PROGNOSTIC ACCURACY OF TRANSIENT ELASTOGRAPHY-BASED PREDICTORS IN DIABETES AND OBESITY: A MULTICENTER INTERNATIONAL COHORT STUDY

Abstract

Background/Aims:

Liver stiffness measurement (LSM) by vibration-controlled transient elastography (VCTE) is recommended for risk stratification of patients with non-alcoholic fatty liver disease (NAFLD). More recently, AGILE3+ and AGILE4 have combined LSM with clinical parameters to identify patients with advanced fibrosis and cirrhosis, respectively. However, there are limited data on prognostic performance of these scores in key at-risk subgroups such as those with diabetes and obesity compared to LSM alone.

Methods:

This is a retrospective cohort study including 1,903 adult patients with NAFLD from tertiary care centers in the United States and Singapore undergoing VCTE between 2015 and 2022. Primary predictors were FAST, LSM, AGILE3+, and AGILE4 scores and the primary outcome was liver-related events (LRE). Patients were further stratified by diabetes and obesity status. Prognostic performance was measured using the time-dependent area under the receiver operating characteristic curve (tAUC) at 5 years.

Results:

In total, 25 LRE occurred and the overall incidence rate of LRE was 4.4 per 1,000 person-years. tAUC for predicting LRE in the overall group was significantly higher with AGILE3+ (0.94 [95% CI: 0.90–0.98]) and AGILE4 (0.94 [95% CI: 0.90–0.98]) compared to LSM (0.87 [95% CI: 0.80–0.94]) (p = 0.001 and 0.009, respectively) and FAST (0.73 [95% CI: 0.59–0.86]) (p < 0.001 for both). Similarly, tAUC was significantly higher in those with T2D for AGILE3+ compared to LSM (0.92 vs 0.86, respectively) (p = 0.015) and FAST (0.92 vs 0.73, respectively) (p = 0.008). Among people with obesity, tAUC was significantly higher for AGILE3+ compared to LSM (0.95 vs 0.89, respectively) (p = 0.005) and FAST (0.95 vs 0.76, respectively) (p = 0.0035). Though AGILE4 had a higher tAUC in these subgroups compared to LSM, it did not reach statistical significance.

Conclusion:

AGILE3+ significantly outperforms LSM and FAST for predicting LRE in patients with NAFLD including in those with diabetes or obesity.

INTRODUCTION:

Non-alcoholic fatty liver disease (NAFLD) is currently one of the leading causes of advanced chronic liver disease worldwide as well as indication for liver transplant.^1–3 However, most patients with NAFLD do not progress to cirrhosis or liver decompensation, and with a global prevalence of nearly 30% and still rising there is an increasing need to accurately identify those who are at high risk for developing major complications while avoiding unnecessary interventions on patients who are very low risk for progressive disease.⁴ Previous studies have clearly demonstrated the increasing risk of liver related events (LRE) including hepatic decompensation and hepatocellular carcinoma, and mortality in those with more advanced fibrosis indicating fibrosis is the primary driver for prognosis.^5,6 The mortality of NAFLD patients who develop advanced fibrosis is increased exponentially compared to the general population with liver-related death a leading cause of mortality in this population.⁷ However, early and reliable recognition of advanced fibrosis and its complications through non-invasive methods continues to be a challenge.

Non-invasive tests (NITs) for diagnosing advanced fibrosis and predicting LRE have largely replaced invasive measure for risk stratification in routine practice.^8–10 Liver stiffness measurement (LSM) by vibration-controlled transient elastography (VCTE) is one widely used NIT both to identify patients with prevalent fibrosis and at high risk of future LRE.^10,11 However, sensitivity of LSM alone for predicting LRE may be lower in those with obesity or T2D.^12–14 Given known limitations of LSM, there has been an interest in combining LSM with serologic tests to more accurately stratify risk of disease. Two such tests include AGILE 3+ (age, sex, AST, ALT, platelets, diabetes status, and LSM) and AGILE-4 (sex, AST, ALT, platelets, diabetes status, and LSM).^15,16 Recent studies have shown these index tests have considerable value in detecting fibrosis in earlier stages and increasing negative predictive value leading to less false negatives, compared to LSM alone.^15,17,18 This allows us to risk stratify patients better, monitor disease progression, and escalate medical therapy appropriately.

