Liver Enzymes and Risk of All-Cause Mortality
Liver Enzymes and Risk of All-Cause Mortality
Our initial search identified 3286 potentially relevant citations. After screening of titles and abstracts, 55 articles remained for further evaluation. Following detailed assessments, 36 articles were excluded of which 6 involved study populations with pre-existing disease. Overall, 19 articles based on 19 unique prospective cohort studies were included in the meta-analysis (Figure 1). One study combined results of three independent cohorts. In aggregate, the included studies comprised 9 238 201 non-overlapping participants and 242 953 all-cause mortality outcomes. A single study which provided results of unpublished data contributed over 90% of data (comprising 8 922 358 participants and 220 216 events) to the present review.
(Enlarge Image)
Figure 1.
Selection of studies included in the meta-analysis. ALP, alkaline phosphatase; ALT, alanine aminotransferase; AST, aspartate aminotransferase; GGT, gamma glutamyltransferase
Table 1 provides details of the eligible studies that evaluated baseline circulating GGT, ALT, AST and ALP levels with risk of all-cause mortality outcomes and their quality assessment scores. All included studies were prospective cohort studies carried out in USA, Europe (UK, The Netherlands, Germany, Italy and Austria), and Asia (Korea and Japan) with an average baseline age of 51 years. Duration of follow-up for all-cause mortality outcomes ranged from 5 to 20 years, with a median follow-up of 10 years. The majority of studies achieved the highest score for quality. Table 2 provides assay characteristics of measured levels of liver enzymes from studies contributing to the analysis. All studies evaluated associations in approximately general populations.
The combined RR (95% CI) for all-cause mortality outcomes in a comparison of individuals in the top thirds with those in the bottom thirds of baseline GGT level for the 11 studies in fully adjusted analysis was 1.60 (1.42–1.80) with substantial heterogeneity between studies (I >70%) (Figure 2). Exclusion of any single study at a time from the meta-analysis had minimal effect on the pooled RRs, which ranged from 1.53 (1.36–1.73) to 1.65 (1.46–1.87). The combined RR excluding the single largest study was 1.60 (1.37–1.86), very similar to the main finding. Little of the heterogeneity in the contributing studies was explained by differences in several study level characteristics other than study size (P for meta-regression = 0.05; Figure 3). A stronger association was observed in studies with ≥500 outcomes: 1.73 (1.49–2.00) compared with smaller studies: 1.31 (1.15–1.50). Between-study heterogeneity was substantial in several study characteristic categories. There was less heterogeneity in studies conducted in North American populations, studies of small size and studies of the highest quality. In further exploration of heterogeneity, this was substantially reduced when analysis was restricted to studies of the highest quality. The strong positive association was consistently observed across several subgroups. The Egger test for publication bias was not significant (P = 0.52), consistent with observed funnel plot symmetry (Supplementary Figure 1, available as Supplementary data at IJE online).
(Enlarge Image)
Figure 2.
Relative risk for all-cause mortality in individuals in the top compared with the bottom third of levels of liver enzymes in eligible studies, 1995–2013. Study acronyms are provided in Table 1. The summary estimate presented was calculated using a random-effects model. Degree of adjustment: +, adjusted for age and/or sex; ++, further adjustment for established vascular risk factors; +++, additional adjustment for alcohol consumption, other liver markers or inflammatory markers; ALP, alkaline phosphatase; ALT, alanine aminotransferase; AST, aspartate aminotransferase; CI, confidence interval (bars); GGT, gamma glutamyltransferase; RR, relative risk
(Enlarge Image)
Figure 3.
Prospective studies of GGT levels with all-cause mortality risk, grouped according to several study characteristics, 1995–2012. The summary estimate presented was calculated using a random-effects model; Sizes of data markers are proportional to the inverse of the variance of the relative ratio; CI, confidence interval (bars); GGT, gamma glutamyltransferase; RR, relative risk; *, P-value for meta-regression
The pooled RR (95% CI) for all-cause mortality in a comparison of extreme thirds of baseline level of ALT was 1.03 (0.76–1.41), with substantial heterogeneity (I > 70%) across the eight studies (Figure 2). Analysis examining the influence of a single study on the combined RR showed similar results. The combined RR on exclusion of the single largest study, which could have potentially influenced our findings, was 1.07 (0.82–1.40). The inconsistency across the studies that evaluated ALT was partially explained by geographical location (P for meta-regression = 0.0002; Supplementary Figure 2, available as Supplementary data at IJE online). The pooled RRs for studies conducted in North America, Europe and Asia were 0.82 (0.78–0.86), 0.95 (0.84–1.07) and 1.43 (1.08–1.90), respectively. Similarly as for studies of GGT, heterogeneity was also less among studies conducted in Europe and North America, smaller studies and studies of the highest quality. In further exploration of heterogeneity, this was substantially reduced when we restricted the analysis to studies of the highest quality. Among the four remaining studies, the pooled RR also did not show evidence of an association of ALT with all-cause mortality, at 0.94 (0.82–1.07).
