests between the first and last tests in the serial model.

^b The LSM is a noninvasive method to evaluate liver fibrosis, using transient elastography to measure liver stiffness, which helps in assessing the extent of fibrosis; controlled attenuation parameter quantifies liver steatosis noninvasively, by measuring the attenuation of ultrasonography waves through the liver, providing an indicator of fat levels.

^c The formulas for the calculation of the noninvasive tests are presented in the eMethods of Supplement 1.

^d Fibrosis stage (0-4) according to the Nonalcoholic Steatohepatitis Clinical Research Network scoring system. F0, no fibrosis; F1, centrilobular pericellular fibrosis; F2: centrilobular and periportal fibrosis; F3: bridging fibrosis; and F4, cirrhosis.

Baseline Model

At a median follow-up of 51.7 (IQR, 25.2-85.2) months, 316 patients (1.9%) developed LREs, including 139 cases of HCC and 209 cases of hepatic decompensation (eTable 1 in Supplement 1). Both the Agile 3+ and Agile 4 scores demonstrated the highest AUROC and AUPRC for predicting LREs (Figure 2A; eFigure 1 in Supplement 1); they classified fewer patients (10.2% [1314 of 12948] for Agile 3+ and 8.7% [1130 of 12948] for Agile 4) in the intermediate-risk group than the other fibrosis scores. The Agile 3+ and Agile 4 scores also demonstrated the highest integrated AUROC and lowest integrated Brier score (eTable 2 in Supplement 1). Similarly, in the 10 678 patients with all studied fibrosis markers available, the Agile 3+ and Agile 4 scores demonstrated highest AUROC and lowest integrated Brier score (eFigures 1 and 2 and eTable 2 in Supplement 1).

Figure 2. Prediction of Liver-Related Events (LREs) by Noninvasive Tests and Liver Histologic Findings.

In panel B, the cut points for Agile 3+ score were based on the original publication. The low cut point (<0.451) achieved sensitivity of at least 85% to rule out patients of fibrosis stage 3 or greater; the high cut point (0.679) achieved specificity of at least 90% to rule in patients of fibrosis stage of 3 or greater. ALT indicates alanine aminotransferase; AST, aspartate aminotransferase; APRI, AST to platelets ratio index; BARD, BMI (body mass index), AST:ALT ratio, and diabetes; FAST, FibroScan-AST; FIB-4, Fibrosis-4 index; NAFLD, nonalcoholic fatty liver disease; NFS, NAFLD fibrosis score.

By pairwise comparison, the AUROC for LREs of both Agile scores was significantly higher than histologic fibrosis staging and other comparator fibrosis tests at 3 and 5 years, with the exception of a similar performance between the Agile scores and LSM at 3 years. The AUROC ranged from 0.87 to 0.91 for Agile 3+ and Agile 4 scores compared with a range from 0.86 to 0.88 for histologic fibrosis stage, 0.86 for LSM, and from 0.66 to 0.86 for the fibrosis scores (eTable 3 in Supplement 1). The calibration was excellent for both Agile scores but was generally unsatisfactory for the simple fibrosis scores (eFigures 3 and 4 in Supplement 1). The Agile scores better reclassified patients with and without LREs at 3 and 5 years according to their risk compared with LSM, while other noninvasive tests generally had a similar or reduced correct reclassification compared with LSM (Table 2). Analyzed separately, all the fibrosis tests were better at the prediction of hepatic decompensation than HCC (eFigure 5, eTables 4 and 5 in Supplement 1).

Table 2. Paired Comparisons of the Agile Scores and Other Noninvasive Tests vs LSM on the NRI for the Prediction of 3-Year and 5-Year LREs in the Baseline Model.