However, there are limited data regarding these tests’ performance in key at-risk subgroups such as individuals with T2D and obesity. Therefore, we assessed whether the prognostic accuracy of these NITs in predicting LRE is affected by T2D or obesity in comparison to those without.

METHODS:

Study design and patient selection

This is an international multi-center, retrospective cohort study that included 1,903 adult NAFLD patients (age ≥ 18 years) consecutively diagnosed from United States (University of Michigan and University of Arizona) and Singapore (Changi General Hospital) between 2015 and 2022. This study was considered exempt by the Institutional Review Boards of University of Michigan, Changi General Hospital, and Banner Health.

In the Arizona cohort, NAFLD was diagnosed using International Classification of Diseases −9 and −10 (ICD) codes with exclusion of other etiologies of liver disease (Supp. Table 1). The authors conducted random manual confirmations of the Arizona cohort through electronic medical record reviews of imaging studies to assess the accuracy of ICD codes in identifying NAFLD. These confirmations were found to be accurate for use given the high positive predictive values (PPV) of >95% based on our prior study.¹⁹

In the Singapore cohort, all cases were manually reviewed for imaging evidence of hepatic steatosis. In the Michigan cohort, NAFLD was diagnosed based on imaging evidence of hepatic steatosis using ultrasound, computed tomography, or magnetic resonance imaging, followed by exclusion of other etiologies of liver disease (Supp. Table 1) using a previously validated natural language processing algorithm with positive predictive value >90%.²⁰

Patients with documented evidence for other liver diseases or significant alcohol intake (>1 drink per day in women or >2 drinks per day in men²¹) were excluded. Time was recorded in days from diagnosis of NAFLD to clinical event of interest.

Non-invasive assessments (LSM by VCTE, FAST, AGILE 3+, AGILE-4)

VCTE were performed by trained health care staff using either a M or XL probe per manufacturer instructions. LSM was measured as a median of at least 10 successful measurements. Unreliable LSMs were defined as interquartile range >30% of the median or <10 successful measurements and were excluded from the study. LSM was stratified by low (<8 kPa), intermediate (8–12 kPa), and high (>12 kPa) risk.^22,23

FAST score was stratified by low (<0.35) and high (>0.67) risk to rule out and rule in stage ≥2 fibrosis, respectively.²⁴ AGILE3+ was stratified by low (<0.48) and high (>0.65) risk to rule out and rule in stage ≥3 fibrosis, respectively.¹⁶ AGILE4 was stratified by low (<0.25) and high (>0.57) risk to rule out and rule in stage 4 fibrosis, respectively.¹⁶ Equations for FAST, AGILE3+, and AGILE4 are below:

$$\mathit{\text{FAST}}=\frac{e^{-1.65+1.07*\text{ln}\left(LSM\right)+2.66*{10}^{-8}*CA{P}^3-63.3*AS{T}^{-1}}}{1+e^{-1.65+1.07*\text{ln}\left(LSM\right)+2.66*{10}^{-8}*CA{P}^3-63.3*AS{T}^{-1}}}$$

$$\mathit{\text{Agile}}3+=\frac{e^{-3.92368+2.29714*\text{ln}\left(LSM\right)-0.00902*PLT-0.98633*AA{R}^{-1}+1.08636*\mathit{\text{Diabetes status}}-0.38581*Sex+0.03018*Age}}{1+e^{-3.92368+2.29714*\text{ln}\left(LSM\right)-0.00902*PLT-0.98633*AA{R}^{-1}+1.08636*\mathit{\text{Diabetes status}}-0.38581*Sex+0.03018*Age}}$$

$$\mathit{\text{Agile}}4=\frac{e^{7.50139-15.42498*\frac{1}{\sqrt{LSM}}-0.01378*PLT-1.41149*AA{R}^{-1}-0.53281*Sex+0.41741*\mathit{\text{Diabetes status}}}}{1+e^{7.50139-15.42498*\frac{1}{\sqrt{LSM}}-0.01378*PLT-1.41149*AA{R}^{-1}-0.53281*Sex+0.41741*\mathit{\text{Diabetes status}}}}$$

