Introduction

High levels of liver enzymes such as γ-glutamyltransferase (GGT), alanine aminotransferase (GPT), and aspartate aminotransferase (GOT) are considered markers of liver dysfunction. These enzymes could indicate fat accumulation in the liver and are known to be elevated in those with insulin resistance [1,2,3]. Previous population-based studies have shown that these enzymes of liver function were associated with an increased risk of diabetes, independent of obesity and daily alcohol consumption [4,5,6,7,8,9]. The reported associations were evident even within normal ranges of liver enzymes. GPT represents the most specific marker of liver function because it is mainly observed in the liver, whereas GGT and GOT are less specific markers because they are also correlated to other conditions. Two meta-analyses of population-based studies showed that both GGT and GPT are independent predictors of diabetes [4, 5]. A Mendelian randomized study showed causal evidence for the association of GPT and GOT with the risk of diabetes [6].

Alcohol consumption has a dose-dependent impact on liver enzymes, and elevated GGT is the most typical biomarker of excess alcohol use [10]. In a meta-analysis of 20 cohort studies, light alcohol consumption was reported to have a protective role against incident diabetes [11]. The combined impact of liver enzymes and alcohol consumption on the risk of diabetes is rarely investigated, despite its potential utility for risk classification for primary prevention of diabetes among the general population.

In this study, we analyzed the data from the Suita study to investigate the combined impact of liver enzymes and alcohol consumption on the risk of diabetes. We hypothesized that liver enzymes are positively associated with the risk of diabetes, but that these associations would likely be altered by daily alcohol consumption.

Methods

Study population

The Suita study is a population-based study of urban residents launched more than 30 years ago, in 1989 [12,13,14]. Residents aged 30–79 years were randomly selected from the municipality population registry and were stratified by sex and 10-year age groups, and ultimately, 8360 men and women underwent at least one health check-up in the National Cerebral and Cardiovascular Center. Participants were enrolled from the original cohort between 1989 and 1996, a secondary cohort between 1996 and 1998, and a volunteer group between 1992 and 2006. After excluding 2,388 participants who had histories of cardiovascular diseases and diabetes, no follow-up data, or age greater than 79 years, 5972 participants were available for this analysis. Informed consent was obtained from all study participants, and the institutional review board approved this study of the National Cerebral and Cardiovascular Center (R21024).

Follow-up and ascertainment of cases

Participants were followed up to (1) the date of first diagnosis with diabetes; (2) the date of the last health examination and medical records; or (3) December 31, 2015 (date of censorship). The median follow-up was 12.6 years with an interquartile range of 5.9–19.2 years. Participants were invited to have a health check-up every 2 years. Diabetes was defined as a fasting blood glucose level ≥ 126 mg/dL, non-fasting blood glucose level ≥ 200 mg/dL, or the use of diabetes medication.

Measurement of liver enzymes and covariates

Blood tests, including liver enzymes, total cholesterol, high-density lipoprotein cholesterol, and blood glucose, were performed as part of the regular health check-up. After at least 5 min of rest, blood pressure was measured in a seated position using a mercury sphygmomanometer and a suitable-sized cuff according to the standard protocol. Body mass index was calculated as weight (kg) divided by height in meters squared (m2). Details regarding lifestyles such as smoking, drinking (status and daily alcohol consumption in current drinkers), and medications use were obtained using a standard-format interview. Participants were grouped by drinking status, first as never-drinkers, former drinkers, and current drinkers; then the category “nondrinkers” was created to encompass both never-drinkers and former drinkers whereas the “current drinkers” group members were further defined by daily alcohol consumption as light drinkers (< 23 g ethanol/day in men and < 11.5 g in women), moderate drinkers (23–45.9 g and 11.5–22.9 g), and heavy drinkers (≥ 46 g and ≥ 23 g). Impaired glucose tolerance was defined as fasting plasma glucose levels of 100–125 mg/dL or non-fasting glucose levels of 140–199 mg/dL. The estimated glomerular filtration rate (eGFR) was calculated using the Modification of Diet in Renal Disease equation modified by the Japanese coefficient:

eGFR = 0.881 × 186 × age−0.203 × serum creatinine −1.154 × 0.742 (for women) [15].

