Liver Stiffness Is Associated With Risk of Decompensation, Liver Cancer, and Death in Patients With Chronic Liver Diseases: A Systematic Review and Meta-analysis

Search Strategy

First, a systematic literature search of Medline (January 1, 1966 through February 1, 2013), Embase (January 1, 1988 through February 1, 2013), and Web of Science (January 1, 1993 through February 1, 2013) databases was conducted with the help of an expert librarian to identify all relevant articles on the role of LSM in predicting clinically relevant outcomes in patients with CLD. Medical subject heading terms used in the search included a combination of “liver” AND “stiff*,” “elastogra*,” “Fibroscan,” combined with “outcome*,” “prognos*,” “predict*,” “course,” “cancer,” “death,” “mortal*,” “transplant*.” The details of the search strategy are included in Supplementary Table 1. The title and abstract of studies identified in the search were reviewed by 2 authors independently (S.S. and L.L.F.) to exclude studies that did not address the research question of interest, based on prespecified inclusion and exclusion criteria (see later). The full text of the remaining articles was examined to determine whether it contained relevant information. Next, the reference sections of the selected articles and review articles on the topic were searched manually for additional studies. Third, a manual search of conference proceedings of major gastroenterology and hepatology conferences (The Liver Meeting, organized by the American Association for the Study of the Liver; The International Liver Congress, organized by the European Association for the Study of the Liver; and Digestive Diseases Week, organized in conjunction with the American Gastroenterological Association) between 2008 and 2012 was conducted to identify additional studies published only in abstract form. This time period was chosen because elastographic techniques began to be used increasingly for the assessment of liver fibrosis in this time period.

Selection Criteria

Considering the prognostic objective of this systematic review, we included prospective and historical cohort studies that met the following inclusion criteria: (1) LSM was performed (using TE or MRE) at time of cohort entry in patients with CLD (regardless of etiology and stage of disease) who were free of reported outcome at time of LSM; (2) systematically assessed the development of clinically relevant outcomes (decompensated cirrhosis, HCC in patients who previously were free of HCC, need for LT in patients who had a native liver and/ or all-cause or liver-related mortality); (3) had a minimum follow-up period of 6 months for all patients; and (4) reported a measure of association—hazard ratio (HR), relative risks (RR), sensitivity, and specificity of LSM for predicting outcomes or provided sufficient data for their calculation. Inclusion was not otherwise restricted by study size, language, or publication type. We excluded the following studies: (1) case-control studies, cross-sectional studies, and case series; (2) studies that assessed the use of LSM in patients with no evidence of CLD; or (3) provided insufficient data to allow estimation of RR with 95% confidence interval (CI) of outcomes. In case of multiple publications from the same cohort, data from the most recent comprehensive report were included. If, however, different outcomes were reported in different publications on the same cohort, then data from these studies were pooled together to interpret as one study with different outcome measures to minimize the risk of double-counting while at the same time maximizing available data. Figure 1 summarizes the process of study identification, inclusion, and exclusion.

Data Abstraction

Data on the following were abstracted onto a standardized form: (1) study characteristics: primary author; time period of study/year of publication; country of the population studied; duration of follow-up evaluation (mean or median, total person-years of follow-up evaluation), as well as attrition; (2) patient characteristics: age, sex, body mass index; etiology of CLD (viral hepatitis, other causes); stage of CLD (early fibrosis defined as stages 0–2, advanced fibrosis, compensated cirrhosis, decompensated cirrhosis); treatment for underlying liver disease (and its distribution among those who developed outcomes and those who did not); (3) exposure assessment: technique of LSM, reported failure of LSM, whether LSM was reported as a continuous variable or categorical variable (and, if so, the categories reported); (4) outcomes reported: development of decompensated cirrhosis (variceal bleeding, ascites, hepatic encephalopathy, hepatorenal syndrome, spontaneous bacterial peritonitis), HCC, need for LT, overall- or liver-related mortality; (5) statistical analysis: HR or RR and 95% CIs with and without adjustment for confounding factors, or sensitivity/specificity of LSM for predicting outcomes; and (6) confounding variables, adjusted for in each study, especially for model for end-stage liver disease (MELD) score (or its components) and treatment of the underlying CLD. The authors of included studies were contacted for missing data. Conflicts in data abstraction were resolved by consensus, referring back to the original article and in consultation with the principal investigator (J.A.T.).

Quality Assessment

The quality assessment of prognostic studies was performed by 2 study investigators (S.S., L.L.F.) independently, using the Quality In Prognosis Studies tool,¹⁴ which evaluates validity and bias in studies of prognostic factors across 6 domains: participation, attrition, prognostic factor measurement, confounding measurement, outcome measurement, and analysis and reporting.