Subgroup identification

Patients with T2D were identified using ICD scores among all three centers. Obesity was defined by the World Health Organization recommendations for Asian and non-Asian BMI cut-offs.^25,26 For non-Asians, underweight was defined as BMI < 18.5, normal was defined as BMI 18.5 – 24.9, overweight was defined as BMI 25 – 29.9, and obesity was defined as BMI > 30. For Asians, underweight was defined as BMI < 17.5, normal was defined as BMI 17.5 – 22.9, overweight was defined as BMI 23–27.49, and obesity was defined as BMI > 27.5.²⁷

Outcomes measures

The primary outcome for this study was LRE defined as new onset hepatocellular carcinoma (HCC), hepatic decompensation (West Haven grade 2–4 hepatic encephalopathy; spontaneous bacterial peritonitis diagnosed by paracentesis; ascites requiring diuretics, paracentesis, or transjugular intrahepatic portosystemic shunt; or variceal bleeding as confirmed endoscopically), or liver related mortality.^28,29 Secondary outcomes included a composite of LRE combined with all-cause mortality, and MACE, which was a composite end point of myocardial infarction, coronary revascularization, heart failure requiring hospitalization, or embolic stroke as per our prior study.²⁸ Given HCC can occur in NAFLD in the absence of cirrhosis or portal hypertension, we also performed a sensitivity analysis among the NITs in predicting portal hypertension associated LRE (hepatic encephalopathy, spontaneous bacterial peritonitis, ascites, or variceal bleeding).^30,31

We screened for clinical outcomes using ICD codes (Supp. Table 1). All clinical outcomes for the Singapore and Michigan cohorts were verified through manual chart review. In the Arizona cohort, ICD codes were used to identify clinical outcomes with random manual confirmations; the codes demonstrated high positive predictive value as previously reported.¹⁹

Patients were excluded (1) if they had documented evidence of LRE prior to NAFLD diagnosis, (2) if an event occurred within the first 6 months to avoid misclassifying prevalent disease as incident, or (3) if they had less than 6 months of follow up. We report the incidence rates (events per 1000-person years) of events to account for the varying follow-up period between individual patients.

Statistical analysis

Categorical variables are reported as percentages while quantitative variables are reported as means with standard deviation. Pearson’s Chi-squared or Fisher’s exact tests along with Wilcox rank sum test were used to compare dichotomous and continuous variables, respectively. The LRE and MACE outcomes were run as Fine-Gray competing risk analyses, with death without LRE or MACE, respectively, as the competing events. The overall mortality and LRE or death outcomes were run as Cox proportional hazards models. The cumulative incidence and incidence rates (events per 1000-person years) were compared between the NITs with stratification by risk category.

The prognostic performance among NITs at predicting LRE, extrahepatic events, and LRE/death was compared using the time-dependent area under the receiver operating characteristic curve (tAUC) at 5 years. High accuracy of the prognostic performance is defined as tAUC > 90%. The Delong test was used to compare tAUC for the various NITs in the overall and subgroup populations based on T2D and obesity. All data analysis was performed using R (version 4.3.3) and R-studio (version 2023.12.1+402) with tidycmprsk, pROC, and survival packages. Statistical significance was set at p < 0.05.

