Optimizing Surveillance of Low-Risk Metabolic Dysfunction Associated Steatotic Liver Disease Using Transient Elastography

Abstract

Background

Most people with metabolic dysfunction associated steatotic liver disease (MASLD) lack significant fibrosis and are considered low-risk. Surveillance strategy for low-risk MASLD remains uncertain.

Aim

Identify which low-risk subjects can avoid follow-up vibration controlled transient elastography (VCTE) within one year.

Methods

Retrospective analysis of two independent low-risk MASLD cohorts (baseline liver stiffness (LS) < 8kPa) with routine 6–12 months follow-up VCTE. The primary outcome was LS ≥ 8kPa on follow-up, requiring referral and further work-up according to current guidance. Predictors of the primary outcome on univariate and multivariate logistic regression were incorporated into a decision algorithm, and validated in an independent cohort.

Results

Of 206 subjects in the derivation cohort, 96 were low-risk. After a median 10 months, 24 (25%) low-risk subjects had LS ≥ 8kPa. Baseline LS (p < 0.01) and ALT change from baseline (p = 0.02) (multivariate AUROC = 0.84 [0.74–0.94]) predicted the primary outcome, and were incorporated to a two-step decision algorithm. Low-risk subjects with baseline LS < 5.5kPa can avoid repeating VCTE in a year, while those with LS > 6.8kPa require one. For intermediate baseline LS (5.5–6.8kPa), repeat VCTE is only indicated when ALT increase > 6U/L. The algorithm had 92% negative predictive value (NPV), 78% specificity, and 78% accuracy in the derivation cohort. In the validation cohort (n = 64), it had 91% NPV, 72% specificity, and 71% accuracy.

Conclusion

In low-risk MASLD, a simple algorithm combining baseline LS and ALT change can be used to safelt avoid a repeat VCTE in a year.

INTRODUCTION

Metabolic dysfunction associated steatotic liver disease (MASLD) is a global health burden and is the most common cause for chronic liver disease. Global prevalence of MASLD is rising at an alarming rate and is currently estimated to be around 32%.(1) Fibrosis stage is the major determinant of long term outcomes in individuals with MASLD, including overall and liver related mortality,(2) and people with significant fibrosis (stage ≥ 2) should be referred to a specialist for further evaluation and management.(3) In contrast, people with low-risk MASLD, defined as fibrosis stage < 2, do not require immediate further evaluation.(3, 4) Multiple factors can determine fibrosis progression in MASLD, of which the presence of insulin resistance and type 2 diabetes mellitus play a major role.(5) Obesity is closely linked to diabetes and insulin resistance, and in turn is associated with MASLD. Prospective studies have shown that an increase in body mass index (BMI) over time can increase the risk of MASLD.(6) A subset of people with low-risk MASLD will progress to more advanced fibrosis, but the follow up interval and optimal surveillance strategy to identify progression are not well defined.

Vibration controlled transient elastography (FibroScan ^®, VCTE) is a well-validated, FDA-approved noninvasive tool to measure liver stiffness (LS), which estimates liver fibrosis stage. VCTE is reliable in detecting significant fibrosis with high sensitivity,(7) while LS < 8kPa can be used to exclude significant fibrosis with high negative predictive value. VCTE has also shown to be a reliable tool to monitor individuals with MASLD.(8, 9) According to current guidance, people with MASLD and LS ≥ 8 kPa should undergo further evaluation by a gastroenterologist or hepatologist(3). However, surveillance strategy for subjects who had a baseline LS < 8kPa is not clearly defined.(4) We aimed to develop a simple clinical decision tool to identify which people with low-risk MASLD can safely avoid a repeat VCTE within a year of their baseline test.

MATERIALS AND METHODS

Study population

Derivation cohort

We performed a retrospective analysis of a prospective cohort of adult subjects (age >18 years) with a clinical diagnosis of MASLD as per the diagnostic criteria,(10) who attended the hepatology clinic at Digestive Disease Associates (Baltimore, Maryland, USA) between January 2019 and January 2021.(11) All subjects underwent VCTE testing routinely at baseline and after 6–12 months.

