Diagnostic accuracy of shear-wave elastography as a non-invasive biomarker of high-risk non-alcoholic steatohepatitis (NASH) in patients with non-alcoholic fatty liver disease (NAFLD)

Abstract

In this study, we evaluated the diagnostic accuracy of shear-wave elastography (SWE) for differentiating high risk non-alcoholic steatohepatitis (hrNASH) from nonalcoholic fatty liver (NAFL) and low risk NASH. Patients with non-alcoholic fatty liver disease (NAFLD) scheduled for liver biopsy underwent pre-biopsy SWE. Ten SWE measurements were obtained. Biopsy samples were reviewed using the NASH-CRN scoring system and patients with hrNASH were identified. Receiver operating characteristic curves (ROC) for SWE-based hrNASH diagnosis were charted. One hundred sixteen adult patients underwent liver biopsy at our institution for the evaluation of NAFLD. The area under the ROC curve of SWE for hrNASH diagnosis was 0.73 (95%CI, 0.61–0.84, p<0.001). The Youden index-based optimal stiffness cut-off value for hrNASH diagnosis was calculated as 8.4 kPA (1.67 m/s), with sensitivity of 77% and specificity of 66%. SWE may be useful for the detection of NASH patients at risk of long-term liver specific morbidity and mortality.

INTRODUCTION

Nonalcoholic fatty liver disease (NAFLD) is a chronic disease with a global prevalence of 25% (Younossi et al. 2016). In the United States, NAFLD and nonalcoholic steatohepatitis (NASH) afflict ~30% and ~5% of the population, respectively (Rinella 2015). NAFLD has an even higher prevalence among obese and diabetic individuals, estimated to be 70% and 90%, respectively (Hannah and Harrison 2016). By 2020, NAFLD is expected to be the major indication for liver transplantation (Charlton 2008).

NAFLD is hepatic fat infiltration in >5% of the liver volume or weight, as evaluated by liver biopsy histopathology, in the absence of alcohol intake exceeding 30g/day for men and 20g/day for women (Abd El-Kader and El-Den Ashmawy 2015; Watanabe et al. 2015). The NAFLD spectrum ranges from intracellular fat accumulation (steatosis) to varying degrees of necrosis, inflammation, and fibrosis, collectively termed nonalcoholic steatohepatitis (NASH). The NAFLD spectrum can be simplified into two conceptual categories: simple steatosis (nonalcoholic fatty liver) (NAFL), as excess liver fat accumulation without inflammation, and nonalcoholic steatohepatitis (NASH), as excess liver fat accumulation with inflammation and tissue injury (Chalasani et al. 2012). In an important subset of NASH patients, ongoing liver injury culminates in liver cirrhosis, with the potential for life-threatening complications, including variceal bleeding, hepatic failure, and hepatocellular carcinoma. NASH is also associated with a ~300% increase in cardiovascular disease mortality (Athyros et al. 2015) and is presently the second major cause of liver transplantation in the United States (Wong et al. 2015).

NAFL/NASH risk factors include dyslipidemia, diabetes mellitus, obesity, and the metabolic syndrome (Watanabe et al. 2015). While NASH pathogenesis remains incompletely understood, numerous therapeutic agents are presently under development for NASH treatment and prevention of its complications. Consequently, there is a pressing need for identification of low-cost non-invasive biomarkers to facilitate better understanding of NAFLD biology, optimal patient selection for clinical decision-making, and NASH drug development.

At present, non-focal liver biopsy is the accepted NASH diagnosis and risk stratification reference standard. However, biopsy is an expensive invasive method which is associated with morbidity and uncommon mortality. Variability and sampling error issues have also been reported (Nalbantoglu and Brunt 2014). In this common disease which affects ~30% of the population, liver biopsy is not practical for diagnosis and risk stratification. As a result, imperfect surrogate markers of NASH are commonly used in clinical practice. Currently available diagnostic tests include serum biomarkers and imaging (Kinner et al. 2016; Vilar-Gomez and Chalasani 2017).