RESULTS:

Baseline demographics

A total of 1,903 patients were included (Figure 1). The overall cohort was predominantly White (55%) and Asian (20%) with an overall mean (± SD) age of 52 (± 14) years with the Asian cohort being significantly older (57 vs 50 years, p <0.001). In general, 62% of the entire cohort had obesity and 40% had T2D. There was no difference in LSM between the two cohorts with a mean of 9.3 kPa in the overall cohort. The mean FAST, AGILE3+, and AGILE4 scores were 0.40, 0.33, and 0.10. There was no difference in FAST or AGILE4 scores between the cohorts, but the Asian cohort had slightly higher AGILE3+ score (0.40 vs. 0.31) than the US cohorts, which is potentially driven by the older age of the Asian cohort.

Figure 1. Study flowchart.

Abbreviations: LSM, liver stiffness measurement; LRE, liver related events

Incidence of intra and extra-hepatic outcomes

Liver related events:

In total, 25 LRE occurred and the overall incidence rate of LRE was 4.4 (95% CI: 2.8–6.6) per 1,000 person-years with a total of 5,627 person-years of follow-up. Increasing LSM was associated with a significantly higher incidence rate (per 1,000 person-year) of LRE: 0.6 (95% CI: 0.1–2.1), 1.8 (95% CI: 0.2–6.6), and 19.3 (95% CI: 12.0–29.5) for LSM <8, 8–12, or >12 kPa, respectively (p < 0.001) (Table 2). This similar trend was also noted with FAST, AGILE3+, and AGILE4 scores where increasing scores were associated with higher risk of LRE (Figure 2). Increasing AGILE3+ score was associated with a significantly higher incidence rate (per 1,000 person-year) for low, medium, and high-risk scores: 0.0 (95% CI: 0.0–0.9), 3.7 (95% CI: 0.5–13.5), 19.6 (95% CI: 12.4–29.3), respectively (p < 0.001). AGILE4 also demonstrated the same trend in incidence rate (per 1,000 person-year) for low, medium, and high-risk score: 0.8 (95% CI: 0.2–2.1), 13.9 (95% CI: 6.0–27.3), 59.6 (95% CI: 31.7–102.0), respectively (p < 0.001). Lastly, increasing FAST score also had a significantly higher incidence rate (per 1,000 person-year) for low, medium, and high-risk scores: 1.9 (95% CI: 0.5–4.8), 1.9 (95% CI: 0.4–5.4), 14.1 (95% CI: 7.2–24.6), respectively (p < 0.001).

Table 2:

Incidence rates (per 1000 person-years) of LRE and LRE/death in the overall cohort stratified based on risk of non-invasive test scores.

Test	Cutoff	N	Events	Incidence rate	95% CI	p-value
Liver-related events
LSM	≤ 8 kPa	1155	2	0.6	.07 – 2.1	< 0.0001
	8–12 kPa	375	2	1.8	0.2 – 6.6
	≥ 12 kPa	373	21	19.3	12.0 – 29.5
FAST	≤ 0.35	773	4	1.9	0.5 – 4.8	< 0.0001
	0.35–0.67	561	3	1.9	0.4 – 5.4
	≥ 0.67	288	12	14.1	7.2 – 24.6
AGILE3+	≤ 0.48	1341	0	0	0 – 0.9	< 0.0001
	0.48–0.65	182	2	3.7	0.5 – 13.5
	≥ 0.65	380	23	19.6	12.4 – 29.3
AGILE4	≤ 0.25	1655	4	0.8	0.2 – 2.1	< 0.0001
	0.25–0.57	168	8	13.9	6.0 – 27.3
	≥ 0.57	80	13	59.6	31.7 – 102.0
Liver-related events or death
LSM	≤ 8 kPa	1155	10	2.8	1.4–5.2	< 0.0001
	8–12 kPa	375	6	5.4	2.0 – 11.7
	≥ 12 kPa	373	26	23.6	15.4 – 34.6
FAST	≤ 0.35	773	11	5.1	2.5 – 9.1	0.0009
	0.35–0.67	561	7	4.2	1.7 – 8.7
	≥ 0.67	288	16	18.5	10.6 – 30.1
AGILE3+	≤ 0.48	1341	6	1.5	0.6–3.3	< 0.0001
	0.48–0.65	182	6	10.9	4.0 – 23.8
	≥ 0.65	380	30	25.1	17.0 – 35.9
AGILE4	≤ 0.25	1655	18	3.6	2.2 – 5.8	< 0.0001
	0.25–0.57	168	9	15.5	7.1 – 29.4
	≥ 0.57	80	15	67.5	37.8 – 111.4

Incidence rates are shown as events per 1,000 person-years of follow-up (95% confidence interval). p-value reflects the differences in incidence rates between the risk stratification of each non-invasive test using the Poisson test. LSM, liver stiffness measurement; FAST, Fibroscan-aspartate aminotransferase score.