In pooled results of the only four studies of AST, there was no strong evidence of an association with all-cause mortality outcomes [RR 1.23 (0.80–1.88)] (Figure 2). The limited number of studies for AST precluded us from investigating the potential sources of heterogeneity. As funnel plots are unlikely to be useful in meta-analysis containing fewer than 10 studies, publication bias was not evaluated for studies of the aminotransferases.
Circulating ALP level was associated with a 38% higher risk of all-cause mortality outcomes (pooled RR: 1.38, 1.17–1.63) in a comparison of extreme thirds of baseline level of ALP (Figure 2). Substantial heterogeneity (I > 70%) was observed across the four studies. Exclusion of any single study at a time (including the largest study) from the meta-analysis did not change the direction of the association, yielding pooled RRs which ranged from 1.26 (1.13–1.41) to 1.44 (1.23–1.70). Meta-regression, stratified analyses and evaluation of publication bias were not conducted because of the limited number of studies.
Figure 4 shows the dose-response relationships of levels of liver enzymes with risk of all-cause mortality outcomes for pooled results of studies providing relevant data. For the dose-response relation between baseline GGT level and all-cause mortality risk, examination of the figure did not suggest substantial departure from linearity though the test for a nonlinear relation was marginally significant (P for nonlinearity = 0.06; Figure 4A). The pooled RR per 5 U/l increment in GGT levels was 1.07 (1.04–1.10). In dose-response analysis stratified by geographical location, monotonous risk increases for all-cause mortality across increasing GGT levels were apparent in both North American and European populations (Figure 4B and C).
(Enlarge Image)
Figure 4.
Dose-response relations between levels of liver enzymes and relative risks of all-cause mortality. (A) Gamma glutamyltransferase; (B) gamma glutamyltransferase in North American populations; (C) gamma glutamyltransferase in European populations; (D) alkaline phosphatase; (E) alanine aminotransferase; (F) alanine aminotransferase in Asian populations; (G) aspartate aminotransferase. Adjusted relative risks and 95% confidence intervals (CIs; dashed lines) are reported. Data were modelled with restricted cubic splines with 3 knots in random-effects dose-response models. The median values of the lowest reference ranges were used to estimate all relative risks. The vertical axes are on log scales
There was evidence for a linear relation between ALP levels and all-cause mortality risk (P for nonlinearity = 0.61; Figure 4D). A 5-U/l increment in ALP levels conferred a RR of 1.03 (1.01–1.06). These findings suggest that the pooling of dose-response estimates using GLST analysis for GGT and ALP with all-cause mortality outcomes was appropriate. In contrast, significant nonlinearity (P for nonlinearity < 0.0001; Figure 4E) was detected for ALT levels and all-cause mortality, with no effect of levels of ALT up to about 17 U/l, followed by an increase in risk beyond these levels. In analysis restricted to Asian populations (two studies), the pooled RR for all-cause mortality per 5-U/l increase in ALT level was 1.11 (1.09–1.12) and examination of the dose-response figure did not suggest substantial departure from linearity though the test for nonlinearity was marginally significant (P for nonlinearity = 0.05; Figure 4F). The few data points precluded us from assessing the dose-response associations in European and North American populations. The cubic spline model that included two studies on AST levels also indicated a nonlinear relation between all-cause mortality risk and AST levels (P for nonlinearity < 0.0001; Figure 4G).
Using the most recent mortality statistics for the USA and Europe, the estimated absolute risk differences for all-cause mortality were 41.4 and 37.7 cases per 100 000 individuals per year for USA and Europe, respectively, for every 5-U/l increment in GGT levels. The corresponding estimates for every 5-U/l increment in ALP levels were 27.6 and 6.3 cases per 100 000 individuals per year, respectively.
