Liver stiffness measurement predicted liver-related events and all-cause mortality: A systematic review and nonlinear dose-response meta-analysis.

Numerous studies have investigated the prognosis value of the liver stiffness measurement (LSM) by transient elastography in assessing the risk of liver-related events (LREs) and all-cause mortality in patients with chronic liver disease (CLD). However, the shape of the dose-response relationship between them remains unclear. We searched PubMed, Embase, the Cochrane Library, and reference lists of articles for studies published up to July 1, 2017, that assessed the LSM in predicting LREs and all-cause mortality among subjects with CLD. Fifty-four observational cohort studies with 35,249 participants were included. Summary relative risks (RRs) were calculated using a random-effects model, and a restricted cubic spline function was used to model the dose-response association. LREs and all-cause mortality were increased in subjects with a high LSM (LRE: RR, 7.90; 95% confidence interval [CI], 5.65, 11.05; I2 = 71.6%; all-cause mortality: RR, 4.15; 95% CI, 2.56, 6.72; I2 = 68.5%). For each unit increment of liver stiffness, the summary RR was 1.06 (95% CI, 1.06, 1.07; I2 = 74.6%) for LREs and 1.06 (95% CI, 1.04, 1.07; I2 = 55.7%) for all-cause mortality. A positive relationship with a nonlinear trend for LSM with LREs and all-cause mortality was examined by a dose-response meta-analysis (P < 0.001). When stratified by etiology, a nonlinear association was also found in patients infected with hepatitis C virus and those coinfected with hepatitis C virus and human immunodeficiency virus. In contrast, there was no evidence of departure from linearity among patients with hepatitis B virus infection (Pnonlinearity = 0.072).
Conclusion: LSM is useful in screening LREs and all-cause mortality in patients with CLD. Further studies are warranted in assessing the application of LSM in monitoring the risk of LREs and all-cause mortality in clinical practice. (Hepatology Communications 2018;2:467-476).

confidence interval

chronic liver disease

hepatitis B virus

hepatocellular carcinoma

hepatitis C virus

human immunodeficiency virus

liver‐related event

liver stiffness measurement

relative risk

sustained virologic response

transient elastography

Introduction

Chronic liver disease (CLD) is an increasing health burden worldwide. Cirrhosis and CLDs accounted for 2% of deaths globally in 2015 and showed a relative increase of 10.3% from 2005.1 It is well known that advanced liver fibrosis and cirrhosis are significantly associated with the risk of development of liver‐related events (LREs), such as hepatocellular carcinoma (HCC), hepatic decompensation, and liver‐related mortality.2 According to the largest study available,3 the total number of deaths from cirrhosis and liver cancer have steadily risen by approximately 50 million per year. Therefore, early diagnosis of cirrhosis and accurate risk stratification are essential for managing CLDs for both interventional strategies and estimating prognosis. Liver biopsy has traditionally been the gold standard for assessing liver fibrosis4; however, liver biopsy has potential complications, inherent sampling error, and intra‐observer and interobserver variability in interpretation of the histology.5 In light of these limitations, attention has turned to noninvasive methods for the assessment of liver fibrosis and disease severity, and these methods have been increasingly implemented in national and international guidelines and extensively validated in the hospital setting.6, 7

Transient elastography (TE), as a surrogate for liver fibrosis, is the most widely used technique for assessing the degree of liver fibrosis, which can be expressed numerically as a continuous variable with accurate, reproducible, and reliable results.8, 9 In patients with liver cirrhosis, liver stiffness measurements (LSMs) were able to distinguish between patients with compensated and decompensated cirrhosis.8 In addition, TE is currently the most promising approach for predicting LREs and all‐cause mortality in patients with CLD6, 7; however, it is still limited to noninvasively assessing complications and prognosis in patients with advanced liver disease.10

An epidemiologic‐based analysis for confirming the usefulness of LSM in predicting the risk of LREs and all‐cause mortality is required. A previous meta‐analysis showed a linear association between liver stiffness and the risk of LREs in patients with CLD11; however, the included studies were limited and the linear relationship was uncertain. More recently, results from large cohort studies have been reported. We therefore had sufficient statistical power to precisely evaluate the shape of the dose–response association between LSM and the risk of LRE development in patients with CLD. Our analysis contributes to the growing evidence that favors the routine clinical application of LSM in predicting LREs in the management of patients with CLD.