Figure 2. Cumulative incidence of liver related events stratified based on risk of (A) LSM, (B) FAST, (C) AGILE3+, and (D) AGILE4 scores.

p-value reflects the differences between the risk stratification of each non-invasive test using Fine-Gray competing risk models. Abbreviations: LSM, liver stiffness measurement; LRE, liver related events; FAST, Fibroscan-aspartate aminotransferase score

Liver related events or all-cause mortality:

In total, 42 LRE or deaths occurred and the overall incidence rate (per 1,000 person-years) of LRE/death was 7.3 (95% CI: 5.3–9.9) with a total of 5,745 person-years of follow-up. All of the NITs correlated with significantly increased incidence rates with increasing scores (Table 2).

Major adverse cardiac events:

In total, 18 MACE occurred and the overall incidence rate (per 1,000 person-years) was 3.7 (95% CI: 2.2–5.8) with a total of 4,877 person-years of follow-up. LSM, FAST, and AGILE4 were not significantly associated with increased risk of MACE in the overall population despite risk stratification of the scores (Table 3). However, only AGILE3+ showed a significant increase in incidence rate of MACE in the overall population with higher scores. Subgroup analysis showed no difference in incidence rates of MACE in the Asian cohort among all NITs despite risk stratification (Supp. Table 2). Interestingly, only the American cohort showed a statistically significant rise in incidence rate of MACE with increasing AGILE3+ score.

Table 3:

Incidence rates (per 1000 person-years) of major adverse cardiac events in the overall cohort stratified based on risk of non-invasive test scores.

Test	Cutoff	N	Events	Incidence rate (95% CI)	p-value
LSM	≤ 8 kPa	1002	9	2.9 (1.3 – 5.5)	0.451
	8–12 kPa	287	3	3.4 (0.7 – 9.9)
	≥ 12 kPa	282	6	6.8 (2.5 – 14.7)
FAST	≤ 0.35	558	8	4.8 (2.1 – 9.4)	0.71
	0.35–0.67	487	5	3.4 (1.1 – 7.9)
	≥ 0.67	241	4	5.4 (1.5 – 13.8)
AGILE3+	≤ 0.48	1122	7	2.2 (0.8 – 4.2)	0.0017
	0.48–0.65	144	3	6.9 (1.4 – 20.0)
	≥ 0.65	305	8	8.2 (3.6 – 16.2)
AGILE4	≤ 0.25	1365	14	3.3 (1.8 – 5.6)	0.235
	0.25–0.57	142	3	6.2 (1.3 – 18.0)
	≥ 0.57	64	1	5.3 (0.1 – 29.3)

Incidence rates are shown as events per 1,000 person-years of follow-up (95% confidence interval). p-value reflects the differences in incidence rates between the risk stratification of each non-invasive test using the Poisson test. LSM, liver stiffness measurement; FAST, Fibroscan-aspartate aminotransferase score.

Prognostic accuracy of AGILE scores:

Predicting liver related events:

AGILE scores were significantly better at predicting LRE compared to LSM and FAST in the overall cohort (Table 4). tAUC for predicting LREs in the overall group was significantly higher with AGILE3+ (0.94 [95% CI: 0.90–0.98]) and AGILE4 (0.94 [95% CI: 0.90–0.98]) compared to LSM (0.87 [95% CI: 0.80–0.94]) (p = 0.001 and 0.009, respectively) and FAST (0.73 [95% CI: 0.59–0.86]) (p < 0.001 for both).