Test	No.	NRI (95% CI)a
		3-y LRE			5-y LRE
		Event	Nonevent	Overall	Event	Nonevent	Overall
Agile 3+	12 948	0.31 (0.14 to 0.49)	0.57 (0.53 to 0.61)	0.88 (0.68 to 1.08)	0.41 (0.27 to 0.54)	0.61 (0.57 to 0.65)	1.02 (0.86 to 1.18)
Agile 4	12 948	0.19 (0.02 to 0.36)	0.81 (0.79 to 0.83)	1.00 (0.82 to 1.18)	0.30 (0.14 to 0.43)	0.84 (0.82 to 0.85)	1.13 (0.96 to 1.28)
LSM		0 [Reference]	0 [Reference]	0 [Reference]	0 [Reference]	0 [Reference]	0 [Reference]
FIB-4	12 950	−0.30 (−0.46 to −0.04)	−0.78 (−0.81 to −0.54)	−1.08 (−1.25 to −0.63)	−0.31 (−0.46 to −0.03)	−0.78 (−0.81 to −0.51)	−1.09 (−1.24 to −0.57)
NFS	12 064	−0.18 (−0.37 to 0.04)	−0.57 (−0.69 to 0.12)	−0.75 (−0.98 to 0.06)	−0.16 (−0.28 to 0.10)	−0.52 (−0.64 to 0.12)	−0.68 (−0.87 to 0.20)
APRI	12 975	−0.40 (−0.56 to −0.20)	−0.79 (−0.82 to −0.75)	−1.19 (−1.35 to −0.98)	−0.43 (−0.55 to −0.26)	−0.80 (−0.83 to −0.77)	−1.23 (−1.35 to −1.05)
FAST	11 541	−0.16 (−0.37 to 0.05)	0.24 (0.14 to 0.31)	0.08 (−0.19 to 0.34)	−0.10 (−0.26 to 0.07)	0.26 (0.17 to 0.34)	0.17 (−0.07 to 0.38)
AST:ALT	13 159	−0.37 (−0.53 to −0.22)	−0.78 (−0.80 to −0.70)	−1.15 (−1.31 to −0.96)	−0.44 (−0.55 to −0.30)	−0.78 (−0.81 to −0.69)	−1.22 (−1.34 to −1.02)
BARD	12 498	−0.36 (−0.51 to −0.18)	−0.14 (−0.32 to −0.11)	−0.50 (−0.74 to −0.31)	−0.32 (−0.46 to −0.17)	−0.12 (−0.27 to −0.09)	−0.44 (−0.68 to −0.28)

Abbreviations: ALT indicates alanine aminotransferase; AST, aspartate aminotransferase; APRI, AST to platelets ratio index; BARD, BMI (body mass index), AST:ALT ratio, and diabetes; FAST, FibroScan-AST; FIB-4, Fibrosis-4 index; LRE, liver-related event; LSM, liver stiffness measurement; NAFLD, nonalcoholic fatty liver disease; NFS, NAFLD fibrosis score; NRI, net reclassification improvement.

^a Event NRI referred to the net proportion of LREs assigned a higher risk, which ranged from −1 to 1. Nonevent NRI referred to the net proportion of non-LREs assigned a lower risk, which ranged from −1 to 1. Overall NRI was the simple sum of event NRI and nonevent NRI, which was a crude summary of event NRI and nonevent NRI, ranged from −2 to 2. A positive NRI referred to an improvement in correct reclassification, while a negative NRI referred to a reduction in correct reclassification. The 95% CI for NRI was estimated using 1000 bootstrap samples.

Among patients with baseline Agile 3+ score less than 0.451, the incidence rate of LREs was 0.7 (95% CI, 0.5-1.0), for a score of 0.451 to 0.678, the incidence rate of LREs was 3.3 (95% CI, 2.1-5.2), and for a score of 0.679 or greater, the incidence rate of LREs was 24.9 (95% CI, 21.7-28.6) per 1000 person-years (P < .001) (Figure 2B; eTable 6 in Supplement 1). Among patients with baseline Agile 4 score less than 0.251, the incidence rate of LREs was 1.2 (95% CI, 1.0-1.5), for a score of 0.251 to 0.842, the incidence rate of LREs was 23.5 (95% CI, 19.6-28.2), and for a score of 0.843 or greater, the incidence rate of LREs was 105.5 (95% CI, 84.4-191.9) per 1000 person-years (P < .001). Among the noninvasive tests, the Agile 4 score classified the highest proportion (89.8%) of patients in the low-risk group with a sensitivity of 0.74 and negative predictive value of 0.99 for 3-year LREs (eTable 7 in Supplement 1). In contrast, it classified the fewest patients (1.4%) in the high-risk group compared with 14.3% for the Agile 3+ score. eFigure 6 in Supplement 1 shows the incidence of LREs in patients categorized by histologic fibrosis staging and other noninvasive tests. Similar to the AUROC analysis, AST to alanine aminotransferase ratio, BARD, and FAST were the least discriminatory.