Changes in liver tissue stiffness are known to occur with liver fibrosis. Various noninvasive elastography techniques, including Transient Elastography (TE), Strain Elastography (SE), 2D-Shear wave elastography (2D-SWE), and point SWE are widely used in clinical practice for liver fibrosis staging (Ferraioli et al. 2015). 2D-SWE is an ultrasound-based imaging technique that non-invasively measures tissue stiffness by generating an acoustic radiation force push pulse and monitoring the propagation of the resultant shear waves at multiple points. The shear wave speed (SWS) (in m/s) or calculated Young’s modulus (in kPA) are displayed in 2D-SWE images providing real time biomarkers of liver fibrosis (Sigrist et al. 2017). SWS values and Young’s modulus values can be algebraically interchanged and higher SWS values or higher Young’s modulus values correspond to higher stages of liver fibrosis (Ferraioli et al. 2015). The reported sensitivity and specificity for clinically significant liver fibrosis of various sonoelastography techniques range from 60% to 96% and from 18% to 100%, respectively (Samir et al. 2015; Tsochatzis et al. 2011).

Fibrosis stage at the time of diagnosis is a key prognostic biomarker for need of liver transplantation and death in NASH (Angulo et al. 2015). Understanding of the prognostic importance of liver fibrosis in NASH has led to the emergence of the concept of high-risk NASH (hrNASH), denoting those patients at highest risk of long-term liver-specific adverse outcomes. Identification of these patients is centrally important both for clinical decision-making and for the efficient performance of research studies aimed at mitigating these adverse outcomes. In this study, our aim was to determine whether SWE may be useful for non-invasive hrNASH diagnosis.

MATERIALS AND METHODS

Study design and subjects

This retrospective single institution study was approved by the local institutional review board (IRB). The study was compliant with the Health Insurance Portability and Accountability Act (HIPAA). We performed SWE scans in diagnosed or suspected NAFLD patients scheduled for ultrasound guided non-focal liver biopsy as part of routine clinical care. We did not obtain informed consent, as the IRB waived this requirement. Non-focal liver biopsies were performed under ultrasound guidance. To reduce histopathology variability blinded single-reader histopathologic assessment of liver fibrosis staging was performed using the fibrosis scoring component of the system proposed by the NASH Clinical Research Network (NASH CRN) (Kleiner et al. 2005). We collected demographic data, clinical history data, and blood biomarkers including liver enzymes.

Shear Wave Elastography (SWE) Acquisitions

SWE scanning was performed using an Aixplorer ultrasound system (SuperSonic Imagine S.A., Aix-en-Provence, France) with a convex abdominal probe (SC6–1). The quantitative unit for stiffness (Young’s modulus) was calculated and reported in kilopascals (kPa). Values in m/s unit were computed afterward and not obtained from the device. Six sonographers with varying ultrasound experience and training in SWE image acquisition performed SWE scans immediately prior to the liver biopsy procedure. During shallow expiration, ten (10) sequential SWE measurements were obtained via an intercostal approach in the upper portion of the right lobe. SWE measurements were collected at least 2 cm below the liver capsule and at a depth of less than 6 cm below the skin surface. Sonographers placed a circle-shaped 1cm region of interest (ROI) in the liver tissue, avoiding liver anatomic structures like large blood vessels. The median value of the 10 SWE acquisitions was used for subsequent analysis (Dietrich et al. 2017)

SWE image quality assurance:

A reviewer (AO) with 2 years’ experience in liver elastography techniques assessed all collected SWE images. The reviewer was blinded to the biopsy results and clinical history. A SWE acquisition was accepted as poor quality if: (1) a minimum of 10 measurements was not made, (2) a Region of Interest was located in a non-standard manner, such as in the area immediately below the liver capsule (which causes image artifact in SWE) or (3) images had poor elastogram filling pattern. An elastogram image was classified as poor quality if the reviewer estimated more than half of the color box was not filled in, which indicates poor shear wave propagation.

Histological Examination

To minimize variability, a single subspecialist board certified pathologist with more than thirty years of liver pathology experience and substantial experience with consensus-based liver fibrosis staging in multicenter trials reviewed the biopsy specimens while blinded to clinical history and SWE results.

