MRE for Prediction of Long-Term Progression and Outcome in Chronic Liver Disease: A Retrospective Study

Abstract

Introduction

While magnetic resonance elastography (MRE) has been well-established for detecting and staging liver fibrosis, its prognostic role in determining outcomes of chronic liver disease (CLD) is mostly unknown.

Methods

This retrospective study comprised of 1269 subjects who underwent MRE between 2007 and 2009 and followed-up until death or last known clinical encounter or end of study period. Charts were reviewed for cirrhosis development, decompensation, and transplant or death. The cohort was split into: baseline non-cirrhosis (group 1), compensated cirrhosis (group 2), and decompensated cirrhosis (group 3). Cox-regression analysis with age, sex, splenomegaly, CLD etiology, Child-Pugh Score (CPS), Fib-4 score, and MELD-adjusted Hazard ratios (HR) for every 1kPa increase in liver stiffness measurement (LSM) were used to assess the predictive performance of MRE on outcomes.

Results

Group 1 (n=821) had baseline median LSM of 2.8kPa and cirrhosis developed in 72 (8.8%) subjects with an overall rate of ~1% cirrhosis/year. Baseline LSM predicted the future cirrhosis with multivariable adjusted HR of 2.38 (p< 0.0001) (concordance, 0.84). In group 2 (n=277) with baseline median LSM of 5.7kPa, 83 (30%) subjects developed decompensation. Baseline LSM predicted the future decompensation in cirrhosis with Fib-4 and MELD-adjusted HR of 1.22 (p<0.0001) (concordance, 0.75). In group 3 (n=171) with median baseline LSM of 6.8kPa (5.2, 8.4), 113 (66%) subjects had either death or transplant. Baseline LSM predicted the future transplant or death with HR of 1.11 (p=0.013) (concordance 0.53) but not in CPS and MELD-adjusted models (p=0.08).

Conclusion

MRE-based LSM is independently predictive of development of future cirrhosis and decompensation, and has predictive value in future transplant/death in CLD patients.

Introduction

Chronic liver disease (CLD) and cirrhosis account for more than 1 million deaths worldwide and 42,838 deaths in the US alone ¹. CLD is the fourth leading cause of death in adults aged 45–65 years ². Currently, there are about 4.5 million people diagnosed with CLD in the US ³. In addition to the increased risk of mortality, patients with CLD also have a very low quality of life, and CLD ranks among the top 20 worst conditions as assessed by Activities of Daily Living (ADLs) ⁴. Furthermore, patients with CLD have higher disability related unemployment (30.5% vs. 6.6%) and higher health care related costs than the national average ($19,390 vs. $5,567 per year), making CLD and cirrhosis a burden at both individual and national levels ⁵. Early diagnosis and risk stratification of CLD has the potential to improve these morbidity and mortality outcomes in CLD.

Traditionally, the diagnosis of liver fibrosis was performed with liver biopsy; however, is invasive and also has several other limitations, including poor interobserver agreement, sampling error and patient compliance⁶. Consequently, there is a high demand for non-invasive tests for the evaluation of liver fibrosis. Among the available non-invasive techniques, studies have indicated that liver stiffness measurement (LSM) with magnetic resonance elastography (MRE) has high accuracy and reproducibility, making it most reliable currently availble technique for evaluation of liver fibrosis ^7,8.

MRE was introduced for clinical practice in 2007 and has been validated in a broad range of CLD ⁹. However, less is known about the prognostic value of LSM by MRE. Recent studies showed that a single LSM has promise in predicting the risk of future events in two different liver etiologies: nonalcoholic fatty liver disease (NAFLD) ¹⁰ and primary biliary cholangitis (PBC) ¹¹. The applicability of the LSM with MRE to other CLDs for prediction of progression of liver fibrosis to cirrhosis and outcome, particularly decompensation, is still unknown. The purpose of this study was to evaluate the prognostic ability of a single LSM with MRE by reviewing the long term (≥10 years) outcomes in subjects who underwent liver MRE between 2007 and 2009.