We excluded subjects who had: 1) significant alcohol use (defined as more than 21 standard drinks per week in men and more than 14 standard drinks per week in women over a 2-year period), 2) fatty liver due to other pathologies, including chronic viral hepatitis, medication induced steatosis, autoimmune hepatitis, hemochromatosis, and Wilson disease, 3) receiving treatment for weight loss or for MASLD, 4) type 2 diabetes receiving treatments other than lifestyle recommendations or metformin; 5) missing BMI or lab results within 5 days of the repeat VCTE test.

Validation cohort

A prospective cohort of adult subjects with MASLD who were followed at the Hepatology clinic at the National Institutes of Health (NIH) Clinical Center between January 2016 and December 2022. Subjects routinely had follow-up VCTE every 6–12 months. Exclusion criteria were identical to the derivation cohort.

Transient elastography

LS was obtained in both cohorts using VCTE by a FibroScan^® model 502 V2 Touch (Echosens, Paris, France) using the medium (M) or extra-large (XL) probe as recommended by an automatic probe selection tool. All scans were performed by certified trained operators. Subjects were fasting for ≥ 4 hours prior to the test. At least 10 measurements with an interquartile range (IQR)/median of ≤ 30% were required for a valid test and the median measurement was used to represent LS. Liver fat content was estimated by the controlled attenuation parameter (CAP) module.

Data collection

Clinical parameters were extracted from the medical record and included age at baseline, gender, ethnicity, and diagnosis of diabetes mellitus. BMI, AST, ALT, and platelet count were collected at baseline and the time of follow-up VCTE test.

Outcome definition

Subjects with baseline LS < 8kPa were defined as low risk. The primary outcome was defined as LS ≥ 8kPa on the follow-up VCTE, meeting the guidance criterion for further workup.

Statistical analysis

Statistical analyses were performed using GraphPad Prism version 9.4.1 (GraphPad Prism Software Inc., San Diego, CA) and SPSS version 28.0 (IBM). Continuous variables are reported as median (interquartile range) or mean ± standard deviation (SD), while categorical variables are reported as proportion of subjects with/without a certain characteristic. Student’s t test or Mann-Whitney U test were used, as appropriate. Group comparisons of categorical variables were performed using Fisher’s exact test. Univariate analysis was followed by multivariate logistic regression with variables included using forward stepwise selection. P-value thresholds of 0.05 and 0.10 determined variable entry into and removal from the model, respectively. Multivariate models were compared using corrected Akaike information criterion (AICc). Test characteristics – sensitivity, specificity, negative predictive value (NPV), positive predictive value (PPV), accuracy, and likelihood ratio - were calculated. For a lower cut off we prioritized sensitivity/NPV and for upper cutoff we prioritized specificity/PPV. All statistical tests were 2-sided using an α of 0.05.

Ethics

Data collection from the derivation cohort was approved by the institutional review board (IRB) at Saint Agnes Healthcare (No. IORG0005451) and exempted from informed consent. The validation cohort subjects participate in a natural history study approved by the National Institutes of Health Institutional Review Board (clnicaltrials.gov NCT00001971) and provided a written informed consent.

RESULTS

Characteristics of the derivation cohort

Of 284 candidates, 206 subjects met inclusion criteria and were included in the derivation cohort (Supplementary Figure 1). 96 subjects with baseline LS < 8kPa, defined as low risk, were the focus of further evaluation. The baseline characteristics of the low-risk subjects are shown in Table 1. Compared to low-risk subjects in the derivation cohort, subjects with LS ≥ 8kPa (n = 110) at baseline had significantly higher AST, ALT and FIB-4 index (Supplementary Table 1).

Table 1.