Liver fibrosis was scored using the NASH CRN staging system’s fibrosis evaluation component (Kleiner et al. 2005). NASH CRN staging was performed for fibrosis staging (F Stage, F0–4) by the same pathologist using the following criteria: (0): No fibrosis; (1): 1a, mild, zone 3 perisinusoidal fibrosis, (1b), moderate, zone 3 perisinusoidal fibrosis, (1c), portal/periportal fibrosis only; (2): Zone 3 perisinusoidal and portal/periportal fibrosis; (3): Bridging fibrosis; (4): Cirrhosis. NAFLD Activity Score (NAS) was evaluated using sum of steatosis, lobular inflammation and hepatocellular ballooning (NAS Score 0–8). Steatosis was scored using the following criteria: (0), <5%; (1): 5–33%, (2):33–66%, (3):>66% to amount of surface area. Lobular inflammation was scored using the following criteria: (0) No foci; (1) <2 foci; (2) 2–4 foci; (3) >4 foci. Hepatocellular ballooning was scored using the following criteria: (0) None; (1) Few balloon cells, (2) Many ballooning cells (Kleiner et al. 2005).

Statistical Analysis

Statistical analysis was performed with RStudio (Version 0.98.1103). Multivariable logistic regression was performed to assess the diagnostic utility of SWE for the prediction of clinically significant liver fibrosis (≥F2), ≥F3 stage fibrosis, and high NAS (≥5) after controlling for key covariates known to correlate with liver disease. For covariates that were expected to be colinear (i.e. BMI and weight or serum glucose and hemoglobin A1C), only one variable was included in the model. P values for the covariates in each model are not reported due to insufficient sample size to adequately account for multiple comparisons. In our models, we included the components of most commonly used noninvasive biomarkers and models like APRI score, BARD score, and FIB-4 index (Vilar-Gomez and Chalasani 2017). Considering the well known accuracy and performance of these biomarker models, in our multivariate models we included; age, BMI, platelets, albumin, ALT, AST, diabetes diagnosis, and median shear wave elastography results.

Receiver operating characteristic (ROC) curve was plotted to compute the diagnostic accuracy of median SWE value to predict hrNASH patients, those patients with F2 or greater fibrosis. We also calculated ROC values to diagnose hrNASH patients with F stage ≥3. An additional ROC curve was plotted to assess diagnostic accuracy of SWE for the prediction of NAS 5 or greater. 95% confidence intervals were computed for parameter estimates. All p-values were two sided. Youden index-based optimal stiffness cut-off in kPA, sensitivity, and specificity values were calculated. p values less than 0.05 were considered statistically significant.

RESULTS

Of 132 patients who underwent liver biopsy at our institution for the evaluation of NAFLD between January 2014 and March 2015, 16 patients were excluded owing to poor quality SWE. The remaining 116 patients were included in the study. Patient demographics including gender, race, diabetes, dyslipidemia, smoking history, hypertension, obstructive sleep apnea, and summary of subject age, BMI, and laboratory data is displayed in Table 1. 54 men and 62 women with a mean age of 50.6±11.8 years were included, with a race distribution of 88 White, 2 American Indian, 6 Asian, 9 Hispanic or Latino, and 1 Black. The race of the other 10 patients was unknown.

Table 1.

Patient Demographic and clinical characteristics

	Item	No. of Patient (n=116)	Percentage (%)	Mean Young’s Modulus (sd)	P values
Gender	M	54	46.6	8.7 (3.3)	0.611
	F	62	53.4	9.1 (4.5)
Race	White	88	75.9	9.0 (3.6)	0.811
	American Indian and Alaska Native	2	1.7	9.0 (3.5)
	Asian	6	5.2	7.9 (6.1)
	Hispanic or Latino	9	7.7	10.0 (4.2)
	Black or African American	1	0.9	12.7
	Unknown	10	8.6	8.1 (6.0)
Diabetes	Yes	38	32.8	9.4 (4.9)	0.351
	No	78	67.2	8.7 (3.5)
Smoker	Yes	11	9.5	7.4 (2.7)	0.194
	No	105	90.5	9.1 (4.1)
Hypertension	Yes	67	57.8	9.4 (4.2)	0.198
	No	49	42.2	8.4 (3.7)
Obstructive sleep apnea	Yes	20	17.2	10.6 (4.5)	0.71
	No	96	82.8	8.6 (3.8)

SWE values ranged from 1.8 to 24.45 kPa (~ 0.77 m/s to 2.85 m/s). On biopsy examination, the subjects had the following fibrosis stage distribution: 0 (n=48, 41.3%), 1a/1b/1c (n=38, 32.7%), 2 (n=11, 9.4%), 3 (n=16, 13.7%), and 4 (n=3, 2.5%).