Methods

Subjects and definitions

This is an IRB approved, HIPAA compliant, single institution (Mayo Clinic), multi-center (Minnesota, Florida, Arizona), retrospective 10-year outcome study of CLD subjects who underwent MRE for evaluation of liver fibrosis between January 2007 and December 2009. The list of individuals who underwent MRE was generated from clinical database and verified with radiology reports.

Demographics, body mass index (BMI), and laboratory tests of the subject population were retrieved from the electronic medical record system. Age, gender, race, and ethnicity were recorded at the time of MRE. BMI and laboratory results (albumin, AST, ALT, platelet count, INR, total bilirubin, serum creatinine and sodium) from the closest date to the MRE study within a 6-month period were recorded. Fibrosis index-4 score (Fib-4), model for end-stage liver disease score (MELD) and Child-Pugh Score (CPS) were calculated using standard formulas, clinical markers (for CPS) and lab values within 24 hours of baseline MRE ^12,13. In addition, manual chart review for every subject was performed for etiology of CLD and for presence or absence of following: cirrhosis, gastroesophageal varices, hepatic decompensation, hepatocellular carcinoma (HCC), cholangiocarcinoma (CCA), and history of liver transplantation. Subjects without CLD or clinical suspicion of liver fibrosis were excluded (n=266).

Cirrhosis was confirmed by either 1) histologic evidence of stage 4 fibrosis at biopsy (present in 67% of the cohort) and/or 2) cirrhotic morphology and/or signs of portal hypertension (splenomegaly or portosystemic shunting) at imaging or endoscopic evidence of portal hypertension (esophageal varices, portal hypertensive gastropathy).

Hepatic decompensation was defined by evidence of at least one decompensation event: esophageal variceal bleeding, ascites, hepatic encephalopathy or jaundice. Esophageal variceal bleeding was confirmed by upper endoscopy and ascites (mild to massive, >100 ml) was confirmed by imaging. Jaundice and encephalopathy were recorded from clinical notes and supportive laboratory tests. Jaundice was not ascertained as initial decompensation in subjects with biliary causes of CLD (primary biliary cirrhosis, primary sclerosing cholangitis and overlap syndromes) and other causes of biliary obstruction (e.g., tumor, calculus) where it can occur without cirrhosis.

Baseline clinical status of each patient (non-cirrhosis, compensated cirrhosis, decompensated cirrhosis) was determined at the time of MRE from individual chart reviews per definitions above. After the exclusion of previously transplanted subjects at the time of MRE, the cohort was divided into 3 groups according to the baseline clinical status: baseline non-cirrhosis (Group 1), baseline non-decompensated cirrhosis (Group 2), and baseline decompensated cirrhosis (Group 3).

Magnetic Resonance Elastography (MRE)

Each subject underwent a standard clinical liver MRE study either with a 1.5T or 3.0 T MRI systems (Various Models, GE Medical Systems, Milwaukee, WI). The indications were suspected liver fibrosis in chronic liver disease or clinical suspicion of liver fibrosis in patients without known chronic liver disease. The technique has been previously described in detail ¹⁴. The standard liver MRE sequence- 2D GRE MRE was used in all the patients. The regions of interest (ROI) was drawn by experts in MRE to include as large liver parenchyma as possible in multiple liver slices. Two expert MRE readers (MY, SKV) verified all the ROIs drawn for this retrospective review, and the liver stiffness measurement (LSM) was calculated by drawn ROIs. MREs with technical failure were excluded, and mean LSM was calculated by averaging LSM in multiple slices. Mean LSM, presence/absence of splenomegaly (spleen >12cm), native or transplant liver were noted. Subjects who received baseline MRE on transplant livers were excluded.