Baseline characteristics of low-risk subjects in the derivation cohort

Subject characteristics (n = 96)	Baseline	Follow up
Age [years]	64 (50–71)
Sex, Male	39 (41%)
Race
Caucasian	69 (72%)
African American	18 (19%)
Other	9 (9%)
Diabetes mellitus	20 (21%)
BMI [kg/m2]	32.4 (27.4–36.0)	32.7 (27.5–36.4)
LS [kPa]	6 (5–6.9)	6.1 (5.1–7.9)
CAP [dB/m]	312 (267–342)	303 (217–302)
AST [U/L]	26 (20–33)	25 (19–34)
ALT [U/L]	31 (24–46)	30 (20–45)
Platelets [X104/μl]	268 (214–300)	252 (217–302)
FIB-4	1.1 (0.7–1.5)	1.2 (0.8–1.6)
Time from baseline to follow up VCTE [months]	10 (6–12)
LS ≥ 8kPa during follow up,	24 (25%)

Data presented as median (IQR) or n (%). BMI- Body Mass Index, LS- Liver stiffness, CAP- Controlled Attenuation Parameter, AST- Aspartate Aminotransferase, ALT- Alanine Aminotransferase, VCTE- Vibration Controlled Transient Elastography, FIB-4 - Fibrosis-4 Score

Factors associated with the primary outcome

At the follow-up visit after a median of 10 months, 24 subjects (25%) met the primary endpoint of LS ≥ 8kPa at follow-up. As expected, subjects who met the primary endpoint had higher baseline LS compared to those who did not (7 vs. 5.6kPa, p < 0.01). Baseline AST as well as BMI and ALT at follow up were also significantly higher in subjects with follow-up LS ≥ 8kPa, with a trend for higher follow-up AST (Table 2).

Table 2.

Factors associated with LS ≥ 8 kPa on follow-up

Variables	Follow up LS ≥ 8kPa (N = 24)	Follow up LS < 8kPa (N = 72)	P value
Age [years]	59 (51–70)	65 (50–72)	0.52
Diabetes mellitus	5 (21%)	15 (21%)	0.99
Baseline
BMI [kg/m2]	34.3 (30.6–37.2)	31.9 (27.1–35.3)	0.09
LS [kPa]	7 (6–7.5)	5.6 (4.6–6.4)	< 0.01
CAP [dB/m]	319 (273–354)	310 (281–339)	0.76
AST [U/L]	30 (23–37)	24 (19–33)	0.03
ALT [U/L]	32 (26–51)	30 (22–44)	0.27
FIB-4	1.16 (0.87–1.7)	1.04 (0.68–1.45)	0.19
Follow up
BMI [kg/m2]	35.7 (31.1–37.6)	32.1 (26.8–35.2)	0.02
AST [U/L]	33 (20–42)	24 (18–33)	0.051
ALT [U/L]	41 (23–59)	28 (20–40)	0.04
FIB-4	1.19 (0.82–1.99)	1.14 (0.74–1.51)	0.25
Change from baseline
BMI change [kg/m2]	0.59 ± 1.83	−0.11 ± 1.4	0.19
Percent weight change [%]	2.08 ± 5.3	−0.19 ± 4.1	0.12
ALT change [U/L]	16.8 ± 45.0	−6.1 ± 20.8	0.09
AST change [U/L]	6.9 ± 25	−2.2 ± 15.2	0.26

Data presented as median (IQR) mean ± standard deviation or n (%).

BMI- Body Mass Index, LS- Liver stiffness, CAP- Controlled Attenuation Parameter, AST- Aspartate Aminotransferase, ALT- Alanine Aminotransferase, FIB-4 - Fibrosis-4 Score

On multivariate logistic regression, baseline LS and follow-up ALT were significant predictors of achieving the primary endpoint (Model 1, Table 3). Although the absolute value of ALT at the follow-up visit was identified as a significant predictor, given the relatively small cohort we were concerned that it may reflect overfitting of the model to our specific dataset, while our aim was to obtain a generalizable model. In contrast, the change in ALT from baseline is plausibly a reflection of the change in liver disease activity over time and may be more generalizable. On multivariate logistic regression with baseline LS and change in ALT (Model 2, Table 3), both variables were significant predictors of a follow-up LS ≥ 8kPa, and the model was numerically superior to Model 1 in AUROC (0.84 vs. 0.82) and AIC (85.2 vs. 88.9, p = 0.16) 1.