Statistical analysis

Analysis of variance was used to compare mean values, and chi-square test was used to compare proportions of baseline characteristics. Cox proportional hazard regression model was used to calculate hazard ratio with 95% confidence interval (CI) according to quartile of liver enzymes or drinking status. Potential confounding factors included age (years), body mass index groups (< 18.5, 18.5–24.9, or ≥ 25.0 kg/m2), systolic blood pressure (mmHg), antihypertensive medication use (no or yes), total cholesterol (mg/dL), high-density lipoprotein cholesterol (mg/dL), antihyperlipidemic medication use (no or yes), eGFR (mL/min/1.73 m2), impaired glucose tolerance (no or yes), exercise habits (no or yes), current smoking (no or yes), drinking status (nondrinkers, light drinkers, moderate drinkers, and heavy drinkers) or alcohol consumption (g/day), and other liver enzymes (U/L). We also calculated the hazard ratio with 95% CI according to 1 SD increment of liver enzymes and clinical threshold, which was defined as more than 50 U/L for men and 30 U/L for women for γ-GTP, and more than 40 U/L for GPT and GOT. P for interaction was calculated by a cross-product term that sex (dichotomous) was multiplied by liver enzymes levels or drinking status (categories). All analyses were performed with SAS Enterprise Guide version 7.1 (SAS Institute, Cary, NC, USA).

Results

Baseline characteristics of participants are shown in Table 1. There were 5972 participants, of whom 2735 (45.8%) were men; their mean age was 54.3 years with a standard deviation of 12.4 years. On average, the participants who developed diabetes during follow-up were younger and had more likely to be men compared to that who did not. Also, the participants with diabetes had higher levels of liver enzymes, blood pressure levels, and body mass index, a higher proportion of impaired glucose tolerance, current smokers, moderate and heavy drinkers, and lower levels of serum HDL cholesterol compared to those who did not develop diabetes.

Table 1 Baseline characteristics according to diabetes status during the follow-up
Full size table

During the 76,793 person-years of follow-up, a total of 597 diabetes cases were diagnosed. The incidence rate of diabetes was 7.8 per 1000 person-years. Hazard ratios (95% CIs) for incident diabetes according to liver enzymes or drinking status are shown in Table 2. Liver enzyme levels were positively associated with the age-adjusted risk of diabetes. After controlling for traditional risk factors, the association remained statistically significant: the multivariable hazard ratios (95% CIs) of the highest versus lowest quartile were 1.98 (1.44–2.72) for GGT, 2.02 (1.48–2.74) for GPT, and 1.47 (1.12–1.95) for GOT. No sex difference was observed; P for sex–interactions were 0.20 for GGT, 0.55 for GPT, and 0.34 for GOT. The corresponding hazard ratios (95% CIs) of 1 SD increment were 1.11 (1.03–1.19), 1.15 (1.03–1.29), and 0.95 (0.83–1.08), respectively. The association between daily alcohol consumption and diabetes took the form of a U-shaped curve; the multivariable hazard ratios (95% CIs) were 0.61 (0.43–0.86) for light drinkers, 0.80 (0.63–1.03) for moderate drinkers, and 0.97 (0.68–1.39) for heavy drinkers compared to nondrinkers, with nonsignificant sex–interactions (P = 0.58).

Table 2 Hazard ratios and 95% confidence intervals for incident diabetes according to liver enzymes and drinking status
Full size table

Multivariable hazard ratios and 95% confidence intervals for incident diabetes according to cross groups between liver enzymes and daily alcohol consumption are shown in Table 3. Compared to nondrinkers with the lowest quartiles of liver enzymes, nondrinkers and current moderate/heavy drinkers with the highest quartiles had increased risk of diabetes. However, current light drinkers with the highest quartiles of liver enzymes had no significant risk of diabetes; the corresponding multivariable hazard ratio (95% CI) was 1.27 (0.68–2.37) for GGT, 1.05 (0.59–1.89) for GPT, and 0.76 (0.40–1.47) for GOT.

Table 3 Multivariable hazard ratios and 95% confidence intervals for incident diabetes according to liver enzymes and daily alcohol consumption, compared to participants who were nondrinkers and had the lowest liver enzymes quartile
Full size table

Discussion

In this study, we observed a positive association between liver enzymes and the risk of diabetes and a U-shaped association between daily alcohol consumption and the risk of diabetes. The associations between GGT and GPT and diabetes were evident in nondrinkers and moderate/heavy alcohol drinkers, but not in light drinkers.

In a meta-analysis of seven population-based studies, GGT and GPT were positively associated with the risk of diabetes; the pooled hazard ratios of the highest versus lowest quartile were 2.94 (95% CI, 1.98–3.88; I2 = 20% and P = 0.26) for GGT and 2.02 (95% CI, 1.59–2.58; I2 = 27% and P = 0.19) for GPT [4]. The association for GOT was not reported in that study. Another meta-analysis of 17 population-based studies showed a consistent result for GPT (pooled hazard ratio = 1.66; 95% CI, 1.31–2.09; I2 = 88% and P < 0.001). However, this result showed considerable heterogeneity, and no explanation could be found among the study characteristics. Besides, publication bias was reported [5]. These unexplained heterogeneities limited meta-analytic results. In this situation, a separate prospective study based on a local general population may provide better evidence for clinical practice guidelines. The pooled association for GOT was first investigated in a meta-analysis of nine population-based studies; the pooled hazard ratio of the highest versus lowest tertile was 1.03 (95% CI, 0.98–1.09; I2 = 65% and P = 0.003), with a similar high heterogeneity [5]. That study of population-based studies of GOT associations was expanded from 9 to 13 in a subsequent meta-analysis performed by the same research team; the revised pooled hazard ratio of the highest versus lowest tertile was 1.09 (95% CI, 1.03–1.14; I2 = 73% and P < 0.001), whereas the statistical significance disappeared in the six studies that GGT and GPT were adjusted [7].