The Agile scores consistently outperformed the other noninvasive tests in predicting LREs at 3 and 5 years in subgroups stratified by age, sex, presence of diabetes, BMI, and reliability of LSM (eFigure 7 in Supplement 1). Both Agile scores had higher AUROC in patients aged 60 years or older than in younger patients. The prognostic performance of the fibrosis scores was largely similar across regions (eTable 8 in Supplement 1).

Serial Model

Among 16 603 patients in the baseline model, 10 920 patients (65.8%) with repeated VCTE examinations at a median interval of 15 (IQR, 11.3-27.7) months were included in the serial model (Figure 1). The clinical characteristics at the first examination of the patients in the serial model were similar to those of patients in the baseline model (Table 1). Between the first and last VCTE examinations, the proportion of patients with diabetes and hypertension increased by approximately 12%. Using published cutoffs, the risk classification by Agile scores was stable when either 2 or 3 examinations were considered (Figure 3A; eFigures 8-11 in Supplement 1). A total of 81.9% of patients (7208 of 8810) had stable Agile 3+ scores and 92.1% of patients (8163 of 8810) had stable Agile 4 scores (same risk categories) at 2 examinations. Patients with a longer time interval between 2 tests were more likely to have increased scores at the second assessment, suggesting genuine fibrosis progression instead of variability in scores on repeated testing (eFigure 10 in Supplement 1). In general, the Agile scores and LSM had a higher stability than the other noninvasive tests (eFigure 11 in Supplement 1).

Figure 3. Agile 3+ Scores in the Serial Model.

LREs indicates liver-related events; VCTE, vibration-controlled transient elastography. Whiskers indicate 95% CIs.

^aRepresents the Agile 3+ score decreasing at least 10%, 20%, or 30% from the first to last test.

^bRepresents the Agile 3+ remaining stable from first to last test in the remaining patients.

^cRepresents the Agile 3+ score increasing at least 10%, 20%, or 30% from the first to last test.

eTables 9 and 10 in Supplement 1 show the incidence of LREs in patients with serial Agile scores. The incidence of LREs was 0.6 per 1000 person-years in patients with persistently low Agile 3+ scores and 30.1 per 1000 person-years in patients with persistently high Agile 3+ scores. In patients with high Agile 3+ score at the first examination but intermediate score at the last examination, the incidence of LREs decreased to 3.3 per 1000 person-years. A similar trend was observed for the Agile 4 score (eTable 10 in Supplement 1) and LSM (eTables 11 and 12 in Supplement 1). In contrast, patients who had worsened Agile 3+ scores at the last examination only had a mild increase in the risk of LREs compared with those who had stable scores (eTable 9 in Supplement 1). eTables 13 to 16 in Supplement 1 show consistent results in sensitivity analyses by including only patients who had 2 noninvasive tests performed within an interval of 3 years.

Apart from classifying patients into crude risk categories, another way to interpret serial test results is to determine their change over time. By restricted spline curve analysis, there was a positive nonlinear association between changes in Agile scores or LSM and the risk of LREs (eFigure 12 in Supplement 1). Regardless of baseline Agile scores and LSM, a 10% or greater relative decrease in the test results was associated with a lower risk of LREs, whereas an increase in the test results was associated with increased risk of events (Figure 3B; eTables 17-20 in Supplement 1). As expected, the greater the change observed in Agile scores or LSM (eg, 30% relative change), the greater the change observed in the incidence of LREs. Compared with patients with stable Agile scores, those with a 30% or greater relative increase in the scores had significant changes in all the components of the scores (eTable 21 in Supplement 1).