Materials and Methods

We conducted this systematic review and meta‐analysis according to guidance provided by the Cochrane Handbook for Systematic Reviews of interventions.12 Our reporting follows the Preferred Reporting Items for Systematic Reviews and Meta‐Analyses guidelines (PRISMA; http://onlinelibrary.wiley.com/doi/10.1002/hep4.1154/full).13

LITERATURE SEARCH STRATEGY

We conducted a computer‐aided systematic literature search of PubMed, Embase, and Cochrane Library databases from inception through July 1, 2017, with the help of an experienced medical librarian. Our objective was to identify all relevant original studies and conference proceedings about LSM in predicting clinically relevant outcomes in patients with CLD. Medical subject heading terms used in the search included a combination of the following: “liver,” “hepatic,” AND “stiff*,” “elastogra*,” “Fibroscan,” “sonoelastography,” “elastography,” “transient elastography,” combined with “outcome*,” “prognos*,” “predict*,” “decompensa*,” “cancer,” “death,” “mortal*,” “diagnos*,” “hepatocellular carcinoma,” “complication,” “cirrhosis,” “LRE,” “HCC,” “liver‐related event.” Details of the search strategy are available in http://onlinelibrary.wiley.com/doi/10.1002/hep4.1154/full. Subsequently, the title and abstract of studies identified in the search were reviewed by two investigators (J.W., J.L.) independently to exclude studies that did not address the research question of interest based on prespecified inclusion and exclusion criteria (see below). The full text of the remaining articles was reviewed again to determine whether it contained relevant information. Next, to improve yield, additional studies were searched manually from the bibliographies of the selected studies and review articles on the topic as well as those listed in the published meta‐analyses for additional articles.11 We then consulted with experts in the field to identify additional published and unpublished primary studies.

SELECTION CRITERIA

Considering the prognostic objective of this systematic review, we included observational cohort studies that met the following inclusion criteria: (1) selected studies were prospective or retrospective cohort studies; (2) LSM was performed using TE at the time of cohort entry on the study participants with CLD (regardless of etiologies and stages of fibrosis) who were free of reported LREs at the time of LSM; (3) outcomes of interest were the development of LREs (hepatic decompensation, HCC, and/or liver‐related mortality) and all‐cause mortality; (4) all participants had a minimum follow‐up period of 12 months; and (5) estimates of the hazard ratio or relative risk (RR) with corresponding 95% confidence intervals (CIs) were reported or the number of cases and total participants were provided for each category of LSM. 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) liver stiffness assessed with noninvasive tools other than TE; (3) the use of TE was assessed in participants with no evidence of CLD at enrollment; (4) insufficient data were provided to allow estimation of relative risk and corresponding 95% CIs of the outcomes; and (5) participants enrolled in the study had pre‐existing LREs, such as HCC and hepatic decompensation, or were undergoing liver transplantation at baseline. Studies were selected by reviewing the title and abstract after duplicated articles were removed from the primary search. The remaining studies were confirmed with the original and validated by applying the inclusion and exclusion criteria. Two independent authors reviewed the results of the search using the inclusion criteria. Disagreements between these authors were resolved by discussion. The process of study identification, inclusion criteria, and exclusion criteria is summarized in Fig. 1A.

Figure 1

Characteristics of the study. (A) Flow diagram of the study selection. (B) Analysis of the general information of the included studies. (C) Quality assessment of included studies using the Quality In Prognosis Studies tool. Abbreviations: H, high risk of bias; IF, impact factor; L, low risk of bias; M, moderate risk of bias.