Follow-up and outcomes

Subjects were followed-up from the time of first (baseline) MRE until death, last known encounter or the end of study period (April 2020). The electronic health record of all study participants were individually reviewed for the primary outcomes of interest in their respective groups: development of cirrhosis, decompensation, transplant or death; and secondary outcomes of interest: HCC and CCA development. Ambiguous and contradicting medical records were consulted to an experienced transplant hepatologist (A.M.A). To avoid misclassification due to delayed reporting, cirrhosis development within 30 days and decompensation development within 15 days of baseline MRE were not recorded as follow up events. These events were categorized as, respectively, baseline cirrhosis and baseline decompensated, The cut-offs (30 day cut-off for cirrhosis, 15 day cut-off for decompensation) were decided after extensive review of all borderline events to minimize the possibility of false positive events.

Statistical Analysis

Patient demographics, LSM, BMI, and laboratory values were summarized as median and quartiles for continuous variables and counts and percentages for categorical variables. All groups had the time zero (baseline) as the time of MRE. Cox regression utilizing a cause-specific hazard approach was used to assess the predictive performance of LSM on outcomes, and the hazard ratio (HR) was reported for every 1 kPa increase in LSM. The HRs were adjusted for age, sex, Fib-4 score, CPS, MELD score, splenomegaly and etiology of liver disease in their respective analysis. Fib-4, CPS and MELD scores were also independently analyzed for predictive performance. Cirrhosis and decompensation probabilities over time accounting for the competing risk of death were graphed as cumulative incidence for every 1 kPa increase in LSM. Transplantation or death were graphed using Kaplan-Meier curves. Median follow-up times were calculated using the reverse Kaplan-Meier method. As a secondary outcome of the study, de-novo development of HCC and CCA was assessed for predictive performance of LSM both in subgroups and overall. To assess statistical significance, the traditional type I error rate of 0.05 was used. The analysis was performed with R software package (version 3.4.2, Vienna, Austria).

Results

The database search for all liver MRE studies in patients with CLD between January 2007 and December 2009 yielded 1376 subjects. Of these, 58 (4.4%) MREs were unsuccessful due to iron deposition (n=32), technical errors (n=15), poor wave propagation (n=7), poor breath hold or motion artifacts (n=4). 49 subjects with liver transplantation were also excluded. The remaining 1269 formed the study cohort (Figure 1).

Figure 1:

Flowsheet of the cohort.

The study cohort had a median age of 55 years (interquartile range [IQR] 46, 64) and 619 (48.8%) were females. The median LSM of the entire cohort was 3.5 kPa (IQR, 2.6, 5.3). The etiologies for CLD were chronic viral hepatitis B and C (n=437), NAFLD (n=375), alcoholic liver disease (n=99), chronic biliary diseases (PBC and PSC) (n=111), autoimmune liver disease (n=62), hereditary hemochromatosis (n=37), and other CLDs (n=146). The details of the etiologies and baseline characteristics are outlined in Table 1.

Table 1:

Baseline patient characteristics at the time of MRE

	Study cohort (n=1269)	Group 1 Baseline non-cirrhosis (n=821)	Group 2 Baseline compensated cirrhosis (n=277)	Group 3 Baseline decompensated cirrhosis (n=171)
Age, median (IQR)	55 (47,64)	53 (43,61)	57 (51,67)	61 (52, 68)
Sex, female, n (%)	619 (48.8%)	424 (51.6%)	120 (43.3 %)	75 (43.9%)
Race
African American	42	20	15	7
A. Indian/Alaskan N.	8	4	2	2
Asian	45	38	5	2
Pacific Islander/N. Hawaii	2	1	0	1
White	1041	668	231	142
Other/mixed	32	24	5	3
Not Disclosed	99	66	19	14
Ethnicity
Hispanic/Latino	40	26	9	5
Non-Hispanic/Latino	921	604	198	119
Unknown/not disclosed	308	191	70	47
LSM, median (IQR)	3.5 (2.6,5.3)	2.8 (2.4, 3.5)	5.72 (4.5, 7.4)	6.8 (5.2, 8.4)
BMI, median (IQR)	28.8 (25.1,33.6)	28.6 (24.9, 33.4)	29.7 (25.6, 34.7)	28.7 (26.1,32.3)
Liver etiology
HBV	59	48	7	4
HCV	378	242	103	33
NAFLD	375	264	63	48
Alcohol	99	44	18	37
PSC	62	36	20	6
PBC	49	38	8	3
AIH	62	34	21	7
HH	37	30	6	1
Drug-induced	15	13	1	1
A1AD	11	4	4	3
Sarcoidosis	8	3	3	2
Wilson’s disease	3	2	1	0
Parenchymal etiology unknown	28	4	9	15
Secondary parenchymal/other	81	59	13	11
Splenomegaly, n (%)
Yes	362 (28.5%)	91 (11.1%)	146 (52.7%)	125 (73.1 %)
No	892	720	127	45
Splenectomy/no spleen	15	10	4	1
Albumin	4.1 (3.7,4.4)	4.3 (4.0,4.5)	3.95 (3.6,4.2)	3.4 (3.1, 3.9)
AST	45 (30,73)	43 (31,68)	62 (42, 96)	58 (39, 87)
ALT	50 (31,90)	59 (36,96)	59.5 (36, 105)	39 (26, 66)
Platelets	203 (145,263)	225 (179,278)	149 (100, 192)	109 (72, 173)
INR	1.06 (1.0,1.1)	1.0 (1.0,1.1)	1.1 (1.0, 1.2)	1.2 (1.1, 1.3)
Bilirubin total	0.7 (0.5,1.1)	0.6 (0.4,0.9)	0.8 (0.6, 1.3)	1.4 (0.8, 2.3)
Creatinine	0.9 (0.7,1.0)	0.8 (0.7, 1.0)	0.8 (0.7, 1.0)	0.9 (0.8, 1.2)
Sodium	140 (138,141)	140 (138, 141)	140 (138, 142)	139 (137, 141)
Fib-4 Score	1.8 (1.1,3.4)	1.39 (0.91,2.18)	3.28 (2.03, 5.75)	4.99 (2.72, 8.35)
MELD score	7 (6,10)	7 (6, 8)	8 (7, 11)	11 (8, 14.5)
Child-Pugh Score (CPS)	5 (5, 6)	5 (5, 5)	5 (5, 6)	8 (7, 9.5)
Baseline HCC	70	9	37	24
Baseline CCA	22	9	8	5

The cohort (n=1269) was divided into three groups: group 1- baseline non-cirrhotic (n=821), group 2- baseline non-decompensated cirrhosis (n=277), and group 3- baseline decompensated cirrhosis (n= 171).

Baseline non-cirrhotic group (Group 1)

Group 1 (n=821) had median age of 53 years (IQR, 43.0, 61.0), 424 (51.6%) females, and a baseline LSM of 2.8 kPa (IQR, 2.4, 3.5). All baseline characteristics are summarized in the Table 1. Over the course of median 7.8 years follow-up, 72 (8.8%) subjects developed cirrhosis at rate of ~ 1% per year. The cumulative probability of cirrhosis at 10 years for subjects with LSM <2 kPa, 2–3 kPa, 3–4 kPa, 4–5 kPa and >5.0 kPa were 0 %, 3.7%, 14.3%, 24.9% and 60.7%, respectively (Figure 2). Among the stratified groups, subjects with LSM > 5 kPa developed cirrhosis in shorter time, with 17% within 1 year.

Figure 2:

Probability of cirrhosis in baseline non-cirrhosis stratified by LSM over time in years

Baseline LSM predicted the future development of cirrhosis with a HR of 2.08 (95% CI 1.82–2.37, p<0.0001) for every 1 kPa increase, and a concordance of 0.82 (Table 2A). Among the tested variables age, MELD and Fib-4 models were predictive of the outcome independently (HR: 1.03, 95% CI 1.01–1.05, p=0.003; HR: 1.08, 95% 1.02–1.14, p=0.004; HR: 1.35, 95% CI 1.23–1.47, p<0.0001, respectively), however they did not contribute to the prediction in bivariate (p=0.05, p=0.15, p=0.64 respectively) (Supplementary table 1) as well as the multivariate LSM models (Table 2A),while LSM remained a significant predictor (p<0.0001).