Discussion

In this large multicenter cohort study, we found that the Agile scores had better performance in predicting LREs in patients with MASLD than commonly used simple fibrosis scores. Although the difference in prognostication between the Agile scores and LSM might be marginal, the Agile scores were stable over time, and changes in the scores over time provide insights that can affect clinical management.

In the baseline model, both the Agile 3+ and Agile 4 scores had the highest overall accuracy in predicting LREs. Although both Agile scores had identical integrated AUROC, it should be noted that the Agile 4 score classified approximately 90% of patients in the low-risk group and in turn missed twice as many patients who would develop LREs as the Agile 3+ score. The Agile 4 score mainly improved classification of patients without LREs, while the Agile 3+ score improved the classification of events. This is understandable as the Agile 3+ score was designed to detect advanced fibrosis and the Agile 4 score was designed to detect cirrhosis.⁷ Therefore, the Agile 3+ score may be preferable for prognostic purposes, whereas the main value of the Agile 4 score is for the diagnosis of MASLD-related cirrhosis. It is also worth noting that the superiority of the Agile scores over LSM alone was marginal. While the calculation of the Agile scores is based on routine parameters and thus does not incur specific additional costs, clinicians who prefer to use LSM alone for the sake of simplicity can also refer to the detailed analysis on the prognostication by LSM in this study.

Analyzed separately, all noninvasive tests of fibrosis were better at predicting hepatic decompensation than HCC. This can be explained by the phenomenon of HCC occurring in a noncirrhotic liver. Although hepatic decompensation almost always develops in the background of cirrhosis, HCC appears to occur in a noncirrhotic liver more often in MASLD (approximately 30%) than other chronic liver diseases.^17,18

Compared with the existing literature,^8,19 our study found that some patients had significant changes in LSM and Agile scores over time. More than 80% of patients, in 2 or 3 assessments, remained within the same risk categories based on published Agile score cutoffs. MASLD progression from no to minimal fibrosis to cirrhosis or LREs typically spans 20 years.²⁰ Among patients with LSM and Agile score changes, reductions were more frequent than increases. Reduced LSM might reflect true fibrosis improvement due to lifestyle changes but most likely resulted from initial false positives, potentially explaining why decreased LSM was associated with a bigger change in LRE risk than increases. False-positive LSM has been reported in patients with factors such as extreme body build, acute hepatitis, congestive heart failure, biliary obstruction, amyloidosis, and recent food intake.²¹ In a previous study with a median 18-week interval between 2 VCTE examinations, 35% of patients with initially high LSM had normal LSM at the second assessment, with most showing no or mild fibrosis on subsequent liver biopsy.²² Similarly, in our study, patients with reduced LSM or Agile scores over time had a lower LRE incidence compared with those with higher readings. Therefore, patients with abnormal LSM or Agile scores should consider repeated examinations before deciding on liver biopsy or treatment.

While customary, interpreting noninvasive tests based on published cutoffs can be crude and misleading. Some individual test results do not cross these thresholds despite progression or regression, while minor fluctuations near cutoffs can lead to misinterpretation. To address this, we performed a restricted spline curve analysis, which revealed that Agile score and LSM changes are positively associated with LRE risk. Prior studies recommended a 20% LSM relative change for prognostication.^23,24 Our study provides detailed data on the prognostic importance of varying Agile score or LSM changes.

