Liver stiffness predicts progression to liver‐related events in patients with chronic liver disease – A cohort study of 14 414 patients

Liver stiffness measurement (LSM) by vibration‐controlled transient elastography (VCTE) is a non‐invasive diagnostic biomarker of liver fibrosis. It is uncertain if LSM can predict risk for future liver‐related outcomes in large, heterogenous populations.


| INTRODUC TI ON
The common end-stage of all chronic liver diseases is cirrhosis.
5][6] However, liver biopsy is an invasive and expensive procedure with risk of sampling error, interobserver variability, complications, and it is not suitable as a main diagnostic tool in larger patient populations.To assess the risk of progression in patients with chronic liver disease, there is a need of reliable non-invasive methods.Liver stiffness measurement (LSM) by vibration-controlled transient elastography (VCTE) is a non-invasive ultrasound-based method to assess fibrosis stage. 7M by VCTE is currently used in clinical practice and has largely replaced liver biopsy in the day-to-day management of chronic liver diseases. 8[12][13][14][15][16][17] However, the prognostic ability of LSM by VCTE in large, heterogeneous populations is uncertain.Most previous studies have been hampered by small sample sizes, not allowing risk estimations across subgroups. 13,16,18,19Here, we evaluated the association between LSM by VCTE and liver-related outcomes in a large cohort of patients with a diverse set of chronic liver diseases.

| Study population
This was a multi-centre cohort study of patients with LSM performed between 2008 and 2020, with various liver diseases, and outcomes ascertained by cross-linkage to Swedish national health registers.By using a unique personal identity number, consisting of the date of birth and a 4-digit code, that is assigned to all residents in Sweden, linkage to Swedish national health registers can be performed 20 (supplementary methods-Data S1).All sites in Sweden who had procured a Fibroscan®-machine as of mid-2019 (n = 24) were invited to participate (Table S1).Measures of liver stiffness, controlled attenuation parameter (CAP), success rates, and dates of examination were extracted from the Fibroscan®-machines.All LSM without a date of birth or a personal identity number were excluded (Figure 1).Also, an examination with less than 10 valid measurements were Results: Included patients had a median age of 46 (interquartile range 34-57), median LSM of 5.9 kPa (4.6-8.0),59% were male, and the majority had hepatitis C (50.1%).
During a median follow-up of 5.9 (4.3-8.0)years, 402 patients (2.7%) developed cirrhosis with portal hypertension.In patients with an LSM ≥25 kPa, 28.7% developed cirrhosis with portal hypertension within 5 years of follow-up, while only .6% of patients with an LSM <10 kPa did.This translated to a HR of 48.3 (95% confidence interval = 37.6-62.0).VCTE had a high discriminative ability, with C-indices above .80for most liver diseases, including .82 for MASLD.Similar findings were seen for incident HCC.
Conclusions: Increased LSM by VCTE was associated with an increased risk of progression to both cirrhosis with portal hypertension, and to HCC, and had a high discriminative ability across different aetiologies of chronic liver diseases.These results support the use of VCTE to guide follow-up and treatment decisions.excluded (Figure 1).If several examinations had been performed at the same date, the LSM with the narrowest interquartile range (IQR) was used.If patients had serial LSM at different dates, only the first LSM was used.
Exclusion criteria were an unavailable or reused personal identity number, a diagnosis of congestive heart failure, a diagnosis related to portal hypertension or hepatocellular carcinoma (HCC) (at, or before, baseline), and death or emigration before baseline.We additionally excluded patients with unreliable LSM, defined as an IQR for kPa >30% of the median value if the median was ≥7.1 kPa. 21Study baseline was set to 90 days after the LSM to ensure that possible exclusion criteria were identified.Data on sex, age at baseline, aetiology of liver disease, presence of type 2 diabetes, cardiovascular disease, kidney failure and previous cancer except HCC were obtained from the national health registers.Aetiology of liver disease and comorbidity data were based on International Classification of Disease (ICD) -10 codes, linked through the personal identity number to the National Patient Register 22 and the Cancer Register 23 (Tables S4 and S5).