After controlling for key clinical covariates (table 2 and 3) such as age, body mass index, liver function tests, and diabetes, a statistically significant association between SWE values and clinically significant liver fibrosis (≥F2) was observed (p = 0.011). When controlling for the same covariates, a statistically significant association between SWE values and ≥F3 stage fibrosis was observed (p=0.005), but no association between SWE and NAS (≥5) was demonstrated (p=0.19). The odds ratio estimates were 1.23 (95% CI, 1.06–1.47), 1.34 (95% CI, 1.12–1.70), and 1.08 (95% CI, 0.96–1.25) to diagnose patients with ≥F2, ≥F3 and NAS≥5, respectively.

Table 2. Characteristics of the patient population.

Wt- Weight, Ht-Height, BMI- Body Mass Index, PLT- Platelets, ALB- Albumin, D. Bil- Direct Bilirubin, I. Bil- Indirect bilirubin, T. Bil- Total Bilirubin, ALP- Alkaline Phosphatase, ALT- Alanine Aminotransferase, AST- Aspartate Aminotransferase, HbA1c- Hemoglobin A1c, HDL- High density lipoprotein, LDL-Low density lipoprotein

Cases	Included cases		Excluded cases
Variable	Mean	StDev	Mean	StDev
Age (years)	50.6	11.8	48.6	12.3
Wt (lbs)	195.8	38.0	198.7	37.6
Ht (inc)	66.1	3.9	66.1	4.0
BMI*	31.4	5.1	31.8	4.7
PLT (k/uL)	237.8	77.1	240.2	46.4
ALB (g/dL)	4.6	0.35	4.6	1.2
Globulin (g/dL)	2.7	0.5	2.7	0.8
D.Bil (mg/dL)	0.11	0.045	0.12	0.07
I.Bil (mg/dL)	0.43	0.24	0.33	0.21
T.Bil (mg/dL)	0.57	0.33	0.45	0.26
ALP (u/L)	92.1	43.2	87.6	48.6
ALT (u/L)	67.1	43.2	64.9	40.7
AST (u/L)	51.0	29.3	50.2	28.4
Fasting Glucose (mg/dL)	109.0	36.4	101.3	42.2
HbA1c (%)	6.2	1.1	6.5	3.0
Cholesterol (mg/dL)	195.0	39.9	209	63.1
Triglycerides (mg/dL)	168.3	76.8	191.4	90.8
HDL (mg/dL)	46.1	16.2	41.7	14.7
LDL (mg/dL)	116.8	37.4	128.9	41.4
Plasma creatinine (mg/dL)	0.85	0.23	0.85	0.19
Subcutaneous tissue (skin to capsule distance-SCD)(cm)	2.4	0.5	2.3	0.44

Table 3.

Comparison of F<2 vs F≥2, F< 3 vs F≥3, NAS<5 vs NAS≥5. Number of patients are presented for categorical variables. Mean values are presented for continuous variables. The variables that are included in multivariate model, are marked with *. P values reflect univariate comparison.