DATA ABSTRACTION

Data were extracted by two reviewers independently (J.W., J.L.). Differences in opinion with regard to the data were resolved by discussion until a consensus was reached. The included data were abstracted onto a standardized form as follows: (1) study characteristics: primary author and year of publication; time period of study; country of the population studied; study type (prospective or retrospective); duration of follow‐up evaluation (mean or median); (2) patient characteristics: mean age, sex (% male), body mass index, number of total participants and cases; etiologies of underlying CLD; fibrosis stage on liver biopsy (and classification system used); treatment for underlying liver disease; number of patients who achieved sustained virologic response (SVR); (3) liver stiffness assessment: baseline LSM and whether LSM was reported as a continuous variable or categorical variable; (4) outcomes reported: development of decompensated cirrhosis (including variceal bleeding, ascites, hepatic encephalopathy, hepatorenal syndrome, and spontaneous bacterial peritonitis), HCC, liver‐related or all‐cause mortality; (5) statistical analysis: hazard ratio or RR and 95% CIs with and without adjustment for confounding factors, or the number of cases and total participants for each category of LSM; and (6) confounding variables adjusted for in each study. 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 (W.H).

QUALITY ASSESSMENT

Two investigators (J.W., J.L.) used the Quality In Prognosis Studies tool11, 14 to determine the quality of the prognostic studies. This tool measures validity and bias in the six parameters of participation, attrition, prognostic factor measurement, confounding measurement, outcome measurement, and analysis and reporting. Details of the quality assessment are shown in http://onlinelibrary.wiley.com/doi/10.1002/hep4.1154/full.

STATISTICAL ANALYSIS

We used random‐effects models to calculate summary RRs and 95% CIs for the highest versus lowest LSM with an increased risk of LREs and all‐cause mortality. For continuous risk estimates, summary RRs and 95% CIs per unit increment in LSM were pooled using a random‐effects model. P < 0.05 was considered statistically significant for a two‐tailed analysis.

The dose–response relationship between liver stiffness and risk of LREs in patients with CLD was estimated using the method described by Greenland and Longnecker.15 The method requires that the distribution of cases and person‐years or non‐cases and the RRs with the variance estimates are known for at least three quantitative exposure categories. Midpoint values were used for all studies, and liver stiffness was presented as intervals. When the lowest category was open ended, the lower boundary was assumed to be zero. When the upper boundary of the highest category was not provided, the median value was assigned as the lower boundary multiplied by 1.5. Forest plots of the linear dose–response results are presented for RRs in 5‐kPa increments of liver stiffness. To examine a potential nonlinear dose–response relationship between LSM and LRE risk, we performed a two‐stage random‐effects meta‐analysis, as summarized.16 In the first stage, we constructed study‐specific restricted cubic spline models to model liver stiffness with four knots at fixed percentiles (5%, 35%, 65%, and 95%) of the distribution.17 In the second stage, we combined the study‐specific estimates and the variance/covariance matrix that had been estimated within each study, taking the random effects model into consideration.18 The difference between linear and nonlinear models was assessed by testing the null hypothesis that the regression coefficients of the spline transformations are all equal to zero.19

Statistical heterogeneity between studies was quantitatively assessed using the I 2 statistic, and we defined low, moderate, substantial, and considerable heterogeneity by the cutoffs of <30%, 30%‐59%, 60%‐75%, and >75%, respectively.20 To identify sources of heterogeneity that may influence liver stiffness and the risk of LREs, we performed subgroup analysis by outcomes, etiologies of CLD, geographic location, follow‐up period, number of participants, publication, and study type. Additionally, we carried out sensitivity analyses by excluding one study at a time to investigate whether the results were driven by a single large study. Possible publication bias was evaluated with Egger's regression asymmetry test21 and Begg's test,22 with the results considered to indicate publication bias when P < 0.10. If statistically significant bias was found, the trim and fill method was used for adjustment. All statistical analyses were conducted using Stata 14.0 version software (Stata Corp, College Station, TX).

