Association of Alanine Transaminase to Aspartate Aminotransferase Ratio and Incident Diabetes in Chinese Adults: A Second Retrospective Cohort Study

Previous studies have indicated that serum alanine aminotransferase (ALT) level, serum aspartate aminotransferase (AST) level or ALT/AST ratio may be bound up with diabetes. But the association of ALT/AST ratio and incident diabetes in Chinese adults have not been elaborated yet. The main objective of the present study was to analyze whether ALT/AST ratio was an independent risk factor which attached to the risk of incident diabetes in Chinese adults.

The liver which plays a signi cant role in the maintenance of normal glucose levels has been involved in the pathologic process of type 2 diabetes (T2DM), and hepatic dysfunction from insulin resistance syndrome has been considered as guiding factor to T2DM (3). Pyruvate and glutamate which formed through alanine and α-ketoglutarate, play a pivotal role in glucose and protein metabolism, and alanine transaminase (ALT) is an negotiator enzyme which reversibly catalyzes transamination (4). ALT and aspartate aminotransferase (AST) are the special marker of hepatic injury that were normally detected in diabetes (5). Many studies have indicated that serum ALT level, serum AST level or ALT/AST ratio may be bound up with diabetes (6)(7)(8)(9). But these studies mostly paid attention to investigate whether ALT, AST or ALT/AST ratio were related to insulin resistance, the association of ALT/AST ratio and incident diabetes have not been elaborated yet, and they did not focus on the Chinese population by the way. Hence, the primary purpose of the present article was to analyze whether ALT/AST ratio was an independent risk factor which attached to the risk of incident diabetes in Chinese adults which conducted a large cohort population.
In the present article, we accomplished a secondary retrospective cohort analysis based on previously published data which the author have investigated the association between BMI and age with incident diabetes (10). Through our analysis, the results suggested that ALT/AST ratio was used as an independent risk factor, and the result variables and covariates in accordance with those in the primal study.

Data source and participants
We acquired raw data from 'DATADRYAD' database (www.Datadryad.org), a website that allowed subscribers to download data at liberty. Abided by Dryad Terms, we quoted Dryad data package in this Data which enrolled 685,277 participants ≥20 years old who received a health check with at least two visits from 2010 to 2016, were provided by the Rich Healthcare Group in China across 32 sites and 11 cities in China(Shanghai, Beijing, Nanjing, Suzhou, Shenzhen, Changzhou, Chengdu, Guangzhou, Hefei, Wuhan, Nantong). The data has been already excluded participants by follow rules: (1) diagnosed with diabetes at baseline, (2) follow-up time less than 2 years, (3) no acquirable message about gender, height, weight or fasting plasma glucose level at baseline, (4) with immoderate BMI values (<15 kg/m 2 or >55 kg/m 2 ), (5) with unde ned diabetes status. Ultimately, 211,833 participants were brought into the analysis by Ying Chen, et al, and the details about the inclusion/exclusion standard and result variables of the trial had been informed in the article of Ying Chen, et al (10). On account of the character of retrospective cohort analysis, informed consent or study approval of the patient was not demanded by the institutional ethics committee. In order to research forwardly, we excluded the missing data of baseline ALT (n=1782) and AST (n=123290). Then, the ALT/AST ratio was calculated as ALT divided by AST, 87,881 subjects were elicited. We also removed the ALT/AST outliers (n=1222) by the rules as follow: (1) lower than means minus three standard deviation (SD), (2) higher than means plus three SD). Ultimately, 86,659 participants including 48,942 males and 37,717 females were brought into our study for further research.

Study excogitation and data measurement
The excogitation of the study was referenced from the article of Ying Chen, et al, and the data we obtained for the present retrospective cohort analysis originated from that article too (10). For clarifying our study process, we would list the procedure of our study as follows. Every single participant who visited the health check center, would nish a questionnaire as required. The detail of the questionnaire included demographic characteristics, personal medical history, family history of chronic disease and lifestyle factors as well. Trained staff would measure and collect the data of height which was measured to the nearest 0.1 cm, weight which was measured in light clothing with no shoes to the nearest 0.1 kg and blood pressure which was measured by standard mercury sphygmomanometers of every single participant. BMI was calculated by deriving from weight in kilograms divided by height in metres squared. An autoanalyzer named Beckman 5800 was used to test FPG, cholesterol, triglyceride, SCR, ALT, AST, HDL and LDL. The target independent variable, ALT/AST was acquired at baseline and the dependent variable, the risk of incident diabetes was acquired during the following up. On account of the character of retrospective cohort analysis, we need to minify the observation bias and selection bias.

Con rmation of incident diabetes
Incident diabetes' con rmation comply with fasting plasma glucose whose over 7.00 mmol/L and/or self-reported diabetes during the follow-up. We evaluated the participants when they were diagnosed or arrived the last visit. The data of those participants who lost follow-up also consisted in the whole study.