Table 3.

Multivariate logistic regression to predict follow up LS ≥ 8kPa

Predictive model	Odds ratio	P value	AUROC
Model 1			0.82 (0.72–0.92)
Baseline LS	2.56 (1.53–4.75)	< 0.01
Follow up ALT	1.03 (1.01–1.06)	0.03
Model 2			0.84 (0.74–0.94)
Baseline LS	2.82 (1.66–5.38)	< 0.001
ALT change	1.04 (1.01–1.08)	0.02

^*Binary logistic regression using forward stepwise selection with follow-up LS ≥ 8 kPa as the dependent variable.

LS- Liver stiffness, ALT- Alanine Aminotransferase

Prediction model development

To transform the multivariate regression into a practical decision tool to identify subjects who can avoid a repeat LS in a year, we utilized a two-step approach. Given the strong influence (as expected) of baseline LS we utilized it as the first step, with ALT change added where baseline LS was not sufficiently predictive.

The optimal lower cutoff for baseline LS was 5.4kPa, demonstrating the highest NPV (97%) and 95% sensitivity (Supplementary Figure 2). Subjects with baseline LS below this cutoff were extremely unlikely to have LS ≥ 8kPa within a year. Similarly, we found 6.8kPa as the optimal upper LS cutoff with peak accuracy and high specificity and PPV (Supplementary Figure 2), suggesting subjects with baseline LS above this threshold should be reevaluated with VCTE within a year. To optimize the test performance for intermediate LS values (5.5–6.8kPa), ALT change was added to the model. An ALT change > 6U/L optimized the model with sensitivity of 79%, specificity of 77.8%, NPV 91.8% and accuracy of 78% (Supplementary Figure 3, Table 4). The resultant decision algorithm is presented in Figure 1.

Table 4 -.

Performance of the decision tool

Test characteristics	Value (95% CI)
	Derivation cohort	Validation cohort
Sensitivity, %	79.2 (59.5 – 90.8)	63.6 (35.4 – 84.8)
Specificity, %	77.8 (66.9 – 85.8)	71.7 (58.4 – 82.0)
PPV, %	54.3 (38.2 – 69.5)	31.8 (16.4 – 52.7)
NPV, %	91.8 (82.2 – 96.5)	90.5 (77.9 – 96.2)
Accuracy, %	78	71
Likelihood ratio	3.6	2.3

Wilson-Brown method was used to calculate the confidence interval of test characteristics.

PPV- Positive Predictive Value, NPV- Negative Predictive Value

Figure 1:

The Decision algorithm

Weight loss was shown to improve steatosis and fibrosis in NASH subjects,(12) and weight gain is associated with MASLD progression.(6) As change in weight is clinically important, we tested whether including weight change from baseline to follow up in our decision tool would improve accuracy but found no impact on the test characteristics (results not shown).

Validation of the decision algorithm

We validated the decision algorithm in an independent cohort of 64 subjects with MASLD and baseline LS < 8kPa (Supplementary Table 2). On repeat VCTE after a median interval of 12 months, 11 subjects (17%) had LS ≥ 8kPa. Applying the decision tool in the validation cohort yielded a NPV of 90.48%, sensitivity of 63.6% and specificity of 71.7% (Table 4). Applying the two-step decision algorithm to the two cohorts combined would have saved the need to repeat a VCTE measurement after 1 year in 94 (75%) subjects; only 9 (25%) subjects meeting the primary endpoint and requiring further workup were misclassified.