Results

We identified a total of 5,317 studies with our search strategy. Of these, 54 studies with 35,249 participants, including 38 full‐text publications and 16 meeting abstracts, were ultimately selected according to our inclusion criteria (see the http://onlinelibrary.wiley.com/doi/10.1002/hep4.1154/full for the 54 included studies as references). Specific details of the excluded records are shown in Fig. 1A, and baseline characteristics of the included studies are presented in http://onlinelibrary.wiley.com/doi/10.1002/hep4.1154/full. The earliest study period started in 1988, and the latest follow‐up period ended in 2015. Thirty‐seven prospective studies and 17 retrospective studies were included in the analysis (Fig. 1B,B1). The majority of the studies were performed with patients with hepatitis B virus (HBV) infection (23, 42.6%), hepatitis C virus (HCV) infection (14, 25.9%), and HCV/human immunodeficiency virus (HIV) coinfection (10, 18.5%) (Fig. 1B,B2). Most of the included studies were from Korea (20, 37.0%), Spain (12, 22.2%), and China (6, 11.1%) (Fig. 1B,B3), and the majority of the studies (34, 63.0%) were published between 2013 and 2016 (Fig. 1B,B4). The distribution of the journals' impact factor is shown in Fig. 1B,B5.

QUALITY OF INCLUDED STUDIES

The results indicated a low to moderate risk of bias in study participation, attrition, prognostic factor measurement, outcome measurement, statistical analysis, and reporting. However, several studies did not adjust for potential residual confounders, such as whether the patients achieved SVR at baseline or received therapy during the follow‐up period. In addition, some studies were at a high risk of bias because they reported only univariate analysis or the data were calculated by the number of cases and participants. The overall quality assessment of the included studies using the Quality In Prognosis Studies tool is shown in Fig. 1C and http://onlinelibrary.wiley.com/doi/10.1002/hep4.1154/full.

LIVER‐RELATED EVENTS

High Versus Low

We included 37 studies with 19,889 participants in the meta‐analysis of LREs (http://onlinelibrary.wiley.com/doi/10.1002/hep4.1154/full). The pooled RRs were estimated for the highest categories versus the lowest categories of liver stiffness. When compared to the lowest liver stiffness, patients with the highest LSM were significantly associated with an increased risk of LREs (RR, 7.90; 95% CI, 5.65, 11.05) (http://onlinelibrary.wiley.com/doi/10.1002/hep4.1154/full). However, substantial heterogeneity (I 2 = 71.6%; P = 0.000) and publication bias were found in the analysis (Begger's test z c = 2.20, P = 0.028; Egger's test t = 3.49, P = 0.001) (http://onlinelibrary.wiley.com/doi/10.1002/hep4.1154/full). Because of this, we conducted a trim‐and‐fill method to recalculate our pooled risk estimate, and 16 missing studies were imputed to produce a symmetrical funnel plot (http://onlinelibrary.wiley.com/doi/10.1002/hep4.1154/full). The pooled analysis continued to reveal a statistically significant association between LSM and LRE risk (RR, 3.92; 95% CI, 2.72, 5.66).

To explore the sources of heterogeneity within these studies, we performed subgroup analyses by outcome, etiology of CLD, year of follow‐up, geographic area, and number of participants. There was some evidence of heterogeneity when stratified by CLD etiology (P heterogeneity = 0.000): summary RR was 3.65 (95% CI, 3.01, 4.43; I 2 = 28.70%; P heterogeneity = 0.142) for patients with HBV infection, 20.14 (95% CI, 12.92, 31.38; I 2 = 36.40%; P heterogeneity = 0.138) for patients with HCV infection, 13.71 (95% CI, 8.75, 21.48; I 2 = 75.50%; P heterogeneity = 0.000) for patients with HCV/HIV coinfection, and 5.93 (95% CI, 3.29, 10.69; I 2 = 2.00%; P heterogeneity = 0.395) for patients with mixed etiologies. Restricting the analysis to studies that adjusted for age and sex did not alter the results. When stratified by outcome, the summary RR was 13.12 (95% CI, 7.85, 21.93; I 2 = 47.80%; P heterogeneity = 0.088) for hepatic decompensation, 4.20 (95% CI, 3.41, 5.18; I 2 = 60.20%; P heterogeneity = 0.004) for HCC, 2.73 (95% CI, 1.74, 4.29; I 2 = 71.70%; P heterogeneity = 0.014) for liver‐related mortality, and 10.05 (95% CI, 7.30, 13.82; I 2 = 63.60%; P heterogeneity = 0.000) for the composite outcomes. In addition, we conducted further subgroup analyses by geographic location and found that the summary RR had greater significance in Europe (RR, 12.73; 95% CI, 9.48, 17.09; I 2 = 71.10%; P heterogeneity = 0.000) than in Asia (RR, 3.94; 95% CI, 3.28, 4.73; I 2 = 41.40%; P heterogeneity = 0.025). There were some evidence of heterogeneity when stratified by follow‐up period, with a stronger association among studies that had a longer follow‐up period (P heterogeneity = 0.001) (Table 1). Finally, we conducted sensitivity analyses by omitting one study at a time; the results were not influenced greatly by any of the studies.