	F<2	F≥2		F< 3	F≥3		NAS<5	NAS≥5
	Number of patients or Mean (StDev)	Number of patients or Mean (StDev)	p-value	Number of patients or Mean (StDev))	Number of patients or Mean (StDev))	p-value	Number of patients or Mean (StDev)	Number of patients or Mean (StDev)	p-value
Gender
Male	41	13	0.68	45	9	0.93	44	10	0.59
Female	45	17		52	10		48	14
Race
White	65	23	0.55	74	14	0.96	71	17	0.79
American Indian and Alaska Native	2	0		2	0		1	1
Asian	4	2		5	1		5	1
Hispanic or Latino	7	2		7	2		6	3
Black or African American	0	1		1	0		1	0
Unknown	8	2		8	2		8	2
Diabetes*
Yes	24	14	0.06	27	11	0.01	26	12	0.04
No	62	16		70	8		66	12
Smoker
Yes	7	4	0.40	10	1	0.49	8	3	0.57
No	79	26		87	18		84	21
Hypertension
Yes	47	20	0.25	53	14	0.12	53	14	0.94
No	39	10		44	5		39	10
Obstructive sleep apnea
Yes	16	4	0.51	16	4	0.63	16	4	0.93
No	70	26		81	15		76	20
Age*	48.5(11.7)	56.3(10)	0.002	49.5(11.9)	56(9.5)	0.02	50.2(11.1)	51.8(14.1)	0.56
Weight	194(39.2)	200.7(34.2)	0.41	193.7(39.1)	206.4(30.1)	0.18	194.9(39.2)	199.2(33.6)	0.62
Height	66.1(4.04)	65.7(3.62)	0.66	66.1(4.07)	65.6(3.14)	0.65	66(3.96)	65.9(3.85)	0.87
BMI*	30.9(5.21)	32.4(4.56)	0.17	30.9(5.11)	33.5(4.3)	0.04	31.1(5.03)	32.1(5.23)	0.39
PLT (k/uL)*	249.9(78.2)	203(62.3)	0.04	246(76.7)	195.9(65.2)	0.009	242.7(82.4)	218.9(48)	0.17
ALB (g/dL)*	4.65(0.33)	4.55(0.38)	0.15	4.65(0.32)	4.51(0.43)	0.11	4.62(0.34)	4.67(0.34)	0.53
Globulin (g/dL)	2.73(0.48)	2.71(0.46)	0.87	2.73(0.48)	2.71(0.5)	0.86	2.72(0.49)	2.75(0.46)	0.80
D.Bil (mg/dL)	0.1(0.04)	0.12(0.04)	0.21	0.11(0.04)	0.11(0.03)	0.63	0.11(0.04)	0.11(0.04)	0.87
I.Bil (mg/dL)	0.41(0.24)	0.45(0.24)	0.45	0.42(0.25)	0.41(0.13)	0.81	0.42(0.22)	0.42(0.29)	0.95
T.Bil (mg/dL)	0.53(0.28)	0.64(0.41)	0.13	0.55(0.3)	0.62(0.44)	0.38	0.55(0.27)	0.61(0.48)	0.44
ALP (u/L)	87.8(35.8)	104.2(59.4)	0.07	88.9(35.3)	107.7(71.6)	0.08	92(46.8)	92.2(28)	0.98
ALT (u/L)*	62(37.7)	81.5(54.1)	0.004	65.8(42.3)	73.2(48.2)	0.5	63.4(42.6)	80.8(43.5)	0.08
AST (u/L)*	45.1(23)	67.6(38.1)	0.01	48(25.6)	65.8(41)	0.01	46.7(25.5)	67.3(36.7)	0.002
Fasting Glucose	105.2(30.7)	119.7(48.2)	0.06	104.5(29.3)	132(56.7)	0.002	105.3(32.2)	122.9(47.5)	0.03
HbA1c (%)	6.1(0.98)	6.49(1.27)	0.09	6.09(0.97)	6.76(1.38)	0.01	6.11(0.96)	6.55(1.38)	0.06
Cholesterol (mg/dL)	195.7(38.3)	192.5(44.6)	0.70	196.4(38.9)	187.5(45.1)	0.38	197.4(38.5)	185.2(44.3)	0.18
Triglycerides (mg/dL)	173.9(81.1)	152.1(61)	0.18	171.3(78.5)	152.4(66.7)	0.32	167.2(78.8)	172.3(69.8)	0.77
HDL (mg/dL)	46.5(17.4)	44.9(12.2)	0.64	46.7(16.5)	42.5(14)	0.29	46.8(17.4)	43.2(10)	0.34
LDL (mg/dL)	116.7(36.3)	117(40.9)	0.97	117.2(36.4)	114.5(43.1)	0.77	119.2(36.2)	107.5(41.1)	0.17
Plasma creatinine (mg/dL)	0.85(0.22)	0.82(0.24)	0.53	0.85(0.22)	0.82(0.26)	0.65	0.84(0.23)	0.86(0.18)	0.76
SCD (cm)	2.33(0.55)	2.45(0.42)	0.28	2.34(0.53)	2.48(0.43)	0.29	2.33(0.49)	2.51(0.61)	0.12