DISCUSSION

Nonalcoholic fatty liver disease is interlinked with metabolic risk factors and with the rising obesity epidemic, MASLD prevalence is expected to rise and affect about half of the adult population by 2040.(13) In the USA, the majority of people with MASLD are without significant fibrosis or estimated have low-risk MASLD by non-invasive tests.(14) Guidelines suggest additional work-up for people estimated to be at risk, with a threshold of 8kPa on VCTE, but there is no evidence to guide the appropriate follow up those below the threshold, deemed to be at low-risk.(4) Importantly, as those who have MASLD and fibrosis likely started as low-risk individuals, a rational follow-up strategy is needed to monitor these individuals, lest they develop more progressive disease. In the current study we developed and validated a simple decision tool that predicts when a repeat VCTE in 6–12 months can be safely avoided in subjects who had a baseline low-risk test.

Our primary focus was to develop a simple tool based on readily available parameters. The chances of having a follow up LS ≥ 8kPa during follow up are affected by two essential components – the risk at baseline and change in disease activity over time, both reflected in our tool. Of baseline risk variables, the baseline LS was found to be the best predictor of having LS ≥ 8kPa during follow-up. This could be a reflection of inherent test variability and regression towards the mean, where individuals closer to the threshold at baseline are more likely to cross it. However, as MASLD is slow to progress(15) it is also plausible that individuals with higher baseline LS are already be on the path for progression. Change in disease activity over time is reflected in our model by the change in ALT from baseline. Although the absolute level of ALT is a poor measure of disease activity in MASLD,(16) a decrease in ALT is associated with response to treatment(17) and conversely, an increase in ALT level can be an independent predictor of MASLD.(18) In our final model, the small change in ALT of >6 U/L should not be interpreted as reflection of worsening disease activity per se, but rather as a way to rule out a decrease in activity that would make a rise in LS less likely. We show that the simple combination of baseline risk assessed by LS and the effects of time assessed by change in ALT is sufficient to rule out a need for repeat VCTE.

Several known factors associated with NASH progression were not significant contributors to our model, including FIB-4 and diabetes. FIB-4 index at baseline was not associated with the primary endpoint, likely because it assesses the same underlying pathology (baseline severity) as the baseline LS, and as such is redundant. FIB-4 at follow-up was not markedly changed from baseline FIB-4, obtained 6–12 months earlier, and did not contribute to the model either, likely because the change in ALT is a better marker of disease activity during that period. Of note, FIB-4 is currently recommended as the initial screening test to determine referral for VCTE(3), although it is mainly calibrated to improve specificity of refered cases and has limited sensitivity. However, FIB-4 may not be effective as a tool to longitudinally monitor disease progression as it may not be sensitive enough to detect changes in relatively early stages of disease.(19, 20) For example, in a population cohort, Graupera et al., found that about 40% of individuals with LS ≥ 8kPa had normal FIB-4 index.(21) Importantly, our study focused on people with MASLD who already had one VCTE baseline test and as such, cannot be used to interpret the usability of FIB-4 as a referral test prior to VCTE.

Type 2 diabetes mellitus (T2DM) is an important risk factor for MASLD and independently predicts disease progression.(22) In our study, the presence of diabetes was not associated with the primary endpoint. This could be because we excluded subjects with diabetes who were treated with agents such as semaglutide, SGLT2 inhibitors, or insulin, which could alter the natural history of the disease.(22, 23) Alternatively, it is possible that our sample size did not provide us with sufficient power to detect an additional independent contribution of diabetes.

Weight change is a clinically relevant factor that predicts MASLD disease process, with weight loss providing benefit and weight gain associated with worsening disease.(6, 12, 24) However, inclusion of weight change in our tool did not improve its predictive characteristics. A likely explanation is that, although weight gain may be a driving factor of liver injury in MASLD, the actual effects on the liver are better captured by change in ALT, especially since it was previously shown that an increase in BMI is associated with increase in ALT.(25)