Statistical analyses
In the rst place, we lled the missing data with the median or mean value when the missing data of the variable was continuous and approached the category variable when it was categorical (11). Secondly, we separated all participants by quartiles of ALT/AST. We used medians (quartiles, skewed distribution) or means ± standard deviations (normal distribution) to represent the continuous variables and expressed the categorical variables by percentages or frequency. The incident diabetes rate was calculated on the whole and use the log-rank test to compare and then used cumulative incidence and person-years incidence rate to represented. In addition, we used the Kruskal-Wallis H (skewed distribution) test, the oneway ANOVA (normal distribution) and chi-square test (categorical variables) to pick out the evidential differences among the means and proportions in the groups. Moreover, the prognostic value of ALT/AST on the risk of incident diabetes was explored by Cox proportional hazard regression models and hazard ratios (HRs) which were adjusted by 95% con dence intervals (CIs) were gured out to assess the incident diabetes risk. Besides, we concurrently manifested the results into unadjusted, minimally adjusted and fully adjusted analyses according to the recommendation of the STROBE statement (12). When the covariances was enrolled in the model and they have changed the matched hazard ratio at least by 10%, we treated them as adjusted (13). We also conducted a sensitivity analysis by regenerating the ALT/AST as a categorical variable and counting the P for trend that would help further con rm as a continuous variable and check the possibility of nonlinearity for the robustness of analysis. Furthermore, generalized additive models (GAM) was applied to determine the non-linear relationship and once we checked a non-linear relationship, we would count the threshold effect of the ALT/AST on incident diabetes according to the smoothing plot by conducting a two-piecewise linear regression model. If a smooth curve emerged between ALT/AST and incident diabetes, the in ection point would be counted voluntarily by using recursive method (maximum pseudo likelihood estimation model). What is more, we used Cox proportional hazard models to investigate other subgroups (age, gender, BMI, FPG, SBP, DBP, HDL, LDL, family history of diabetes, smoking status, drinking status) for making sure of the characteristic of analysis. We replaced the continuous variable categorically in terms of their clinical demarcation point and adjusted every single strati cation for all the mixed factors. We also used likelihood ration tests to examine the modi catory and interaction of the subgroups. Through the Kaplan-Meier method speci cally the time-to-rst event for each endpoint, we canvassed the survival estimate and cumulative event rate. As for adverse event, we used the log-rank test to examine the Kaplan-Meier hazard ratios (HRs) and the 95% con dence intervals (CIs) of them.
All statistical analysis mentioned above was utilized by Empower-Stats (http://www.empowerstats.com, X&Y Solutions, Inc., Boston, MA) or the R Project (http://www.R-project.org, The R Foundation) and statistically signi cant was con rmed when the P values less than 0.05.

Baseline characteristics
Quartiles of the ALT/AST and the total population's baseline characteristics were shown in Table 1. We distributed the total population into subgroup according to ALT/AST quartiles (≤ 0.64, 0.64-0.83, 0.83-1.10, > 1.10). In higher ALT/AST group, participants were observed mostly have higher BMI, fasting plasma glucose, SBP, DBP, cholesterol, SCR and more current smoking and drinking status. On the contrary, none of family history status was considered statistically signi cant difference amongst Quartiles of the ALT/AST.   The consequence of univariate analyses was demonstrated in Table 3 that age, BMI, FPG, SBP, DBP, cholesterol, triglyceride, SCR, LDL were undeniably related to incident diabetes. Besides, the outcome also manifested that men had a higher risk of incident diabetes than women, the participants with family history of diabetes have a higher risk of incident diabetes than without family history, and the participants with current drinking or smoking status have a higher risk of incident diabetes than the participants with never or ever drinking or smoking status. Values are n (%), median(Q1-Q3) or mean ± sd The Kaplan-Meier curves of the cumulative hazards of incident diabetes risk was depicted in Fig. 1 when strati ed by quartiles of ALT/AST. Incident diabetes risk between each quartile of ALT/AST was evidentially various (log-rank test, p < 0.0001). When ALT/AST increased, the cumulative incident diabetes risk raised by degrees, and the Q4 ALT/AST (top of the quartiles) was notarized having maximum risk of incident diabetes.

Association of ALT/AST and incident diabetes
The association of ALT/AST and incident diabetes was judged by Cox proportional hazard regression model in which  As ALT/AST ratio was a continuous variable, the non-linear relationship of ALT/AST and incident diabetes was identi ed through generalized additive models (GAM). In Table 5 and Fig. 2 to analyze the relation between ALT/AST and incident diabetes is appropriate. In Table 6, subgroup analysis was used to examined if other variables (age, gender, BMI, FPG, SBP, DBP, HDL, LDL, family history of diabetes, smoking status, drinking status) would affect the relation between ALT/AST and incident diabetes. By treating them as strati ed variables, we could study their impact on the association of ALT/AST and incident diabetes. As the output result depicted, age, BMI and SBP were detected interactional through the priori speci cation test (all P values for interaction < 0.05). The population with age (30 to < 40 years), BMI (< 24 kg/m2) and SBP (< 140 mmHg) was found to be the stronger association. On the contrary, the population with age (20 to < 30 years, > 50 years), BMI (≥ 24 kg/m 2 ) and SBP (≥ 140 mmHg) was found to be the weaker association.