The area under the ROC curve (AUROC) to diagnose hrNASH patients (F≥2) (n=30) was 0.73 (95%CI, 0.61–0.84, p<0.001) (Figure 1). The Youden index-based optimal stiffness cut-off value for hrNASH diagnosis was calculated as 8.4 kPA (1.67 m/s), with sensitivity of 77% and specificity of 66%. The area under the ROC curve (AUROC) to diagnose hrNASH patients with F≥3 (n=19), was 0.82 (95%CI, 0.71–0.93, p<0.001) (Figure 1). The Youden index based optimal stiffness cut-off value for diagnosis of hrNASH patients with F≥3, was calculated as 9.3 kPA (1.76 m/s), with sensitivity of 84% and specificity of 70%. A SWE image from NAFL patient and an image from hrNASH patients are presented in Figure 2 for comparison. The area under the ROC curve (AUROC) to diagnose patients with NAS≥5, was 0.67 (95%CI, 0.55–0.79, p<0.001) (Figure 1). The Youden index based optimal stiffness cut-off value for diagnosis of patients with NAS≥5, was calculated as 8.5 kPA (1.68 m/s), with sensitivity of 71% and specificity of 64%.

Figure 1.

Receiver operator characteristic curves for shear wave elastography to detect patients with F≥2 fibrosis, F≥3 fibrosis, and NAS≥5. Area under the ROC curve (AUROC) to diagnose F≥2 was 0.73 , F≥3 was 0.82, and NAS≥5 was 0.67. Optimal cutoff values using Youden index were: F≥2 – 8.4 kPA (1.67 m/s), F≥3 – 9.3 kPA (1.76 m/s), and NAS≥5 – 8.5 kPA (1.68 m/s).

Figure 2.

SWE images of patients with NAFL (figure 2a) and hrNASH (figure 2b). In figure 2a, SWE shows a Young’s Modulus value of 5.2 kPA (~1.31 m/s) corresponding to NAFL without hrNASH. Corresponding NASH CRN biopsy results show no fibrosis: F=0, S=1. In figure 2b, SWE shows a Young’s Modulus value of 13.2 kPA (~2.09 m/s) corresponding to hrNASH. Corresponding NASH CRN results show clinically significant liver fibrosis: F=2, S=2.

DISCUSSION

Liver fibrosis stage at the time of diagnosis has been shown to be a key prognostic marker, with survival rates diminishing as fibrosis stage increases (Angulo et al. 2015). Even if liver biopsy is traditionally considered the reference method for fibrosis detection and staging, it is invasive and limited by sampling and interpretative variability (Nalbantoglu and Brunt 2014). Serum biomarkers have been proposed to diagnose NASH, however, these biomarkers can be confounded by patient related factors including age and comorbidities (McPherson et al. 2017). A number of technologies have been studied for liver fibrosis staging, including transient elastography, which is limited by a lack of anatomic imaging. Magnetic Resonance Elastography (MRE) has been shown to be useful for liver fibrosis staging but is relatively expensive and not widely available. As a result, there is an urgent unmet need for a technology to safely and cost-effectively detect and risk stratify NAFLD patients.(Li et al. 2018)

Although it is relatively inexpensive, the utility of conventional B mode ultrasound for NASH diagnosis is likely to be limited by its inability to reliably differentiate steatosis and fibrosis or detect small changes in liver fat content (Dasarathy et al. 2009; Ozturk et al. 2018; Shannon et al. 2011). In a retrospective study of 94 subjects, Zardi et al, attempted to diagnose NASH using a conventional B-mode ultrasound based scoring system (Zardi et al. 2011). This system distinguished NASH from steatosis with a sensitivity of 91.6% and specificity of 75% but did not specifically detect the minority of subjects with hrNASH. Iijima et al. used contrast enhanced ultrasound (CEUS) to diagnose NASH, claiming an overall accuracy 79.6% (Iijima et al. 2007). Their report requires further validation before CEUS is widely adopted for NASH diagnosis. The study did not specifically detect the minority of NASH subjects with hrNASH.