During a median follow up of 10–12 months, 24 (25%) subjects in the derivation cohort and 11 (17%) subjects in the validation cohort had LS ≥ 8kPa during follow up. In a study from Hong Kong, serial LS measurement in individuals with diabetes showed that about 4% of people with baseline LS < 10kPa had a follow up LS ≥ 10kPa during a median follow up of 3.5 years. Similar to our study, change in ALT was associated with LS rise to ≥ 10kPa on follow up.(26) The lower percentage of LS increase over time in that cohort compared to our study could be due to the higher LS cutoff used in their study, as well as the fact that the Hong Kong cohort also included subjects with diabetes but without MASLD. In a retrospective study from Sweden, 30% of MASLD subjects had disease progression during a mean follow up of 12.6 years. Follow-up ALT was associated with increase in liver stiffness from baseline in this study, similar to our study. The lower rate of disease progression in Swedish cohort is likely due to difference in criteria defining progression.(27)

An increase in LS over time does not necessarily imply disease progression and could also reflect the inherent variability in LS measurements, especially in subjects with a higher baseline. Furthermore, a threshold of 8kPa does not identify people with advanced fibrosis. Regardless of whether an increase in apparent LS reflects disease progression or measurement variability, individuals surpassing the threshold during follow-up will necessitate referral and further evaluation as per current guidelines and thus, understanding the predictors of crossing the threshold are important by themselves, irrespective of underlying cause.

The primary aim of our study was to minimize unnecessary repetition of VCTE in low-risk individuals, since unnecessary testing can cause patient harm, discomfort, and increased health care costs.(28) Given the rising burden of MASLD, judicious use of available noninvasive diagnostic tools including VCTE is necessary to avoid unnecessary health care expenditure. Indeed, applying the tool could have avoided a repeat scan in 75% of subjects in our combined cohorts with high accuracy. Other major advantages of our model are its easy applicability in daily practice. An additional potential implication of the tool is that individuals who are identified as able to avoid follow-up VCTE in a year could be transitioned to the care of primary care physicians. Subsequently, these subjects could be periodically reassessed according to their evolving clinical trajectory. However, this implication requires validation. Importantly, our primary focus was on optimizing NPV to ensure reliable identification of individuals who can safely avoid follow test. This focus results in low PPV and the model should not be interpreted to be useful as a rule-in tool to suggest follow-up. Of note, a similar strategy of prioritization of rule-out approach and acceptance of low PPV has been used in selecting threshold values for other NITs as well.(27)

A major strength of the study is the use of two cohorts in which all subjects had a repeat VCTE routinely and prospectively, irrespective of clinical findings, decreasing selection bias. An additional strength is the ability to validate the algorithm in an independent cohort with a different demographic background, suggesting generalizability of the tool.

There are several inherent limitations to our study. Our derivation and validation cohorts were both seen in specialty hepatology clinics, hence there is a possibility of selection and referral bias, and the decision tool may need to be validated separately in primary-care setting. Both cohorts were relatively small, limiting our ability to detect additional contributing factors that may fine tune the model. We excluded people with diabetes receiving treatment other than metformin, so caution should be exercised when applying the decision tool to that population. We cannot confirm that meeting the primary endpoint in our study truly meant progression of disease, but that would require a paired biopsy study design which is beyond the scope of this work. Finally, our study was limited to 6–12 month follow up and cannot be extrapolated to longer durations.

In conclusion, we propose a simple decision tool combining baseline LS and change in ALT over time that can be used predict which people with low-risk MASLD can safely avoid a repeat VCTE in 6–12 months.

List of abbreviations:

AICc

Akaike information criterion

ALT

Alanine Aminotransferase

AST

Aspartate Aminotransferase

BMI

Body mass index

CAP

Controlled attenuation parameter

FIB-4

Fibrosis-4 Score

IQR

Interquartile range

IRB

Institutional review board

LS

Liver stiffness

MASLD

Metabolic dysfunction associated steatotic liver disease

NIH

National Institutes of Health

NPV

Negative predictive value

PPV

Positive predictive value

VCTE

Vibration controlled transient elastography

Data availability statement:

Data in aggregate available upon request from the corresponding author.