Table 4:

Time-dependent area under the curve comparison at predicting liver-related events in the overall, type 2 diabetes, and obese cohorts.

Test	Cohort
	Overall	Type 2 Diabetes	Obese
LSM	0.867 (0.797–0.938) +	0.861 (0.791–0.930) +	0.893 (0.832–0.953)
FAST	0.725 (0.590–0.859)	0.732 (0.557–0.908)	0.763 (0.601–0.925)
AGILE3+	0.938 (0.901–0.975) *+	0.918 (0.861–0.974) *+	0.947 (0.903–0.992) *+
AGILE4	0.938 (0.900–0.976) *+	0.911 (0.846–0.977) +	0.934 (0.869–0.998) +

Values are reported as time-dependent area under the curve (95% confidence interval). p-value reflects the differences between tAUC for each non-invasive test using the Delong’s test. LSM, liver stiffness measurement; FAST, fibroscan-aspartate aminotransferase score.

^*p < 0.05 vs LSM.

⁺p < 0.05 vs FAST.

Similarly, tAUC was significantly higher in those with T2D for AGILE3+ (0.92 [95% CI: 0.86–0.97]) compared to LSM (0.86 [95% CI: 0.79–0.93]) (p = 0.015) and FAST (0.73 [95% CI: 0.56–0.91]) (p = 0.008). In people with T2D, tAUC for AGILE4 (0.91 [95% CI: 0.85–0.98]) also has borderline higher tAUC compared to LSM, but it did not reach statistical significance (p = 0.10). AGILE4 however did significantly outperform FAST (p = 0.007) in those with T2D.

Among people with obesity, tAUC was significantly higher for AGILE3+ (0.95 [95% CI: 0.90–1]) and AGILE4 (0.93 [95% CI: 0.87–1]) compared to FAST (0.76 [95% CI: 0.60–0.93]) (p = 0.0035 and 0.0036, respectively). In this subgroup, only AGILE3+ had significantly higher tAUC than LSM (0.89 [95% CI: 0.83–0.95]) for predicting LRE (p = 0.005), but not AGILE4 (p = 0.126).

Predicting portal hypertension associated liver related events:

AGILE scores were significantly better at predicting portal hypertension associated LRE compared to LSM and FAST in the overall cohort (Supp. Table 4). Despite AGILE scores having a higher tAUC among patients with T2D and obesity compared to FAST and LSM, they were only significantly better than FAST. Notably, only AGILE3+ showed a significant improvement over LSM in patients with obesity, but not in patients with T2D.

Predicting major adverse cardiac events:

None of the NITs meaningfully predict MACE with tAUC (Supp. Figure 1) (Supp. Table 3). However, both AGILE scores significantly outperformed LSM at predicting MACE in obese cohorts, but not in the T2D cohort. The AGILE scores also did not significantly outperform FAST at predicting MACE.

DISCUSSION:

The findings from our study at predicting LRE through use of AGILE scores are consistent with previous European and Asian studies.^15,17 We found that AGILE scores have a significantly higher accuracy in prognostication compared to LSM or FAST. This was also true in subgroup analysis of at-risk patients with diabetes or obesity where AGILE3+ outperformed both LSM and FAST. However, AGILE4 was not significantly better than LSM in the diabetes and obese populations even though it had a higher tAUC. When comparing AGILE scores’ performance in predicting only portal hypertension-associated LRE, their utility may be limited in the diabetic population when compared to LSM. From a clinical perspective, these findings may support the use of AGILE scores—especially AGILE3+ in which few patients fell in the indeterminate range—in practice for risk stratification regardless of patient comorbidities and ethnicities, and thus allowing for increased surveillance in high risk NAFLD patients.