Table 2:

Hazard ratio (HR), p-value and concordance of LSM and relevant variables for prediction of cirrhosis, decompensation and death or transplant in patients with CLD

Table 2-A: Cirrhosis prediction in non-cirrhotic CLD
Model	Variable	HR	p-value	concordance
LSM	LSM (per 1 kPa increase)	2.08 (1.82–2.37)	p<0.0001	0.82
Age	Age (per 1 year increase)	1.03 (1.01–1.05)	p=0.0027	0.60
Sex	Sex (M)	1.42 (0.89–2.25)	p=0.143	0.55
MELD	MELD (per 1 increase)	1.08 (1.02–1.14)	p=0.004	0.62
FIB-4	FIB-4 (per 1 increase)	1.35 (1.23–1.47)	p<0.0001	0.72
NAFLD etiology	NAFLD (vs. non-NAFLD)	0.41 (0.23–0.75)	p=0.004	0.58
LSM + age + Fib-4 + NAFLD etiology + MELD	LSM	2.38 (1.91–2.97)	p<0.0001	0.84
	Age	1.00 (0.97–1.02)	p=0.81
	Fib-4	1.17 (0.95–1.43)	p=0.14
	NAFLD	0.84 (0.38–1.82)	p=0.65
	MELD	1.04 (0.97–1.11)	p=0.24
Table 2-B: Decompensation prediction in compensated cirrhosis
Model	Variable	HR	p-value	concordance
LSM	LSM	1.26 (1.16–1.36)	p<0.0001	0.67
Age	Age	0.98 (0.96–0.99)	p=0.010	0.57
Sex	Sex (M)	1.19 (0.77–1.85)	p=0.43	0.53
MELD	MELD	1.10 (1.05–1.14)	p<0.0001	0.66
Splenomegaly	Splenomegaly (+)	3.37 (2.07–5.48)	p<0.0001	0.65
Fib-4	Fib-4	1.14 (1.09–1.19)	p<0.0001	0.67
Child Pugh Score (CPS)	CPS (per 1 increase)	1.78 (1.35–2.34)	P<0.0001	0.63
LSM + Fib4 + MELD	LSM	1.22 (1.11–1.35)	p<0.0001	0.75
	Fib4	1.11 (1.05–1.17)	p=0.0001
	MELD	1.09 (1.04–1.15)	p=0.0003
LSM + Fib4 + MELD + CPS + splenomegaly	LSM	1.16 (1.05–1.29)	P=0.004	0.77
	Fib4	1.10 (1.04–1.16)	P=0.001
	MELD	1.07 (1.01–1.14)	P=0.015
	CPS	1.25 (0.91–1.72)	P= 0.17
	Splenomegaly	2.29 (1.27–4.14)	P=0.006
Table 2-C: Transplantation/death in hepatic decompensation
Model	Variable	HR	p-value	concordance
LSM	LSM	1.11 (1.02–1.21)	p=0.013	0.58
Age	Age	1.01 (0.99–1.03)	p=0.18	0.51
MELD	MELD	1.12 (1.07–1.16)	p<0.0001	0.68
CPS	CPS	1.32 (1.16–1.50)	P<0.0001	0.63
LSM + MELD	LSM	1.09 (1.00–1.19)	p=0.06	0.69
	MELD	1.11 (1.07–1.16)	p<0.0001
LSM + CPS + MELD	LSM	1.07 (0.99–1.20)	P=0.08	0.69
	CPS	1.11 (0.94–1.27)	P=0.26
	MELD	1.09 (1.04–1.15)	P=0.0006

^*The HRs of splenomegaly and NAFLD are per presence of the variable (present vs. non-present) whereas the HRs of LSM, CPS, MELD and age are per 1 unit increments.