| Liver disease classification
The etiologies of liver disease were divided into the following groups: hepatitis C; hepatitis B; MASLD; autoimmune liver disease including autoimmune hepatitis, primary biliary cholangitis, or primary sclerosing cholangitis; ALD; and other rare diseases including hereditary hemochromatosis, alfa 1-antitrypsin deficiency, Wilson's disease, Budd-Chiari syndrome, toxic liver disease, unspecified liver disease, granulomatous hepatitis and Echinococcosis, or 'No recorded diagnosis'.In this study, one single diagnosis per patient was defined.Patients with multiple aetiologies were defined according to the following hierarchy: 1. Hepatitis C, 2. Hepatitis B, 3. ALD, 4. Autoimmune liver disease, 5. Other causes and 6.MASLD.Thus, a patient with both ALD and hepatitis C was defined as having hepatitis C. Patients without an ICD-code suggestive of any specific liver disease, but with an ICD-code for diabetes mellitus type 2, obesity, or a CAP ≥240 dB/m, were categorized as MASLD.A CAP <240 dB/m excluded MASLD, hence a patient with type 2 diabetes but a CAP below 240 dB/m was not placed in the MASLD group, and instead placed in the 'No recorded diagnosis' group. 24This group could potentially contain patients with burned-out MASLD-cirrhosis, as well as patients under evaluation of suspected liver disease.We chose this approach to achieve a higher sensitivity of the MASLD diagnosis.Furthermore, patients with an ICD-code suggestive of alcohol abuse were categorized as ALD if no ICD-code of a specific liver disease was present (Table S4).

| Outcomes
Register data were used to assess outcomes (Table S6).We evaluated the prognostic ability of LSM by VCTE in predicting progression to either cirrhosis with portal hypertension or HCC.Cirrhosis with portal hypertension was defined as a diagnosis of either portal hypertension, ascites, oesophageal varices (bleeding or non-bleeding) or hepatorenal syndrome, death caused by any of the above diagnoses, that is death by a portal hypertension event, or liver transplantation.
The first of these outcomes, if several were present, was chosen as the time of the outcome.HCC was investigated separately.To assess F I G U R E 1 Flow chart for inclusion and exclusion of patients.outcomes, the cohort was linked through the personal identity number to the National Patient Register, 22 the Cause of Death Register 25 and the Cancer Register. 23Outcome data from these registers was obtained until 2 July 2023.Outcomes were based on ICD-10 codes from register data, and LSM was not used to define outcomes.
Patients were censored if they died from another cause than cirrhosis with portal hypertension or HCC, or if they reached the end of follow-up without a recorded event or censoring criteria, or if they emigrated from Sweden.

| Statistical analysis
Cumulative incidences of outcomes at 2 and 5 years, and incidence rates per 1000 person-years, were calculated and stratified on different categories of LSM.Competing events (death from neither cirrhosis with portal hypertension nor HCC) were considered in the calculations of cumulative incidence using the Aalen-Johansen estimator.The consensus at the Baveno VII-meeting suggested the use of LSM values based on a 'rule of five' to grade severity of chronic liver disease, <10 kPa, 10-14.9kPa, 15-19.9kPa, 20-24.9kPa and ≥25 kPa. 26Our approach was to use these LSM cut-offs to assess the risk of incident cirrhosis with portal hypertension or HCC.
Cox regression models were performed for time to event analysis and unadjusted hazard ratios (HR) were calculated.To assess the discriminative ability of LSM by VCTE, we stratified the cohort on the underlying liver disease and calculated Harrell's C-index for the full cohort and by each aetiology. 27Statistical analyses were performed in Stata (StataCorp LLC, version IC/16.1).
A sensitivity analysis was performed to examine the role of treatment of hepatitis C. We included patient-time and events before the initiation of direct acting antivirals (DAA) in patients with hepatitis C, using the Swedish Prescribed Drug Register. 28Patients were censored at the time of initiation of any DAA therapy.A second sensitivity analysis was performed to examine sex-specific differences in progression to cirrhosis with portal hypertension or HCC by stratifying the cohort by sex and by examination of liver stiffness and sex as an interaction term.Also, a third sensitivity analysis was performed using cut-offs of LSM with a higher granularity below 15 kPa: <6 kPa, 6-7.9 kPa, 8-9.9 kPa, 10-12.4kPa and 12.5-14.9kPa. 9