Significant fibrosis in NAFLD is currently diagnosed using LSM in combination with various other NITs such as Fibrosis-4.³² Recent studies demonstrated discordance in stratifying key at-risk patients using Fibrosis-4 and LSM, and the rising need for alternative predictive methods in such populations.³³ The AGILE scores were created to address this gap and there is growing evidence for their potential as predictors for both advanced fibrosis and prognosis while replacing the need for a liver biopsy.^15,17,18 In our multicenter international study that included 1,903 patients with NAFLD, we found that the AGILE scores had excellent prognostic accuracy even in racially diverse populations when compared to LSM regardless of diabetes or obesity status. This was proven by the increased incidence rate of LRE and LRE/death as well as higher tAUC for predicting LRE with AGILE scores when directly compared to LSM and FAST.

Previous studies on Asian populations only have not shown significant benefit for predicting MACE using AGILE scores.¹⁷ Our findings from the Singapore cohort were consistent with these findings, but in the US cohort we found a significantly higher incidence rate of MACE in those with higher AGILE3+ scores. The reasons for this finding are unclear. Fibrosis has been consistently associated with risk of cardiovascular disease in multiple studies, and perhaps the higher accuracy of AGILE scores for fibrosis (vs. LSM alone) partially explains this finding. The higher prevalence of obesity in the US cohorts may also be relevant since in subgroup analysis AGILE scores outperformed LSM for predicting MACE only in the obesity sub-cohorts. The difference in statistical power between the two cohorts could also account for this finding. Regardless, our data on prognosticating MACE through AGILE scores must be interpreted with caution given relatively low tAUC.

The strengths of this study include manual verification and validation of clinical events through chart review in the Michigan and Singapore cohorts. Prior studies on using AGILE scores for predicting LRE were done mostly in the European and Asian populations. This study is one of the first to report AGILE scores’ ability for prognostication purposes in the racially diverse American population as well as in key at risk subgroups. One main limitation is that our study included patients who are part of tertiary care centers and the use of LSM by VCTE is currently limited to such centers. The incidence of cirrhosis and LRE in these patients can be expected to be higher due to referrals than those in primary or secondary care centers. Though the generalizability of these findings to primary care settings can be challenging, this limitation is fairly natural to all studies using VCTE as it is rarely used in the primary care setting. Second, it was not realistically feasible to obtain the exact timing of when NAFLD developed in these patients, so we relied on the timing of earliest available imaging evidence per electronic health record. Another limitation is the generalizability of our findings to Hispanic populations, who are disproportionately affected by NAFLD and its outcomes.³⁴ More studies that validate the use of these scores in Hispanic populations would be required. Additionally, the variable methods in identifying patients with NAFLD among the three institutions can introduce bias and must be taken into account while interpreting and generalizing the data. Lastly, this was not a prospective study so alcohol biomarkers and validated questionnaires were not available, which may under-estimate alcohol-use in our cohort.

In conclusion, AGILE scores have excellent prognostic ability for predicting LRE not only in racially diverse populations, but also in key at risk subgroups. These scores have implications in tertiary care settings by allowing us to more accurately risk stratify patients for improved individualized patient care. Stratifying such patients can help guide treatment strategies and deciding which patients can be referred back to primary care for follow up. Furthermore, these NITs can also be used in selecting patients for clinical trials without the need for histological evidence for advanced fibrosis. These scores can help alleviate the healthcare burden associated with NAFLD complications especially given the growing incidence and prevalence of NAFLD worldwide.

Figure 3. Time-dependent area under the curve comparison at predicting liver related events in the (A) overall, (B) type 2 diabetes, and (C) obesity cohorts using non-invasive tests.

Abbreviations: NAFLD, non-alcoholic fatty liver disease; LSM, liver stiffness measurement; FAST, fibroscan-aspartate aminotransferase score; T2D, type 2 diabetes; AUROC, area under the receiver operating characteristic curve; CI, confidence interval

Table 1:

Baseline characteristics of the overall, American, and Singapore cohorts.