Among CLD etiologies, only NAFLD, was associated with a smaller rate of cirrhosis development compared to other major etiologies with adequate sample size (HCV, HBV, alcohol, NAFLD, PSC) with HR of 0.37 (95% CI 0.19–0.71), p=0.003, concordance = 0.65. After adjustment for age, Fib-4, NAFLD etiology and MELD in the multivariate model, LSM retained its predictive ability with a HR of 2.38 (95% CI 1.91–2.97, p<0.0001) per 1 kPa, and concordance of 0.84 (Table 2A). The covariates did not contribute to the predictive ability of LSM in the multivariate model.

Baseline non-decompensated cirrhosis group (Group 2)

Group 2 (n=277) had a median LSM of 5.72 kPa (IQR, 4.50–7.40) at baseline. Other baseline characteristics are summarized in Table 1. Over the course of 10.2 years of median follow-up, 83 (30%) subjects developed decompensation at 4.2% per year, reaching the highest in the first 3 years at 7% per year. Ascites was the most common (n=78) decompensation event followed by hepatic encephalopathy (n=54), jaundice (n=38) and variceal bleeding (n=21). Cumulative decompensation at 10 years for LSM <4 kPa, 4–5 kPa, 5–6 kPa, 6–7 kPa, 7–8 kPa, and >8 kPa were 13.4 %, 21.6%, 55.0%, 45.7%, 75.9% and 78.4%, respectively (Figure 3). Each stratified LSM curve was statistically different (p<0.01) from the rest but the immediately adjacent LSM (Figure 3).

Figure 3:

Probability of decompensation in baseline non-decompensated cirrhosis stratified by LSM over time in years

Baseline LSM predicted the future development of decompensation with an HR of 1.26 (95% CI 1.16–1.36, p<0.0001) for every 1 kPa increase, and a concordance of 0.67 (Table 2B). Fib-4 score (HR: 1.14, 95% CI 1.09–1.19, p<0.0001), MELD score (HR: 1.10, 95% CI 1.05–1.14, p<0.0001), CPS (HR:1.78, 95% CI 1.35–2.34, p <0.0001) and splenomegaly (HR: 3.37, 95% CI 2.07–5.48, p<0.0001), also had good predictive ability with similar concordance in both univariate and bivariate models (Supplementary table 2). After adjustment for MELD score and Fib-4 score, the LSM retained its predictive ability with HR of 1.22 (95% CI 1.11–1.35, p<0.0001), and model performance increased to a concordance of 0.75 (Table 2B). The addition of splenomegaly to the model contributed to the model (HR: 2.29, 95% CI 1.27–4.14, p=0.006) increasing the performance (0.77 concordance), while addition of CPS did not (p=0.17) (Table 2B).

Baseline decompensated cirrhosis group (Group 3)

Group 3 (n=171) had a median LSM of 6.8 kPa (IQR, 5.2–8.4) at baseline. Other baseline characteristics are summarized in Table 1. Over the course of 10.3 years of median follow-up, 28 (16.3%) subjects had liver transplant, 10 (5.64%) died after transplantation, and 75 (44.0%) subjects died without transplantation. The rate of transplant/death was 6.8% per year on average over 12 years. However, it was non-uniform throughout the years with the highest in the first year with 30% per year, followed by the second year (14% per year) and the third year (10% per year) (Figure 4). Only 45 subjects had an identifiable cause of death. Among these, 27 (60%) were liver–related deaths (liver failure, decompensation, infections, hepatic malignancies and other liver-related causes). Cumulative transplant/death for LSM <4 kPa, 4–5 kPa, 5–6 kPa, 6–7 kPa, 7–8 kPa, and >8 kPa at 10 years they were 63.1%, 56.2%, 77.9%, 87.8%, 70.2%, 86.3%, respectively (Figure 4).