Outcomes

The primary analysis focused on assessing the prognostic use of LSM in predicting each clinically relevant outcome in patients with CLD: decompensated cirrhosis, HCC, and mortality. Because these outcomes are interdependent and not independent of each other, we report the association between LSM and each of these outcomes separately. In addition, some studies reported the risk of a composite outcome (hepatic decompensation, HCC, need for LT, and/or mortality), and these were pooled separately. Anticipating potential heterogeneity in the direction and magnitude of effect among the studies, we performed preplanned subgroup analyses on study-related variables to explore sources of heterogeneity. For each individual outcome, these included the following: (1) patient characteristics: location of study (Asian population vs Western population), etiology (viral hepatitis vs mixed), and stage (compensated cirrhosis vs all stages) of underlying liver disease, and (2) elastography-related variables: measure of LSM used (ultrasound-based vs magnetic resonance–based).

Statistical Analysis

We used the random-effects model using the DerSimonian and Laird¹⁵ method to calculate meta-analytic RR and the 95% CI for each outcome. We assumed HR to be equivalent to RR. When studies reported sensitivity and specificity of LSM for predicting outcomes, we calculated an unadjusted RR by formulating a 2 × 2 table.

When studies reported exposure grouped into categories to provide a dose-specific RR (using the lowest category as the referent category), we transformed this into a risk estimate per unit of LSM (called imputed RR), using linear-trend meta-analytic statistical methodology.^16,17 Briefly, we assigned the midpoint of the cut-off points of the class as the dose value. For studies with open-ended categories, we used the lowest and highest reported LSM from the study to calculate the midpoint. We then calculated the RR for that range of LSM (subtracting the midpoints from the highest risk category with the lowest risk category) to estimate a per-unit RR, after log-transformation. This methodology assumes a linear relationship between LSM and logarithm of RR. If studies reported the relationship between LSM and outcomes as per unit, these were preferred for estimation (called reported RR).

We assessed heterogeneity between study-specific estimates using 2 methods: first, the Cochran Q test, which tests the null hypothesis that all studies in a meta-analysis have the same underlying magnitude of effect, was measured.¹⁸ Because this test is underpowered to detect moderate degrees of heterogeneity,¹⁹ a P value of less than .10 was considered suggestive of significant heterogeneity. Second, to estimate what proportion of total variation across studies was related to heterogeneity rather than chance, the I² statistic was calculated. A value of greater than 50% was suggestive of considerable heterogeneity.¹⁸ We explored potential sources of heterogeneity using multivariable meta-regression according to predefined characteristics: study location (Asian vs Western population), etiology (viral vs mixed) and stage of CLD (compensated cirrhosis vs all stages), method of estimating RR (reported vs imputed), and whether analysis was adjusted for hepatic synthetic function, HVPG, or MELD (yes vs no). A P value less than .05 was suggestive of the grouping variable being a significant source of heterogeneity.

Because of the limited number of studies, a formal estimation of publication bias was not performed.²⁰ Analyses and graphs were conducted using Comprehensive Meta-Analysis version 2 (Biostat, Englewood, NJ) and Stata version 12 (StataCorp, College Station, TX).

Characteristics of Included Studies

All studies, except one,³³ used ultrasound-based TE for measurement of liver stiffness; one study used MRE for assessing liver stiffness. The range of LSM in patients in individual studies varied from 1.5 to 75 kPa. The earliest study period started in 2004 and the latest follow-up period ended in 2012. The median follow-up period in individual studies ranged from 16 to 71 months, with the minimum patient follow-up period being at least 7 months. The majority of the studies were conducted in the European population (9 studies)^{11,22,23,28–32,34}; 3 studies were performed in the North American population^26,33,36 and 5 studies were conducted in the Asian population.^{21,24,25,27,35} In 9 of the studies, HCV, with or without human immunodeficiency virus, was the leading cause of CLD.^{22,27,28,30–32,34–36} Two Asian studies were performed exclusively in patients with chronic hepatitis B.^24,25 One study was performed exclusively in patients with primary biliary cirrhosis.¹¹ All of the remaining studies were performed on patients with mixed etiologies of CLD; nonalcoholic steatohepatitis was the leading cause of CLD (41%) in only one of the included studies performed using MRE.³³ Five studies were conducted exclusively in patients with compensated cirrhosis,^{23,28,30,33,34} whereas others included a mix of patient populations at varying stages of fibrosis or cirrhosis (defined based on liver biopsy or LSM). Of note, in studies that included patients with decompensated cirrhosis at entry into the cohort, the subsequent risk of decompensated cirrhosis was not a valid outcome and was not reported; these patients were still at risk for HCC and mortality. Tables 1 and 2 show the baseline characteristics of the included studies.

Quality of Included Studies

The included studies were usually at low to moderate risk of bias with regard to study participation, attrition from study, measurement of liver stiffness, and defining outcomes (except with regard to features of hepatic decompensation). Several studies did not adjust for important confounders, especially with regard to hepatic synthetic function, severity of liver disease, and whether patients were treated for their underlying CLD. In addition, some studies were at high risk of bias because they reported only univariate and/or unadjusted analysis. Figure 2 and Supplementary Table 2 show the overall quality assessment of the included studies using the Quality In Prognosis Studies tool. The agreement between the 2 reviewers for quality assessment (88.5%) was very good.