Results
Study Identification and Selection
Our initial search identified 3286 potentially relevant citations. After screening of titles and abstracts, 55 articles remained for further evaluation. Following detailed assessments, 36 articles were excluded of which 6 involved study populations with pre-existing disease. Overall, 19 articles based on 19 unique prospective cohort studies were included in the meta-analysis (Figure 1). One study combined results of three independent cohorts. In aggregate, the included studies comprised 9 238 201 non-overlapping participants and 242 953 all-cause mortality outcomes. A single study which provided results of unpublished data contributed over 90% of data (comprising 8 922 358 participants and 220 216 events) to the present review.
(Enlarge Image)
Figure 1.
Selection of studies included in the meta-analysis. ALP, alkaline phosphatase; ALT, alanine aminotransferase; AST, aspartate aminotransferase; GGT, gamma glutamyltransferase
Study Characteristics and Study Quality
Table 1 provides details of the eligible studies that evaluated baseline circulating GGT, ALT, AST and ALP levels with risk of all-cause mortality outcomes and their quality assessment scores. All included studies were prospective cohort studies carried out in USA, Europe (UK, The Netherlands, Germany, Italy and Austria), and Asia (Korea and Japan) with an average baseline age of 51 years. Duration of follow-up for all-cause mortality outcomes ranged from 5 to 20 years, with a median follow-up of 10 years. The majority of studies achieved the highest score for quality. Table 2 provides assay characteristics of measured levels of liver enzymes from studies contributing to the analysis. All studies evaluated associations in approximately general populations.
Gamma Glutamyltransferase and All-Cause Mortality
The combined RR (95% CI) for all-cause mortality outcomes in a comparison of individuals in the top thirds with those in the bottom thirds of baseline GGT level for the 11 studies in fully adjusted analysis was 1.60 (1.42–1.80) with substantial heterogeneity between studies (I >70%) (Figure 2). Exclusion of any single study at a time from the meta-analysis had minimal effect on the pooled RRs, which ranged from 1.53 (1.36–1.73) to 1.65 (1.46–1.87). The combined RR excluding the single largest study was 1.60 (1.37–1.86), very similar to the main finding. Little of the heterogeneity in the contributing studies was explained by differences in several study level characteristics other than study size (P for meta-regression = 0.05; Figure 3). A stronger association was observed in studies with ≥500 outcomes: 1.73 (1.49–2.00) compared with smaller studies: 1.31 (1.15–1.50). Between-study heterogeneity was substantial in several study characteristic categories. There was less heterogeneity in studies conducted in North American populations, studies of small size and studies of the highest quality. In further exploration of heterogeneity, this was substantially reduced when analysis was restricted to studies of the highest quality. The strong positive association was consistently observed across several subgroups. The Egger test for publication bias was not significant (P = 0.52), consistent with observed funnel plot symmetry (Supplementary Figure 1, available as Supplementary data at IJE online).
(Enlarge Image)
Figure 2.
Relative risk for all-cause mortality in individuals in the top compared with the bottom third of levels of liver enzymes in eligible studies, 1995–2013. Study acronyms are provided in Table 1. The summary estimate presented was calculated using a random-effects model. Degree of adjustment: +, adjusted for age and/or sex; ++, further adjustment for established vascular risk factors; +++, additional adjustment for alcohol consumption, other liver markers or inflammatory markers; ALP, alkaline phosphatase; ALT, alanine aminotransferase; AST, aspartate aminotransferase; CI, confidence interval (bars); GGT, gamma glutamyltransferase; RR, relative risk
(Enlarge Image)
Figure 3.
Prospective studies of GGT levels with all-cause mortality risk, grouped according to several study characteristics, 1995–2012. The summary estimate presented was calculated using a random-effects model; Sizes of data markers are proportional to the inverse of the variance of the relative ratio; CI, confidence interval (bars); GGT, gamma glutamyltransferase; RR, relative risk; *, P-value for meta-regression
Aminotransferases and All-Cause Mortality
The pooled RR (95% CI) for all-cause mortality in a comparison of extreme thirds of baseline level of ALT was 1.03 (0.76–1.41), with substantial heterogeneity (I > 70%) across the eight studies (Figure 2). Analysis examining the influence of a single study on the combined RR showed similar results. The combined RR on exclusion of the single largest study, which could have potentially influenced our findings, was 1.07 (0.82–1.40). The inconsistency across the studies that evaluated ALT was partially explained by geographical location (P for meta-regression = 0.0002; Supplementary Figure 2, available as Supplementary data at IJE online). The pooled RRs for studies conducted in North America, Europe and Asia were 0.82 (0.78–0.86), 0.95 (0.84–1.07) and 1.43 (1.08–1.90), respectively. Similarly as for studies of GGT, heterogeneity was also less among studies conducted in Europe and North America, smaller studies and studies of the highest quality. In further exploration of heterogeneity, this was substantially reduced when we restricted the analysis to studies of the highest quality. Among the four remaining studies, the pooled RR also did not show evidence of an association of ALT with all-cause mortality, at 0.94 (0.82–1.07).