Table 1

Stratified Analysis of the Association Between Baseline LSM and Subsequent Risk of LREs

	High Versus Low					Per Unit kPa
Subgroups	Number of Studies	RR (95% CI)	I 2 (%)	P heterogensity	P interaction	Number of Studies	RR (95% CI)	I 2 (%)	P heterogensity	P interaction
All studies	37	7.904 (5.654‐11.048)	71.60	0.000	‐	50	1.065 (1.056‐1.074)	74.6	0.000	‐
Outcomes
HD	6	13.116 (7.845‐21.928)	47.80	0.088	0.000	12	1.062 (1.054‐1.070)	80.5	0.000	0.009
HCC	12	4.201 (3.409‐5.175)	60.20	0.004		15	1.048 (1.041‐1.055)	53.3	0.008
Liver‐related mortality	4	2.734 (1.744‐4.285)	71.70	0.014		4	1.085 (1.055‐1.115)	79.7	0.002
Composite outcome	15	10.047 (7.303‐13.822)	63.60	0.000		19	1.053 (1.047‐1.058)	77.6	0.000
Etiologies of CLD
HBV	15	3.652 (3.013‐4.427)	28.70	0.142	0.000	25	1.046 (1.041‐1.052)	76.0	0.000	0.001
HCV	8	20.135 (12.920‐31.378)	36.40	0.138		11	1.057 (1.051‐1.062)	0.0	0.725
HIV/HCV	7	13.713 (8.754‐21.482)	75.50	0.000		9	1.068 (1.058‐1.078)	88.2	0.000
Mixed	5	5.925 (3.285‐10.687)	2.00	0.395		5	1.077 (1.042‐1.114)	0.0	0.768
Location
Asia	21	3.941 (3.281‐4.734)	41.40	0.025	0.000	27	1.047 (1.042‐1.053)	74.4	0.000	0.001
Europe	15	12.727 (9.477‐17.091)	71.10	0.000		21	1.060 (1.055‐1.065)	75.1	0.000
Follow‐up period
≥3 years	23	9.930 (7.808‐12.629)	69.90	0.000	0.001	33	1.054 (1.050‐1.058)	76.7	0.000	0.900
<3 years	11	4.742 (3.280‐6.856)	0.00	0.899		17	1.054 (1.047‐1.062)	71.2	0.000
Patients
N ≥ 1,000	11	5.634 (4.592‐6.912)	89.20	0.000	0.611	16	1.048 (1.044‐1.053)	85.2	0.000	0.000
N < 1,000	26	5.193 (4.092‐6.590)	25.30	0.120		34	1.067 (1.061‐1.074)	53.3	0.000
Publication type
Full text	29	6.236 (5.239‐7.422)	73.90	0.000	0.001	40	1.054 (1.050‐1.058)	78.6	0.000	0.966
Conference proceedings	8	3.215 (2.283‐4.528)	14.00	0.321		10	1.054 (1.041‐1.067)	16.7	0.289
Study type
Prospective	23	4.873 (4.052‐5.860)	69.8	0.000	0.030	33	1.056 (1.052‐1.060)	51.0	0.000	0.085
Retrospective	14	7.110 (5.337‐9.473)	73.5	0.000		17	1.049 (1.042‐1.056)	87.2	0.000

Abbreviation: HD, hepatic decompensation.