In comparison, serial FIB-4 has also been shown to be prognostic in the general population and hospital settings.^25,26 However, FIB-4 is inferior to LSM and other specific fibrosis biomarkers in the diagnosis of advanced fibrosis.²⁷ FIB-4 also has suboptimal performance at extremes of age.^28,29

According to the US Food and Drug Administration, to replace liver histologic findings as a surrogate end point in clinical trials, a biomarker should demonstrate the ability to diagnose the fibrosis stage, predict prognosis, monitor disease progression, and reflect response to treatment.⁵ Based on this and other studies, VCTE and the Agile scores have already fulfilled the first 3 requirements, but the latter requires a finding of correlation between histologic response and changes in noninvasive tests in clinical trials involving an effective treatment. There have already been efforts to fill this knowledge gap using data from several clinical trials,^30,31 and we expect an acceleration in the validation of response biomarkers when some of the ongoing phase 3 trials show positive results. Meanwhile, we believe the existing noninvasive tests can largely replace liver biopsies in routine practice.

Limitations

The study has several limitations. First, variable patient assessment intervals affect serial data interpretation, yet we analyzed noninvasive test changes and clinical outcomes after VCTE examinations interval stratification. Second, despite a sufficient sample size for clinical outcome evaluation, the 51.7-month median follow-up may be considered short, given the lengthy progression of chronic liver disease to cirrhosis and complications.³² Third, this was a natural history cohort. When effective treatment for steatohepatitis becomes available, studies should be conducted to identify suitable response biomarkers. Fourth, data of this study were from tertiary referral centers. The prognostic performance of VCTE and the Agile scores should be confirmed in a more general setting in the future. Although the Agile scores were compared with a number of simple fibrosis scores, future studies should compare the Agile scores with other specific biomarkers of fibrosis and/or steatohepatitis such as the enhanced liver fibrosis and blood-based biomarker scores.

Conclusions

In this cohort study, the VCTE-based Agile scores were found to be highly accurate in predicting LREs in patients with MASLD. In the short-term to medium-term context, the Agile scores have high stability on repeated testing. In less than half of patients with an early change in Agile scores, the lower score between 2 serial measurements more faithfully reflects the risk of LREs. In this situation, repeating Agile score measurements or testing another specific fibrosis biomarker should be contemplated before making a decision on liver biopsy or treatment. The findings suggest that VCTE-based Agile scores may be suitable alternatives to liver biopsy in routine clinical practice and in phase 2b and 3 clinical trials for steatohepatitis.

Supplement 1.

eMethods. Abbreviations, the Formula of the Non-Invasive Tests, eMethod Details

eAppendix. The VCTE-Prognosis Study Group: Collaborators

eFigure 1. AUROC and for the Prediction of Liver-Related Events by the Agile Scores and Other Non-Invasive Tests

eFigure 2. AUROC and AUPRC for the Prediction of Liver-Related Events by the Agile Scores and Other Non-Invasive Tests in Patients With Available Data for All Non-Invasive Tests (N = 10 678)

eFigure 3. Calibration Curve for Prediction of 3-Year Liver-Related Events

eFigure 4. Calibration Curve for Prediction of 5-Year Liver-Related Events

eFigure 5. AUROC and AUPRC for the Prediction of HCC and Hepatic Decompensation Risk by Agile Scores, Histological Fibrosis Stage and Other Non-Invasive Tests in the Baseline Model

eFigure 6. Cumulative Incidence of Liver-Related Events Stratified by Different Non-Invasive Tests

eFigure 7. Subgroup Analysis of Area Under The Receiver-Operating Characteristics Curve for Liver-Related Events at 3 and 5 Years by Age (Panel A), Diabetes Status (Panel B), Sex (Panel C), Body Mass Index (kg/m², Panel D) and LSM Reliable Criteria (Panel E)

eFigure 8. Sankey Diagrams Indicating the Change in Agile 3+ in Three Times of Vibration-Controlled Transient Elastography Examinations (N = 3 776)

eFigure 9. Sankey Diagrams Indicating the Change in Agile 4 in Three Times of Vibration-Controlled Transient Elastography Examinations (N = 3 776)

eFigure 10. Sankey Diagrams Indicating the Change in Agile Scores According to the Time Intervals Between Two Vibration-Controlled Transient Elastography Examinations (N = 3 307 Within 1 year, N = 4 490 1-3 Year, N = 903 3-5 Year)

eFigure 11. Sankey Diagrams Indicating the Change in Non-Invasive Tests at Two Examinations