Figure 4:

Probability of transplant/death in baseline decompensated cirrhosis stratified by LSM over time in years

Baseline LSM independently predicted the future transplant or death with an HR of 1.11 (95% CI 1.02–1.21, p=0.013) for every 1 kPa increase, and a concordance of 0.58. Among the variables tested MELD (HR: 1.12, 95% CI 1.07– 1.16, p<0.0001, concordance = 0.68) and CPS (1.32, 95% CI 1.16–1.50, p<0.0001, concordance = 0.63) were predictive of transplant or death independently. After adjustment for MELD, LSM did not reach the significance level of 0.05 (HR: 1.09, 95% CI 1.00–1.19, p=0.06, concordance= 0.69 ) however, MELD remained as a strong predictor (Table 2C). Addition of CPS score did not contribute the predictive ability of the model (concordance =0.69), while MELD was significantly predictive of the outcome (p=0.006)

HCC and CCA development

Overall, 39 subjects developed HCC in 8.8 years of median follow-up (n=1199). LSM (median 3.4 kPa) was predictive of future HCC development with HR of 1.46. (95% CI 1.33–1.61) per 1 kPa increase (p<0.0001) and concordance of 0.87. LSM remained predictive of HCC development even after adjustment for Fib-4 and MELD (HR: 1.24 per 1 kPa, 95% CI 1.08–1.43, p=0.003, concordance 0.88) (Supplementary table 3). Results of the group specific HCC outcomes are further discussed in Supplementary results. The number of CCA events (n=10) was not high enough for a reliable statistical analysis (Supplementary results).

Discussion

In this longitudinal retrospective study of 1269 subjects, a single baseline LSM by MRE was predictive of liver-related outcomes and death in CLD.

In CLD subjects without cirrhosis, LSM by MRE was able to predict the future development of cirrhosis with high concordance. This finding is novel in two ways: LSM is predictive of future cirrhosis in CLD on a long-term basis from multiple etiologies as shown separately in the previous studies ^10,11. Secondly, the risk stratification with different LSMs highlights a possible new role for MRE in the management of CLD subjects. For instance, individually tailored screening could be implemented for subjects with <3 kPa, 3–4 kPa, 4–5 kPa and >5kPa as the risk of cirrhosis in 5 years are 1.5%, 6.6%, 19% and 51%. Our results show that patients with different baseline stiffness have different risks of developing cirrhosis with exponential increased risk after 3.0 kPa suggesting this number a possible threshold for guideline and policymaking. However, LSM is not categorical but rather a continuous variable, with the overall risk of developing cirrhosis doubled for every 1 kPa increase on average in non-cirrhotic livers.

In the cohort with baseline cirrhosis, MRE was also predictive of future decompensation. At 5-year, the chance of developing decompensation increased dramatically from 10% to 53% based on the baseline LSM. However, 5-year period is an exceedingly long time and current guidelines recommend 6-months follow-ups for patients with cirrhosis. In that case, 1-year outcomes may be utilized as the risk increases exponentially (1.3%, 5.7%, 23.1% in 1 year for <5kPa, 5–7kPa, >7kPa respectively). The association of baseline LSM with a risk of decompensation sheds new light on the prognosis of cirrhosis. Compared to the first cohort, the risk in the second cohort per 1 kPa was smaller, implying that only 1 kPa difference may not be large enough to draw clinical conclusions. On a closer look at the Figure 3, a coupling of the consecutive lines is noticed. While <4 kPa and 4–5 kPa are not significantly different from each other, they are significantly different (p<0.001) from the 5–6 kPa and 6–7 kPa baseline LSM which again are coupled. This brings up the question whether once cirrhotic, a change of 2 kPa is better suited instead of 1 kPa change until >8 kPa. In this specific group, the risk of decompensation is 26% higher per 1 kPa increase on average, however it may be more appropriate to report 59% higher risk for every 2 kPa increase, as the progression in post-cirrhosis may not be as smooth as pre-cirrhosis.