Dose–Response Analysis

Data from 50 studies (23,041 participants) were included in the linear dose–response analysis (http://onlinelibrary.wiley.com/doi/10.1002/hep4.1154/full). The association between estimates of liver stiffness (per unit kPa) and LREs is shown in http://onlinelibrary.wiley.com/doi/10.1002/hep4.1154/full. The pooled RR per unit increment in liver stiffness was 1.06 (95% CI, 1.06, 1.07) with the random‐effects model, with substantial heterogeneity among studies (I 2 = 74.6%, P heterogeneity = 0.000). Results of the funnel plot (http://onlinelibrary.wiley.com/doi/10.1002/hep4.1154/full) and Begg's test (z c = 2.73, P = 0.006) (http://onlinelibrary.wiley.com/doi/10.1002/hep4.1154/full), suggest that publication bias might exist, although the Egger's test was not statistically significant (P = 0.794). Because of this, we used the trim‐and‐fill method to recalculate our pooled risk estimate, and 16 missing studies were imputed to produce a symmetrical funnel plot (http://onlinelibrary.wiley.com/doi/10.1002/hep4.1154/full). The analysis suggested that the imputed risk estimate was 1.05 (95% CI, 1.04, 1.06), which indicated a slightly decreased risk but was still similar to our original risk estimate.

The subgroup analysis showed some evidence of heterogeneity when stratified by the etiologies of CLD (P heterogeneity = 0.001): the summary RR was 1.05 (95% CI, 1.04, 1.05; I 2 = 76.0%; P heterogeneity = 0.000) for patients with HBV infection, 1.06 (95% CI, 1.05, 1.06; I 2 = 0.00%; P heterogeneity = 0.725) for patients with HCV infection, 1.07 (95% CI, 1.06, 1.08; I 2 = 88.2%; P heterogeneity = 0.000) for patients with HCV/HIV coinfection, and 1.08 (95% CI, 1.04, 1.11; I 2 = 0.00%; P heterogeneity = 0.768) for patients with mixed etiologies. When stratified by outcome (P heterogeneity = 0.009), the summary RR was 1.06 (95% CI, 1.05, 1.07; I 2 = 80.5%; P heterogeneity = 0.002) for hepatic decompensation, 1.05 (95% CI, 1.04, 1.06; I 2 = 53.3%; P heterogeneity = 0.008) for HCC, 1.09 (95% CI, 1.06, 1.12; I 2 = 79.7%; P heterogeneity = 0.002) for liver‐related mortality, and 1.05 (95% CI, 1.05, 1.06; I 2 = 77.6%; P heterogeneity = 0.000) for the composite outcomes. We also conducted further subgroup analyses by geographic location and found that the summary RR was higher in Europe (summary RR, 1.06; 95% CI, 1.06, 1.07; I 2 = 75.1%; P heterogeneity = 0.000) than in Asia (summary RR, 1.05; 95% CI, 1.04, 1.05; I 2 = 74.4%; P heterogeneity = 0.000). In addition, the results were in general consistent across subgroups of age and follow‐up period (Table 1). In sensitivity analyses omitting one study at a time, the results were not greatly influenced by any of the studies.