Variable	Overall (N = 1,903)	American (N = 1,469)	Asian (N = 434)	p-value
Male	921 (48%)	667 (45%)	254 (59%)	<0.001
Age	52 ± 14	50 ± 14	57 ± 13	<0.001
Race
African American	151 (7.9%)	81 (5.5%)	70 (16%)	-
Asian	389 (20%)	79 (5.4%)	310 (71%)	-
Caucasian	1,050 (55%)	1,050 (71%)	0 (0%)	-
Hispanic	228 (12%)	202 (14%)	26 (6.0%)	-
Other*	84 (4.4%)	57 (3.9%)	27 (6.2%)	-
Comorbidities
Type 2 diabetes	755 (40%)	526 (36%)	229 (53%)	<0.001
Hypertension	911 (48%)	634 (43%)	277 (64%)	<0.001
Dyslipidemia	1,103 (58%)	769 (52%)	334 (77%)	<0.001
BMI (kg/m 2 )	34 ± 9	35 ± 9	28 ± 5	<0.001
Underweight	11 (0.6%)	8 (0.6%)	3 (0.7%)	-
Normal	128 (7%)	74 (5.3%)	54 (13%)	-
Overweight	475 (26%)	314 (22%)	161 (37%)	-
Obese	1,214 (66%)	1,000 (72%)	214 (50%)	-
Laboratory values
ALP (U/L)	90 ± 52	94 ± 56	77 ± 31	<0.001
AST (U/L)	42 ± 40	43 ± 44	41 ± 25	>0.9
ALT (U/L)	61 ± 88	64 ± 98	54 ± 37	0.048
Albumin (mg/dL)	4.42 ± 0.39	4.41 ± 0.37	4.47 ± 0.44	<0.001
Platelets, (10^3/μL)	249 ± 77	250 ± 76	243 ± 81	0.012
FAST	0.40 ± 0.25	0.40 ± 0.25	0.38 ± 0.26	0.086
≤ 0.35	773 (48%)	640 (47%)	133 (51%)	-
0.35–0.67	561 (35%)	481 (35%)	80 (30%)	-
≥ 0.67	288 (18%)	238 (18%)	50 (19%)	-
LSM (kPa)	9.3 ± 7.8	9.4 ± 8.1	9.1 ± 6.9	0.6
≤ 8 kPa	1,155 (61%)	886 (60%)	269 (62%)	-
8–12 kPa	375 (20%)	308 (21%)	67 (15%)	-
≥ 12 kPa	373 (20%)	275 (19%)	98 (23%)	-
CAP (dB/m)	315 (58%)	317 (58%)	303 (57%)	<0.001
Agile 3+	0.33 ± 0.31	0.31 ± 0.30	0.40 ± 0.34	<0.001
≤ 0.48	1,341 (70%)	1,073 (73%)	268 (62%)	-
0.48–0.65	182 (9.6%)	144 (9.8%)	38 (8.8%)	-
≥ 0.65	380 (20%)	252 (17%)	128 (29%)	-
Agile 4	0.10 ± 0.18	0.09 ± 0.17	0.12 ± 0.20	0.12
≤ 0.25	1,655 (87%)	1,303 (89%)	352 (81%)	-
0.25–0.57	168 (8.8%)	110 (7.5%)	58 (13%)	-
≥ 0.57	80 (4.2%)	56 (3.8%)	24 (5.5%)	-

Continuous variables are reported as mean (standard deviation) and compared using a rank sum test. Categorical variables are reported as N (%) and compared using a chi-squared test. Abbreviations: FAST, fibroscan-aspartate aminotransferase score; AST, aspartate aminotransferase; ALT, alanine aminotransaminase; ALP, alkaline phosphatase.

^*= other races include Native American or multiracial

ABBREVIATIONS:

NAFLD

non-alcoholic fatty liver disease

ICD

International Classification of Diseases

LSM

liver stiffness measurement

VCTE

vibration-controlled transient elastography

NIT

non-invasive tests

FAST

Fibroscan-aspartate aminotransferase score

LRE

liver related events

MACE

major adverse cardiac events

tAUC

time-dependent area under the operative characteristic curve

AST

aspartate aminotransferase

ALT

alanine aminotransaminase

ALP

alkaline phosphatase

T2D

type 2 diabetes

BMI

body mass index