In the decompensated cohort, for the analysis purposes, we grouped transplant and death together, as having transplant is shown to prolong death and by nature are competing events for the outcome. The LSM was predictive of transplant and/or death in univariate analysis but not in multivariate analyses. We acknowledge some limitations in this cohort. Group 3 was not as large as the other cohorts (n=171). Subjects were grouped together irrespective of the number of previous decompensation events. Although decompensated cirrhosis has been shown to increase mortality significantly, our cohort could not identify liver-related deaths vs. non-liver deaths since only 20% of the cause of deaths were reported in the database. Also, it is challenging to assume a direct relationship between liver stiffness and transplant / death, as the latter is more multifactorial than, for example, developing cirrhosis. Hence, it was not surprising that the concordance was lower, the p-value of LSM failed to cross the 0.05 significance level in adjusted models. In this group, our model was not better than easily calculated MELD, which has been verified for prognosis and determination of transplantation ¹⁵.

The varying degree of power, confidence interval and clinical implication in three different sub-group analyses show us that LSM by MRE is a sensitive and strong predictor of liver outcomes early in the disease before developing cirrhosis. As the liver fibrosis progress into cirrhosis and decompensation, the prognostic value of LSM diminishes.

This suggests that in a clinical setting, MRE is probably more suitable as a tool in screening and risk stratification of patients with CLD before developing cirrhosis. The use of MRE for decompensation prediction can be performed to determine prognosis if larger changes in LSM, eg. 2 kPa as demonstrated by our study, is utilized. The use of MRE in predicting transplant/death is limited and not better than easily accessible proven ways of scoring such as MELD score. That being said MELD score is a measure of liver function rather than liver stiffness and fibrosis, Thus, it can be hypothesized that MRE is likely not as predictive once impairment of liver function is noted, as opposed to elevated stiffness.

Our data add on to the knowledge of risk stratification using LSM, and reconstructs LSM utilization in outcome prediction. For instance, studies by Kim et al ¹⁶and Nakagomi et al ¹⁷showed that liver stiffness by transient elastography was associated with liver related outcomes in chronic HBV (p<0.05) and overall mortality in HCV (p<0.001), respectively. We extended this knowledge to MRE in a larger cohort of CLD with baseline specific outcomes and hazard ratios. A study by Trebicka et al ¹⁸ showed that the cut-off of 20 kPa in 2D-shear wave elastography and MELD≥10 together predicted decompensation and mortality in a large cohort (AUC=0.8). Our study refrained using such cut-off, as we have observed that risk increased gradually as LSM increased. With 1 kPa increment approach, we were able to show for the first time that smaller changes in LSM increase the risk exponentially in pre-cirrhosis, while significant increase in risk occurs with larger changes in LSM (>1 kPa) in post-cirrhosis. Asrani et al ¹⁹, also utilizing MRE for prediction, showed that LSM was preditive of decompensation with HR 1.42 per 1 kPa increase (p<0.01). The finding is not very different from our study (HR: 1.26 per 1 kPa increase, p<0.0001) however in the aforementioned study there were only 12 events in 2 years of median follow-up which limited further analysis. Our long (10 years) follow-up time in a larger cohort yielded 83 decompensation events allowing multivariate regression with age, MELD, CPS, and splenomegaly.

In summary, the results of this study demonstrate that MRE is useful in the prediction of long-term events in a large cohort with multiple etiologies of CLD. Liver stiffness measurement with MRE has a significant implication in the long term prognosis of patients with CLD. Prospective studies are needed to validate these findings.

Abbreviation list:

IRB

Institutional review board

HIPAA

Health Insurance Portability and Accountability Act

MRE

Magnetic Resonance Elastography

BMI

Body mass index

AST

Aspartate aminotransferase

ALT

Alanine aminotransferase

INR

International normalized ratio

LSM

Liver stiffness measurement

kPa

kilopascals

CLD

Chronic liver disease

IQR

Inter-quartile range

MELD

Model for end-stage liver disease

CPS

Child-Pugh score

HCC

Hepatocellular carcinoma

CCA

Cholangiocarcinoma