A nonlinear dose–response analysis requires at least three categories of LSM, and studies reporting only a continuous estimate or did not provide the number of cases and participants in each category could not be included. This resulted in only 18 studies with 11,931 participants being included in the analysis (http://onlinelibrary.wiley.com/doi/10.1002/hep4.1154/full). The nonlinear dose–response trend showed a statistically significant increased risk of developing LREs with increasing liver stiffness (P nonlinearity = 0.000) (Fig. 2A) and showed the most pronounced increase in LRE risk among persons with a lower range of liver stiffness and relative risk stabilization at approximately 34.5 kPa. Our dose–response analysis indicated that a 5‐kPa increase in liver stiffness was significantly associated with an increase of approximately 53% in the risk of developing LREs (RR, 1.53; 95% CI, 1.41, 1.65) (http://onlinelibrary.wiley.com/doi/10.1002/hep4.1154/full), with considerable heterogeneity (I 2 = 82.0%; P = 0.000). However, when pooling etiology‐specific results between baseline liver stiffness and the risk of LREs, the shape of the nonlinear curve was steeper among patients with HCV infection or HCV/HIV coinfection (P nonlinearity = 0.000) (Fig. 2B,C). In contrast, there was no evidence of departure from linearity among patients with HBV infection (P nonlinearity = 0.072), indicating that a linear model showed a better fit for liver stiffness and LREs in patients with HBV infection (Fig. 2D).

Figure 2

Dose–response meta‐analyses between LSM and the risk of LREs in patients with (A) CLD, (B) HCV infection, (C) HCV/HIV coinfection, and (D) HBV infection. (E) Dose–response meta‐analyses between LSM and the risk of all‐cause mortality in patients with CLD. Relative risk (solid line) and 95% CI (long dashed lines) from the restricted cubic splines model. Short dashed line represents RR from the linear model.

ALL‐CAUSE MORTALITY

Elevated LSM was associated with an increase in all‐cause mortality in a random‐effects model comparing the highest to the lowest liver stiffness groups (RR, 4.15; 95% CI, 2.56, 6.72), with substantial heterogeneity (I 2 = 68.5%; P = 0.007) (http://onlinelibrary.wiley.com/doi/10.1002/hep4.1154/full). Results were not influenced greatly by any of the studies in the sensitivity analyses when omitting one study at a time (data not shown).

The pooled RR for each unit increment in LSM was 1.06 (95% CI, 1.04, 1.07) (http://onlinelibrary.wiley.com/doi/10.1002/hep4.1154/full); moderate heterogeneity was found (I 2 = 55.7%; P = 0.079) but no evidence of publication bias (Begger's test z c = –0.34, P = 1.000; Egger's test t = 1.71, P = 0.230) (data not shown). Sensitivity analyses showed that the results were not influenced greatly by omitting one study at a time. Three studies with 5,716 participants were included in the nonlinear dose–response analysis (http://onlinelibrary.wiley.com/doi/10.1002/hep4.1154/full); a nonlinear association between LSM and all‐cause mortality was found (P nonlinearity = 0.0005) (Fig. 2E).

Discussion

To our knowledge, this is the first dose–response meta‐analysis conducted in the context of CLD to reveal the clinical usefulness of TE in predicting LREs and all‐cause mortality. Notably, we found that TE can effectively predict LREs and all‐cause mortality in an apparent nonlinear positive association, where liver stiffness at a range below 34.5 kPa is significantly associated with increased risk of LREs. Our results provide an accurate and individualized prediction of LREs by noninvasive diagnostic methods. We recommend that the development of risk algorithms be used in the field of CLD to identify high‐risk patients for progression to LREs. In addition, our findings validate LSM as a strong predictive noninvasive tool for follow‐up patients with CLD in clinical practice.

Our results are in line with those from a pooled meta‐analysis11 that showed a significant linear association per unit increment of liver stiffness with the incident of hepatic decompensation (RR, 1.07), HCC (RR, 1.11), overall mortality (RR, 1.22), or a composite of these outcomes (RR, 1.32) in patients with CLD. However, some evidence of a nonlinear association between LSM and the risk of LREs was found in our dose–response meta‐analysis, which is inconsistent with previous conclusions.11 Possible reasons for this discrepancy include the larger number of additional studies in our analysis that provided sufficient data to explore this nonlinear association. In addition, CLD etiology was a statistically significant effect modifier of this relationship as we found a nonlinear association in patients infected with HCV or HCV/HIV and a linear association only for patients with HBV infection. Considerable evidence supports a similar or higher rate of hepatic decompensation, HCC occurrence, and mortality in patients with liver cirrhosis due to HCV compared to those with liver cirrhosis from other causes.23 Our subgroup analysis also indicated somewhat stronger associations among HCV‐infected patients than among patients with other causes of CLD.