Fatty liver and NASH imaging with SWE have been studied in animal models (Kang et al. 2015; Lu et al. 2014; Ogawa et al. 2016). In rats, using Supersonic Aixplorer system, Kang et al showed that shear wave elastography is an efficient technique to differentiate NASH and cirrhosis from less severe NAFLD, with AUROC 0.96, sensitivity 86.7% and specificity 100% (Kang et al. 2015). In a study with 23 NASH patients and 3 controls, using Siemens ACUSON S2000 system, Osaki et al. noted an AUROC, sensitivity, and specificity of 94.2%, 100%, and 75%, respectively, at a shear wave speed cutoff value of 1.47 m/s (p=0.0092) (equivalent to approximately 6.5 kPa) to differentiate fibrosis stage ≥3 from stage 0–1 in patients with NASH (Osaki et al. 2010).

To mitigate measurement variability, we collected 10 measurements from each patient and used the median value for data analysis. This technique has been reported in previous studies and is used in routine clinical care (Zhao et al. 2014)(Dhyani et al. 2017)(Sande et al. 2017). The feasibility and reliability of fewer measurements have been suggested although additional study is required to further justify reduced measurement number (Sporea et al. 2013).

In our study, SWE had an AUC of 0.79 for hrNASH diagnosis, with a threshold value of 8.37 kPa (~1.67 m/s) yielding sensitivity of 90% and specificity of 65%. The difference in this threshold value with that reported by Osaki et al. may be accounted for by utilization of different ultrasound machines, different histologic outcome markers, or different patient samples. Garcovich et al assessed SWE (with Supersonic Aixplorer system) for NASH diagnosis using the Brunt scoring system (Brunt et al. 1999) in a biopsy proven pediatric population and reported AUC value of 0.96 to diagnose F≥2 stage fibrosis with cut-off value of 6.7kPA (Sensitivity 87%, Specificity 96%)(Garcovich et al. 2017). Although high AUC values have been reported, histopathologic features of pediatric NAFLD might be different from those of adults (Schwimmer et al. 2005), and Garcovich et al’s study results may not generalize to the adult population. Our study provides complementary data in a larger sample of adults.

It is also important to note that the threshold value to diagnose hrNASH (F≥2) appears to be higher at 8.37 kPA (~1.67 m/s) as compared to the threshold value of 7.29 kPA (~1.55 m/s) to detect ≥F2 fibrosis at the same institution in patients with a varied spectrum of liver diseases using the same ultrasound system (SuperSonic Imagine S.A., Aix-en-Provence, France ), same SC6–1 transducer, and same imaging protocol (Dhyani et al. 2017; Dhyani et al. 2018; Samir et al. 2015). The observed difference is presumed to be related to selection bias with a higher percentage of patients with liver inflammation in our patient cohort (Diehl and Day 2017).

The limitations of this study include its retrospective design. While the use of a single ultrasound machine allowed data homogeneity, the cutoff values will not necessarily generalize to other vendors. More studies are necessary to validate our results and generate appropriate cutoff values for different devices. Another limitation is the inherent subjectivity of histologic analysis of needle core biopsies. We addressed this by engaging a single highly experienced blinded reviewer who has extensive experience serving on pathology review panels. However, we did not perform double pathology reads as this was impractical given the size of our cohort and the available resources. SWE scans were performed by six sonographers. High number of sonographers might cause increased data variability, but improves generalizability that would otherwise be affected by operator dependence. Lastly, our cohort included relatively few subjects with very advanced liver fibrosis. Despite these limitations, our results indicate SWE may be a useful risk stratification tool in NAFLD patients.

SUMMARY

Detecting hrNASH is crucial to facilitate early diagnosis of patients at risk of long-term liver-specific adverse outcomes. Our results offer evidence that SWE may be a low-cost non-invasive tool suitable for widespread adoption for this purpose.