Discussion
In the present study, by using the generalized linear models (Cox proportional hazard models), ALT/AST was observed positively associate with incident diabetes when other variables were adjusted. Moreover, the analyses of generalized additive models (GAM) demonstrated that the relation between ALT/AST and incident diabetes is only linear. Subgroup analysis was conducted to further disclosure the association of ALT/AST and incident diabetes in hierarchical populations. The population with age (30 to < 40 years), BMI (< 24 kg/m2) and SBP (< 140 mmHg) was found to be the stronger association. On the contrary, the population with age (20 to < 30 years, > 50 years), BMI (≥ 24 kg/m 2 ) and SBP (≥ 140 mmHg) was found to be the weaker association.
In former study, ALT/AST has been brought into evaluate insulin resistance (7,14,15). Hanley et al. have informed that ALT/AST could be used for assessing metabolic syndrome through decide insulin sensitivity and acute insulin response (16,17). Besides, former researches have con rmed that hepatic diseases like non-alcoholic fatty liver disease (NAFLD) were related to glucose metabolism and the prevalence of diabetes (18,19). And the elevation of liver enzymes such as ALT, AST and gammaglutamyl transpeptidase (GGT) or the ALT/AST ratio are treated as surrogate markers of NAFLD in common (20)(21)(22) (24). Former research showed that ALT could be used for assessing the risk of incident diabetes independent of the fasting glucose, adiposity and family history of diabetes (25). And in a prospective cohort study which performed an analyses of 6484 participants with age (≥ 40 years old) in Korean indicated that individuals with high aminotransferase levels, the risk of incident diabetes was three times higher in those who have highest quartile of BMI than those in the lowest quartile of BMI (26). Yet, only a few investigations have focused considerably the association between ALT/AST and incident diabetes in the Chinese population. Thus, our study had huge objects (685,277) across 32 sites and 11 cities in China that could represent Chinese population rigorously. And the results creatively suggested that ALT/AST is an independent risk factor for incident diabetes in Chinese adults.
Subgroup analysis is extraordinary important for a retrospective cohort analysis. Subgroup analysis was generally used for investigating the coherence of the clinical trial's result among various subpopulations de ned by each of multiple baseline characteristics of the patients or assessing the treatment effect of special characteristics of patients that usually was treated as the chie y or second research aim (27,28).
In the present study, we used age, gender, BMI, FPG, SBP, DBP, HDL, LDL, family history of diabetes, smoking status, drinking status as the strati ed variables, of which only age, BMI and SBP were found that they could strengthen or weaken the association between ALT/AST and incident diabetes. The possible reason for the above results is worth further exploration. Maybe the age, BMI or SBP is the intermediate factor between ALT/AST and incident diabetes in Chinese adults that they modi ed the effect of ALT/AST ratio on incident diabetes or the different levels of themselves could have imparity effect on ALT/AST ratio or incident diabetes. For example, a study which enrolled 1732 young adults with (n = 287) and without (n = 1445) insulin resistance indicated that insulin resistance is related to elevated transaminases and low AST/ALT ratio in young adults with normal weight (29). The clinical value of this study is as follows: (1) As far as we know, it is the rst time to reveal the association of ALT/AST ratio and incident diabetes in Chinese adults; (2) The results of the present article would contribute to further study on the structure of diagnostic or predictive models of incident diabetes.
There were several strengths in the present study. In the rst place, compared with previous similar studies, the sample size of our study is relatively bigger. What's more, we assessed the linearity during research. Besides, there would be prone to confusion as the nature of observational study. For this reason, rigorous adjusting of the component confounders was conducted for reducing confusion.
Moreover, in order to decrease the natural event during analysis and strengthen the characteristic of outcome, ALT/AST was conducted as not only continuous variable but also categorical variable. Last but not least, the analysis of the modi cation factors could make sure the results more typical in various subgroups.
Nonetheless, some limitations deserved to be noted. To begin with, the data we analyzed have already been examined by Chen et al, our study was a second retrospective cohort study which was conducted by the Rich Healthcare Group (10). Owing to the con ned raw data, our ndings might be not generalizable to other groups. In the meantime, the present research was a second retrospective cohort study based on published data, so the variable which not consisted in the data set was unable to be adjusted. Furthermore, as ALT/AST was measured at baseline, the changes during following up are not obsessed in the present study. Additionally, even if an extensive set of confounding factors have been adjusted, but the in uence of the residual confounding factors which because of the measuring errors in the categorization or the unmeasured factors such as dietary habits or physical exercise can't be obviated.
Consequently, a further investigation in a longer following up with more fastidious regulations is required.

Conclusions
In Chinese adults, ALT/AST ratio is an independent risk factor for incident diabetes. The population with age (30 to < 40 years), BMI (< 24 kg/m2) and SBP (< 140 mmHg) was found to be the stronger association. On the contrary, the population with age (20 to < 30 years, > 50 years), BMI (≥ 24 kg/m 2 ) and SBP (≥ 140 mmHg) was found to be the weaker association.