eFigure 12. Restricted Cubic Spline Curves for the Association Between The Relative Change of Agile Scores or LSM and Risk of Liver-Related Events (Adjusted for Baseline Agile Scores and LSM)

eTable 1. Liver-Related Events

eTable 2. Integrated Area Under The Receiver-Operating Characteristic Curves and Brier Scores of Agile Scores, Histological Fibrosis Stage and Other Non-Invasive Tests in the Baseline Model

eTable 3. Paired Comparisons of the Agile Scores Versus Other Non-Invasive Tests and Liver Biopsy on the Discriminatory Performance for the Prediction of 3-Year and 5-Year Liver-Related Events in the Baseline Model

eTable 4. Paired Comparisons of the Agile Scores Versus Other Non-Invasive Tests and Liver Biopsy on the Discriminatory Performance for the Prediction of 3-Year and 5-Year HCC in the Baseline Model

eTable 5. Paired Comparisons of the Agile Scores Versus Other Non-Invasive Tests and Liver Biopsy on the Discriminatory Performance for the Prediction of 3-Year and 5-Year Hepatic Decompensation in the Baseline Model

eTable 6. Incident LREs in Patients With Different Agile Scores, Histological Fibrosis Stage and Other Non-Invasive Tests in the Baseline Model

eTable 7. Predictive Performance on 3-Year and 5-Year Liver-Related Events by Agile Scores, Histological Fibrosis Stage and Other Non-Invasive Tests Based on Published Cut-Offs in the Baseline Model

eTable 8. Predictive Performance on 3-Year and 5-Year Liver-Related Events by Agile Scores and Other Non-Invasive Tests in Western or Asian Countries

eTable 9. Transitioning Between Risk Groups Based on Agile 3+ (N = 8700) and Incident Liver-Related Events After the Last Test in the Serial Model

eTable 10. Transitioning Between Risk Groups Based on Agile 4 (N = 8 700) and Incident Liver-Related Events After the Last Test in the Serial Model

eTable 11. Transitioning Between Risk Groups Based on LSM at the Cut-Offs of 10 and 15 kPa (N = 10 920) and Incident Liver-Related Events After the Last Test in the Serial Model

eTable 12. Transitioning Between Risk Groups Based on LSM at the Cut-Offs of 8 and 12 kPa (N = 10 920) and Incident Liver-Related Events After the Last Test in the Serial Model

eTable 13. Transitioning Between Risk Groups Based on Agile 3+ (N = 7 797) and Incident Liver-Related Events After the Last Test in the Serial Model (Time Interval Between Tests Less Than 3 Years)

eTable 14. Transitioning Between Risk Groups Based on Agile 4 (N = 7 797) and Incident Liver-Related Events After the Last Test in the Serial Model (Time Interval Between Tests Less Than 3 Years)

eTable 15. Transitioning Between Risk Groups Based on LSM at the Cut-Offs of 10 and 15 kPa (N = 9 105) and Incident Liver-Related Events After the Last Test in the Serial Model (Time Interval Between Tests Less Than 3 Years)

eTable 16. Transitioning Between Risk Groups Based on LSM at the Cut-Offs of 8 and 12 kPa (N = 9 105) and Incident Liver-Related Events After the Last Test in the Serial Model (Time Interval Between Tests Less Than 3 Years)

eTable 17. The Relative Change of Agile 3+ (N = 8 700) and Incident Liver-Related Events After the Last Test in the Serial Model

eTable 18. The Relative Change of Agile 4 (N = 8 700) and Incident Liver-Related Events After the Last Test in the Serial Model

eTable 19. The Relative Change of LSM at the Cut-Offs of 10 and 15 kPa (N = 10 920) and Incident Liver-Related Events After the Last Test in the Serial Model

eTable 20. The Relative Change of LSM at the Cut-Offs of 8 and 12 kPa (N = 10 920) and Incident Liver-Related Events After the Last Test in the Serial Model

eTable 21. Factors Associated With the Relative Changes in Agile Scores and LSM

Supplement 2.

Data Sharing Statement