Risk of Hepatic Decompensation

On meta-analysis of 6 studies (216 cases of hepatic decompensation),^{22,25,28,29,33,34} each unit increase in LSM was associated with a 7% higher risk of developing decompensated cirrhosis on follow-up evaluation (Figure 3A). The results were stable across etiology and stage of CLD, across populations, and in studies using TE or MRE for measurement of liver stiffness (Table 3). The results also were similar after restricting the analysis to studies that adjusted for conventional measures of hepatic function and prognostic markers. When we replaced the study by Chon et al²⁵ with another study from the same cohort,³⁸ the results were stable (RR, 1.10; 95% CI, 1.05–1.16).

There was considerable heterogeneity observed in this analysis (P < .01, I² = 84%); this was seen primarily in the magnitude of effect, but not in the direction of effect. In the multivariable meta-regression analysis, the heterogeneity was attributable to the technique of LSM used. In the one study using MRE, the risk estimates were significantly higher (RR, 1.57; 95% CI, 1.24–1.99) as compared with 5 studies performed using TE (RR, 1.06; 95% CI, 1.03–1.09). The results were stable when restricting analysis to studies in which per-unit analysis was reported, rather than imputed (5 studies; RR, 1.07; 95% CI, 1.03–1.11).

Risk of HCC

Of the 9 included studies (214 cases of HCC),^{21,22,24,25,27,28,30,34,35} 8 reported a significant association between baseline LSM and subsequent risk of developing HCC. On meta-analysis, each unit increase in LSM was associated with an 11% higher risk of HCC (Figure 3B). The results were consistent across Asian and Western populations,aswell as across studies that included patients with compensated cirrhosis only or at all stages of CLD (Table 4). All studies were performed in patients with viral hepatitis as the etiology of CLD, and all studies used TE as a modality for LSM. The results were consistent when replacing the studybyChonetal²⁵ withanother study from the same cohort(RR, 1.14; 95% CI, 1.07–1.22).¹⁰ Considerable heterogeneity was observed in the overall analysis (P < .01, I² = 94%); meta-regression analysis was not able to explain heterogeneity based on stage of CLD, study location, level of adjustment in original studies, or method of imputation. The relationship between LSM and future risk of HCC continued to be significant on restricting analysis to studies in which per-unit analysis was reported, rather than imputed (5 studies; RR, 1.05; 95% CI, 1.00–1.09).

Risk of Mortality

Meta-analysis of 5 studies (144 deaths) showed that baseline LSM significantly predicted the risk of mortality in patients with CLD (Figure 3C).^{22,24,28,31,32} Four studies showed a significant positive association with effect sizes varying from 1.09 to 1.44. All studies were performed in patients with viral hepatitis with or without human immunodeficiency virus. The results were stable across location. Significant heterogeneity was observed in the analysis (P < .01; I² = 88%). The heterogeneity might be explained on the basis of analysis—studies in which the imputed per-unit RR showed a significant difference from those that reported per-unit RR (Table 5). In a sensitivity analysis limited to 2 studies that reported per-unit analysis, one of the studies noted the presence of a dose-response relationship,²² and the other study did not observe such an association.²⁸ Overall, a significant relationship was not observed (RR, 1.03; 95% CI, 0.89–1.19).

Risk of Composite Outcome

In 7 studies (all performed in Western populations),^{11,23,26,29,33,36} the relationship between LSM and a composite outcome (hepatic decompensation, HCC, need for LT, or mortality) was reported. Meta-analysis of these studies showed a significant positive association in 6 of these 7 studies, with a summary estimate of 1.32 (Figure 3D). The results were stable across study location, etiology, and stage of CLD, as well as based on whether analysis was adjusted for other known predictors of outcomes (Table 6). Considerable heterogeneity (P < .01; I² = 85%) was identified. However, meta-regression analysis could not identify any obvious sources of heterogeneity based on study, patient, or statistical characteristics.

Sensitivity Analyses

Because meeting abstracts have limited reporting of events and do not undergo rigorous peer-review, we performed sensitivity analyses restricted only to articles published in peer-reviewed journals. A meta-analysis of these studies showed a stable association between baseline LSM and risk of hepatic decompensation (N = 6 studies including our own study; RR, 1.08; 95% CI, 1.03–1.13), HCC (N = 7 studies; RR, 1.11; 95% CI, 1.04–1.18), and composite outcome (N = 6 studies; RR, 1.31; 95% CI, 1.14–1.51). All studies assessing the relationship between LSM and mortality were published as full-text articles.

On limiting analyses to studies that were at low risk for bias based on statistical analysis and reporting and appropriately controlled for confounding factors (hepatic synthetic function and treatment of CLD), the association between LSM and risk of hepatic decompensation (N = 3 studies; RR, 1.05; 95% CI, 1.00–1.11), HCC (N = 3 studies; RR, 1.05; 95% CI, 1.02–1.08), and composite outcome (N = 2 studies; RR, 1.16; 95% CI, 1.01–1.34) remained stable. Only one study was deemed to be at low risk of bias based on appropriate controlling for confounding and adequate statistical analysis and reporting, for assessing the association between LSM and future risk of mortality, and this study was negative (RR, 0.94; 95% CI, 0.80–1.10).