In contrast to Singh et al.11 who found no difference in the summary estimates when stratified by study location, our study showed that the association between liver stiffness and the risk of LREs was significantly higher in Europe than in Asia. CLD is predominantly caused by HBV infection in Asia, whereas in Europe, HCV infection is the major etiology of CLD.24 Therefore, the main determinants for this phenomenon may be dependent on the etiology of CLD rather than geographic location.

Several longitudinal studies have investigated whether dynamic changes in LSM could predict long‐term prognosis in HBV‐infected patients receiving antiviral therapy.25, 26 Wu et al.,26 for example, found a 9.3% increased risk of LREs for every 10% increase in liver stiffness from baseline in patients with compensated cirrhosis undergoing 26 weeks of entecavir treatment. They also found that the effect of Δ%LSM26w‐0w was superior to baseline LSM and other clinical parameters. Additional studies with a larger sample size are needed to confirm the predictive effect of serial LSM changes on LRE risk.

Our meta‐analysis has some limitations. First, inherent limitations occur in the original studies, including heterogeneity of patients, variability in etiologies and stages of CLD, treatment for CLD, study location, sample size, outcome assessments, whether the studies adjusted for important covariates, and the discrepancies in cutoffs of liver stiffness between studies; consequently, substantial heterogeneity was observed in our meta‐analysis. Second, LSM was only evaluated at baseline and not evaluated during follow‐up in most of the studies; however, a greater magnitude of decline in liver stiffness was observed in patients who achieved SVR.27, 28 The impact of magnitude and kinetics of decline in liver stiffness on improvement in LREs needs to be further investigated. Third, LSM might be influenced by nonfasting status, flare of transaminases, etiology of liver disease, type and position of probe, and operator experience.29 Thus, these measures may not reflect usual liver stiffness and may result in some misclassification of long‐term exposure. Additionally, residual confounding by inadequately measured covariates, such as patients who achieved SVR, may influence the risk of LREs; however, very limited data were used to adjust this. As a consequence, the potential impact of this confounding on the change in liver stiffness remains to be elucidated.

The strengths of our meta‐analysis include a relatively large number of studies with approximately 3,424 cases and 35,249 participants in the analysis; thus, we had statistical power to explore the associations between LSM and the risk of LREs in patients with CLD. We conducted several subgroup analyses and found the results to be stable in most; the findings were also robust in sensitivity analyses. We also explored the nonlinear association of LSM with LREs and all‐cause mortality found in patients with CLD; to our knowledge, this is the first time this has been investigated. In addition, inclusion of all available cohort studies provide stronger evidence regarding an association than case‐control studies because they are less prone to differential recall of selection bias and do not restrict analysis based on publication type or language, hence minimizing the risk of selection bias.

In summary, our meta‐analysis, which is the most up‐to‐date review of the evidence, highlights the prognostic value of TE in predicting long‐term clinically relevant outcomes in patients with CLD. Our data suggest that LSM is significantly associated with the risk of LREs in a nonlinear dose–response relationship; this provides a cost‐effective method for annual surveillance of patients with CLD and will be an invaluable aid to physicians in the prediction of patient prognosis in routine clinical application. We recommend that all patients with CLD undergo a noninvasive evaluation of liver fibrosis at baseline and that those patients with a higher liver stiffness be seen more frequently. Further high‐quality studies in patients with similar clinical backgrounds are required to elucidate the prognostic utility of LSM and assess whether it can become a standard prognostic tool for predicting the long‐term clinical outcome in patients with CLD.

Author names in bold designate shared co‐first authorship

Additional Supporting Information may be found at http://onlinelibrary.wiley.com/doi/10.1002/hep4.1154/full.

Supporting Information 1

Click here for additional data file.

Supporting Information Figure S1

Click here for additional data file.

Supporting Information Figure S2

Click here for additional data file.

Supporting Information Figure S3

Click here for additional data file.