| RE SULTS
A total of 16 departments participated in the study.Both tertiary (n = 8) and secondary care centres (n = 8) participated.Reasons for non-participation and number of participants per department are given in Tables S1 and S2.We identified 42 174 examinations by VCTE.A total of 14 414 unique patients met inclusion criteria (Figure 1).

| Cirrhosis with portal hypertension outcomes
The median follow-up time was 5.9  was associated with an increased rate of cirrhosis with portal hypertension (Table 2).For instance, patients with an LSM <10 kPa had an incidence rate of 1.3 (95% confidence interval [CI] = 1.1-1.6)events per 1000 person-years, while patients with an LSM ≥25 kPa, had an incidence rate of 66.3 (95% CI = 56.7-77.6)per 1000 person-years.
In the sensitivity analysis aimed at investigating the differences in risk based on intervals with higher granularity, an LSM <6 kPa had an incidence rate of 1.0 (95% CI = .7-1.3) per 1000 person-years.
In the sensitivity analysis among patients with hepatitis C, naïve to DAA treatment, the incidence rates of HCC per 1000 person-years was .7 (95% CI = .4-1.2) and 34.5 (95% CI = 18.6-64.2) in patients with an LSM <10 kPa and an LSM ≥25 kPa, respectively.This translated into a 53-fold (HR 53.2, 95% CI = 22.2-127.5)increased rate of HCC in patients with an LSM of ≥25 kPa (Table S13).In the sensitivity analysis comparing rates of HCC in women and men, the incidence rates per 1000 person-years was 1.5 (95% CI = 1.1-1.9) in women and 3.0 (95% CI = 2.6-3.5) in men (Tables S14 and S15).The C-index was .91 (95% CI = .86-.95) and .83(95% CI = .79-.87) in women and men, respectively.We found no evidence of interaction between sex and liver stiffness on the risk of progression to HCC (p = .13).In the sensitivity analysis aimed at investigating the potential difference in risk based on LSM-intervals with higher granularity, patients with an LSM <6 kPa had .4(95% CI = .2-.6) events of HCC per 1000 person-years.In patients with an LSM of 6-7.9 kPa, 8-9.9 kPa, 10-12.4,and 12.5-14.9kPa it was .9(95% CI = .6-1.4), 2.7 (95% CI = 1.8-4.1),3.3 (95% CI = 2.1-5.1), and 10.4 (95% CI = 7.0-15.3),respectively (Table S16).This translated to a 29-fold (HR 28.8, 95% CI = 15.5-53.3)increased rate of HCC in patients with an LSM between 12.5-14.9kPa compared to an LSM <6 kPa.The cumulative incidence at 2 and 5 years ranged between .08% (95% CI = .03-.2) and 1.9% (95% CI = .8-3.7), and .2%(95% CI = .1-.3) and 4.6% (95% CI = 2.7-7.2) in patients with an LSM <6 kPa and 12.5 to 14.9 kPa, respectively.When visualizing the incidence of progression to cirrhosis with portal hypertension per 1 kPa increase in liver stiffness, we found that the rate increased in an exponential manner by increasing liver stiffness.This is consistent with the findings in a study that investigated the ability of LSM by VCTE to predict outcomes in patients with PBC, even though they did not present data >30 kPa and had a composite outcome which also included all-cause mortality. 29However, a meta-analysis of 6368 patients found that the relative risk of hepatic decompensation seemed to plateau at an LSM >25 kPa, but they did not present data of an LSM >40 kPa. 11This suggests that the risk of outcomes secondary to increased pressure can be directly estimated by VCTE, and the risk of these outcomes may continue to increase with increasing pressure, here estimated by liver stiffness.The incidence of HCC increased with increased LSM to a liver stiffness of approximately 20-25 kPa, after which the risk plateaued.The risk of HCC seemed to mostly depend on progression to cirrhosis, and not a continuous increase in portal pressure, as estimated by liver stiffness.This is in line with a similar graphical visualization performed in the meta-analysis mentioned above, 11 even though it has been implied that portal hypertension is an independent risk factor of HCC. 30 When investigating if the risk of outcomes differed across aetiologies of liver disease, we found that for each kPa increase   concluded that LSM by VCTE was equally good as a liver biopsy in predicting a composite outcome of liver-related events and allcause mortality. 31This was also the conclusion in a study investigating the prognostic ability of LSM by VCTE in patients with MASLD, which showed a good prognostic ability with a C-index above .80,non-inferior to the prognostic ability of a liver biopsy. 32milar findings have been seen in patients with ALD, hepatitis C, and primary biliary cholangitis. 15,29,33 A British study of 3028 patients investigating the natural history of advanced chronic liver disease by LSM also presented absolute risks of decompensation events. 12Their 5-year cumulative incidence in patients with an LSM of <15 kPa, 15-25 kPa, and >25 kPa was 3.7%, 8.6%, and 19.0%, respectively.Our 5-year cumulative incidence ranged from 2.9% to 28.7% in patients with an LSM of 10-15 kPa to ≥25 kPa.In general, the cumulative incidence in patients with an LSM ≥15 kPa in our study was higher.This could partly be due to differences in the definition of outcomes.Another possible explanation is the differences in the case-mix of the populations, as our study had a large proportion of patients with hepatitis C (50.1%), which have been shown to have more liver-related complications than MASLD. 34Other studies 13,16,18,19 have also demonstrated a prognostic ability of LSM by VCTE.However, they had either studied small cohorts, 19 selected populations at tertiary centres, 13,16,18 or cohorts with a higher comorbidity burden. 18me previous research suggests a sex difference in fibrogenesis 35 ; however, we found no differences in the prognostic capacity of LSM between men and women.Even if the risk of fibrosis progression might be higher in men, the same prognostic information can be derived from LSM.