A bidirectional Mendelian randomization study showed associations of GPT and GOT with an increased risk of diabetes; the odds ratios (95% CIs) were 1.45 (1.10–1.92) for GPT, 1.25 (1.14–1.38) for GOT. Fasting insulin, but not diabetes, was associated with increased GPT. However, unlike previous observational studies, weak causal evidence was observed for GGT (odds ratio = 0.92; 95% CI, 0.80–1.06) [6], which was consistent with another Mendelian randomization study (odds ratio = 1.01; 95% CI, 0.99–1.02) [8].

Several mechanisms have been proposed to explain the apparent association between serum liver enzymes and incident diabetes. It is well-known that increased liver enzymes, especially GPT, reflect liver fat accumulation, which contributes to pathophysiological changes that are relevant to the development of diabetes [16, 17]. Besides, GPT is considered a marker of both insulin resistance and atherosclerotic vascular disease [18]. GGT is an enzyme responsible for the extracellular catabolism of antioxidant glutathione and its high value reflects oxidative stress [19]. Data from 3,086 women without diabetes, GPT, and GGT were positively associated with fasting glucose, fasting insulin, and HbA1c. These correlations were not substantially changed after excluding hyperinsulinemic women, and similar findings were observed in both obese and non-obese non-diabetic women [2]. Data from 1309 non-diabetic participants showed that GGT and GPT were inversely associated with insulin sensitivity, and GGT and GPT were positively associated with higher insulin secretion rates and reduced endogenous clearance of insulin and liver insulin extraction during the oral glucose tolerance test in both sexes [3]. Given that GGT is a non-specific marker and is correlated to many conditions, it could more reflect the potential intermediary related to diabetes pathogenesis. GOT was shown to reflect fat accumulation although the association was much weaker than that for GPT [20].

Increased liver enzymes are observed in current drinkers. GGT, in particular, is a classical biomarker of daily alcohol consumption. However, a meta-analysis of 20 population-based studies showed the protective role of moderate alcohol consumption against the risk of diabetes in both sexes [11]. The reduced diabetic risk could be explained by improved insulin sensitivity with moderate alcohol consumption [21, 22]. Data from 8576 non-diabetic male employees aged 40–55 years showed that both nondrinkers and heavy drinkers with the highest tertiles of liver enzymes had higher risks of diabetes than moderate drinkers with the lowest tertiles after a 4-year follow-up [9]. These findings were similar to ours here, with some exceptions. That study found an inverse linear association between daily alcohol consumption and diabetes; the multivariable odds ratio (95% CI) was 0.93 (0.71–1.22) for light drinkers, 0.69 (0.53–0.89) for moderate drinkers, and 0.71 (0.54–0.93) for heavy drinkers, compared to nondrinkers. The almost liner association was inconsistent with the U-shaped findings both in a previous meta-analysis [11], and in the present study. The difference could be due to the different population characteristics. Our study consisted of both sexes and enrolled participants from the general population so that our findings in turn could be extrapolated to the general population.

The strength of our study was the cohort design and cohort quality control. Cohort members were randomly enrolled from the population registry with stratified sex and age groups so that the possibility of selection bias should be small. However, the study also had several limitations. First, we could not classify the types of diabetes because we did not have sufficient diagnosis basis. However, the incidence of type 1 diabetes in Japan is very low [23], and most participants in our study were older than 40 years, so the cases of type 1 diabetes might be small. Second, we could not rule out the possibility of recall bias in drinking status because it was surveyed by a standard-form interview. However, we have a team of well-trained nurses and staff to perform the interview. Third, physical activity was not fully investigated. We controlled for the impact of physical activity by exercise habits only, not by exercise and/or type of daily work. Finally, residual confounding cannot be excluded because of the observational study design.

In conclusion, higher levels of liver enzymes were associated with an increased risk of diabetes in the general population. An excessive diabetic risk was observed in nondrinkers and current moderate/heavy alcohol drinkers with high levels of liver enzymes, whereas no increased diabetic risk was observed in current light drinkers, even in these with high levels of liver enzymes. The findings of our study highlight the impact of daily alcohol consumption on the association of liver enzymes with diabetic risk, and provide evidence for clinical guidance on how best to classify individuals’ diabetes risk when they have high levels of liver enzymes.