In pooled results of the only four studies of AST, there was no strong evidence of an association with all-cause mortality outcomes [RR 1.23 (0.80–1.88)] (Figure 2). The limited number of studies for AST precluded us from investigating the potential sources of heterogeneity. As funnel plots are unlikely to be useful in meta-analysis containing fewer than 10 studies, publication bias was not evaluated for studies of the aminotransferases.
Alkaline Phosphatase and All-Cause Mortality
Circulating ALP level was associated with a 38% higher risk of all-cause mortality outcomes (pooled RR: 1.38, 1.17–1.63) in a comparison of extreme thirds of baseline level of ALP (Figure 2). Substantial heterogeneity (I > 70%) was observed across the four studies. Exclusion of any single study at a time (including the largest study) from the meta-analysis did not change the direction of the association, yielding pooled RRs which ranged from 1.26 (1.13–1.41) to 1.44 (1.23–1.70). Meta-regression, stratified analyses and evaluation of publication bias were not conducted because of the limited number of studies.
Dose-Response Analyses
Figure 4 shows the dose-response relationships of levels of liver enzymes with risk of all-cause mortality outcomes for pooled results of studies providing relevant data. For the dose-response relation between baseline GGT level and all-cause mortality risk, examination of the figure did not suggest substantial departure from linearity though the test for a nonlinear relation was marginally significant (P for nonlinearity = 0.06; Figure 4A). The pooled RR per 5 U/l increment in GGT levels was 1.07 (1.04–1.10). In dose-response analysis stratified by geographical location, monotonous risk increases for all-cause mortality across increasing GGT levels were apparent in both North American and European populations (Figure 4B and C).
(Enlarge Image)
Figure 4.
Dose-response relations between levels of liver enzymes and relative risks of all-cause mortality. (A) Gamma glutamyltransferase; (B) gamma glutamyltransferase in North American populations; (C) gamma glutamyltransferase in European populations; (D) alkaline phosphatase; (E) alanine aminotransferase; (F) alanine aminotransferase in Asian populations; (G) aspartate aminotransferase. Adjusted relative risks and 95% confidence intervals (CIs; dashed lines) are reported. Data were modelled with restricted cubic splines with 3 knots in random-effects dose-response models. The median values of the lowest reference ranges were used to estimate all relative risks. The vertical axes are on log scales
There was evidence for a linear relation between ALP levels and all-cause mortality risk (P for nonlinearity = 0.61; Figure 4D). A 5-U/l increment in ALP levels conferred a RR of 1.03 (1.01–1.06). These findings suggest that the pooling of dose-response estimates using GLST analysis for GGT and ALP with all-cause mortality outcomes was appropriate. In contrast, significant nonlinearity (P for nonlinearity < 0.0001; Figure 4E) was detected for ALT levels and all-cause mortality, with no effect of levels of ALT up to about 17 U/l, followed by an increase in risk beyond these levels. In analysis restricted to Asian populations (two studies), the pooled RR for all-cause mortality per 5-U/l increase in ALT level was 1.11 (1.09–1.12) and examination of the dose-response figure did not suggest substantial departure from linearity though the test for nonlinearity was marginally significant (P for nonlinearity = 0.05; Figure 4F). The few data points precluded us from assessing the dose-response associations in European and North American populations. The cubic spline model that included two studies on AST levels also indicated a nonlinear relation between all-cause mortality risk and AST levels (P for nonlinearity < 0.0001; Figure 4G).
Absolute Risk Differences Associated With Increasing Levels of Liver Enzymes
Using the most recent mortality statistics for the USA and Europe, the estimated absolute risk differences for all-cause mortality were 41.4 and 37.7 cases per 100 000 individuals per year for USA and Europe, respectively, for every 5-U/l increment in GGT levels. The corresponding estimates for every 5-U/l increment in ALP levels were 27.6 and 6.3 cases per 100 000 individuals per year, respectively.
Source...