| DISCUSS ION
The intervals of liver stiffness used to define different stages of liver diseases differ among studies.Recent studies have suggested a wide range of cut-offs, e.g.<7 kPa or <8 kPa, 8-12 kPa, >12 kPa and >12.5 kPa in viral hepatitis to define groups with different risks among patients with chronic liver disease. 9,36In our sensitivity analysis of LSM-intervals with a higher granularity than suggested in the Baveno documents, 26 we could show that patients with a mildly increased liver stiffness of 8-9.9 kPa compared to <6 kPa had a slightly increased risk of cirrhosis with portal hypertension, both in relative and absolute terms.However, the cumulative incidence at 2 years was below .5% in patients with an LSM <10 kPa whereas the absolute risk in patients with an LSM ≥10 kPa was found to be more pronounced.

| Strengths and limitations
The strengths of our study were a large sample size, even larger than the number of patients included in each of the subgroup analyses in the previously mentioned meta-analysis. 11Also, we had almost no loss of follow-up.This yields more precise estimations of the risk, although some subgroups were small and had few outcomes.
Some caution should therefore be made when interpreting the results from such subgroups.Patients from both tertiary and secondary care centres were included, which improves the generalizability of the results.Moreover, outcomes related to portal hypertension and diagnostic codes of HCC have previously been validated in a Swedish setting. 37They were shown to have a positive predictive value of 90%-95%, and the risk of misclassification of outcomes in our study is therefore low.
Treatment of liver disease at, or after, baseline may affect the results and is difficult to fully account for.A major effect modifier is likely DAA therapy in hepatitis C. In our sensitivity analysis to assess the association between LSM and risk of cirrhosis with portal hypertension independent of DAA, we found that risk estimates were modestly higher in patients unexposed to DAA with an LSM ≥10 kPa.This is expected since patients that reach a sustained viral response after treatment have a reduced risk of portal hypertension. 38This association was not as pronounced in HCC, but there were few outcomes in LSM-subgroups in the DAA-unexposed cohort.We did not consider treatments for specific liver diseases other than DAA for HCV due to lack of data.Another limitation is that the follow-up time was shorter in aetiologies other than viral hepatitis, and as a result there were few outcomes in these subgroups, especially for HCC.Also, patients were diagnosed at different stages of their chronic liver disease at the time of their baseline LSM.The time a patient has their baseline LSM might depend on the aetiology of disease and other parameters and should be taken into consideration when interpreting the results.Unfortunately, we lacked laboratory data for a majority of the cohort, thus being unable to compare the prognostic ability of LSM in combination with lab parameters such as Agile 3+, 39 or with the prognostic ability of other non-invasive biomarkers such as FIB-4.Since this was a register-based study, clinical parameters that can be important risk factors of progression of chronic liver disease, such as alcohol consumption or BMI, were also lacking.We could estimate the prevalence of some parameters, such as obesity based on ICD-coding, but since this is most likely underreported as diagnostic codes, the prevalence of obesity is likely under-estimated.Also, we choose to define patients with a diagnosis of NAFLD as MASLD since 99% of patients match the MASLD criteria, despite not having sufficient data to perform a comparison in all patients. 40Lastly, there were few patients with a very high LSM, making estimates uncertain and they should therefore be interpreted with caution.

| CONCLUSIONS
Increased liver stiffness measured by VCTE is strongly associated with an increased risk for progression to cirrhosis with portal hypertension or HCC and has a high prognostic ability across different aetiologies of chronic liver disease.These results further support that LSM by VCTE can be used to guide risk stratification for patients with chronic liver disease.
(4.3-8.0)years, corresponding to 92 407 person-years.During follow-up, 402 (2.7%) patients developed cirrhosis with portal hypertension.An increased liver stiffness TA B L E 1 Baseline characteristics of patients examined with vibration-controlled transient elastography.
(4.3-8.1)years, corresponding to 93 094 person-years.During follow-up, 220 (1.5%) patients developed HCC.An increased liver stiffness was associated with an increased rate of HCC (Table severity, from early liver disease with normal liver stiffness, to compensated cirrhosis, indicating low risk of selection bias compared to single-centre studies from tertiary care.Hence, our data highlights VCTE as a method to estimate prognosis in patients with chronic liver diseases.The cumulative risk estimates obtained from this study can be used to inform patient monitoring, follow-up and treatment decisions.

F I G U R E 2 F I G U R E 3 TA B L E 4
Hazard ratios (HRs) for development of cirrhosis with portal hypertension (A) or hepatocellular carcinoma (B) in patients examined with vibration-controlled transient elastography.The horizontal axis shows liver stiffness in kPa and the vertical axis shows the HR for different values of liver stiffness using a liver stiffness of 10 kPa as reference.The dashed lines indicate the 95% confidence interval.Cumulative incidence curves of cirrhosis with portal hypertension (A) and hepatocellular carcinoma (B) stratified on liver stiffness measurement (LSM) by vibrationcontrolled transient elastography.The horizontal axis shows time in years since LSM and the vertical axis shows the cumulative incidences in percent (%).Risk of hepatocellular carcinoma among patients a liver stiffness measurement at baseline examined with vibrationcontrolled transient elastography.
in liver stiffness, the HR for development of cirrhosis with portal hypertension was similar among all aetiologies.The C-indices were above .8 in hepatitis C, hepatitis B, ALD, MASLD and autoimmune liver disease, and above .7 in patients with other liver diseases, suggesting that the overall results on the absolute risk of progression to portal hypertension can be used for different liver diseases.A recent meta-analysis of 2518 patients with MASLD

Median (IQR) or n (%)
a Autoimmune liver diseases are defined as either autoimmune hepatitis, primary sclerosing cholangitis, and primary biliary cholangitis.bNo code for any specific liver disease recorded in the used registers.cAvailable only in 2256 cases.