Alpha-glucosidase inhibitors and hepatotoxicity in type 2 diabetes: a systematic review and meta-analysis

Alpha-glucosidase inhibitors (AGIs) was reported to be associated with several rare adverse hepatic events, but with inconsistent results. We aimed to investigate the risk of hepatotoxicity associated with the use of AGIs in patients with type 2 diabetes mellitus (T2DM), and performed a systematic review and meta-analysis. Fourteen studies (n = 2881) were eligible, all of which were RCTs. Meta-analysis of data regarding elevation of more than 3-fold the upper limit of normal (ULN) of AST and ALT showed statistically significant differences between AGIs treatment versus control (OR 6.86, 95% CI 2.50 to 18.80; OR 6.48, 95% CI 2.40 to 17.49). Subgroup analyses of elevation of more than 1.8-fold ULN of AST and ALT by dose of AGIs showed differential effects on AST and ALT (AST: OR 0.38 vs 7.31, interaction P = 0.003; ALT: OR 0.32 vs 4.55, interaction p = 0.02). Meta-analysis showed that AGIs might increase the risk of hepatotoxicity, and higher dose appeared to be associated with higher risk of hepatotoxicity. However, the evidence is limited with surrogate measures (i.e. ALT and AST), and no clinically important adverse events were observed.

High-grade AST elevation. Six trials 13,15,39,[43][44][45] reported 17 cases elevations of more than 1.8-fold ULN of AST levels in 1505 patients who used at least one medication (raw event rate 1.0%). Pooling of these trials showed no statistically significant difference in the risk of elevations more than 1.8-fold ULN of AST levels between AGIs treatment and control (Peto OR: 2.12, 95% CI 0.80 to 5.60; I-square = 51%). The subgroup analyses by type of AGIs agent, type of control, length of follow-up and mode of treatment suggested no apparent differences, but differential effects were present among the varying doses of AGIs (AGIs ≤ 100 mg t.i.d. (Peto OR: 0.32, 95%CI 0.05 to 1.97) vs > 100 mg t.i.d. (Peto OR: 7.31, 95%CI 2.05 to 26.08); interaction P = 0.003) (Fig. 2). The sensitivity analysis using alternative effect measures (OR vs RR), analysis models (random vs fixed) and pooling methods (Peto vs. Mantel-Hanszel method) did not show important changes in the pooled effects.
Seven trials [13][14][15][16][17][18]44 reported 16 cases of elevation of more than 3-fold ULN of AST levels occurred in 1603 patients who used at least one medication (raw event rate 1.0%). Pooling of these trials showed an increased risk of elevations more than three times ULN of AST levels in patients taking AGIs versus control (Peto OR: 6.86, 95% CI 2.50 to 18.80; I-square = 0%) (Fig. 3). The subgroup analyses showed no any significant differences. The sensitivity analysis did not show important changes in pooled effects. In one of sensitivity analyses, we removed studies with potential overlap of study population across publications (e.g. Chniff 1995a and Chniff 1995b); the pooled estimates showed no significant change.
High-grade ALT elevation. Seven trials 13,14,39,41,[43][44][45] reported 18 cases elevations more than 1.8 times ULN of ALT levels occurred in 1601 patients who used at least one medication (raw event rate 1.1%). Pooling of these trials showed no statistically significant difference in the risk of elevations more than 1.8 times ULN of ALT levels between AGIs treatment and control (Peto OR: 2.10, 95% CI 0.79 to 5.61; I-square = 51%). The subgroup analysis by dose of AGIs ( AGIs ≤ 100 mg t.i.d. vs > 100 mg t.i.d.) showed a relatively apparent differential effects (interaction P = 0.02, Peto OR: 0.32 (0.05 to 1.97) vs 4.55 (1.42 to 14.58) (Fig. 4). The sensitivity analysis did not show important changes in the pooled effects.
Seven trials [13][14][15][16][17][18]44 reported 17 cases elevations more than three times ULN of ALT levels occurred in 1611 patients who used at least one medication (raw event rate 1.0%). Pooling of these trials showed an increase in the risk of elevations more than three times ULN of ALT levels between AGIs treatment and control (Peto OR: 6.48, 95% CI 2.40 to 17.49; I-square = 0%) (Fig. 5). The subgroup analyses did not show any significant differences. The sensitivity analysis did not show important changes in the pooled effects, including the one analysis by removing studies with potential overlap of study population across publications (e.g. Chniff 1995a and Chniff 1995b).

Author(year)
International study

Discussion
In this study, we demonstrated the risk of hepatic AEs associated with AGIs use. Overall, the risk of developing hepatic AEs cannot be determined drawing from the 14 randomized trials, given the relatively small doses of   acarbose ( ≤ 100 mg t.i.d), relatively small sample sizes and short follow-up. However, we found evidence showing increased risk of liver transaminases (AST and ALT) associated with the use of AGIs in patients with type 2 diabetes. In our study, there was a significant increase in odds of hepatotoxicity due to the elevation of AST and ALT. The odds of elevations greater than 3.0-fold ULN of AST and ALT levels were 6.86 and 6.48 times higher in patients receiving AGIs compared to patients in the control arms. For the pooling effects of elevations 1.8-fold ULN of AST and ALT, the results did not show a significant difference, but the subgroup analysis by dose of AGIs showed an apparent difference, which probably suggested that the risk of hepatotoxicity would be greatly increased when patients using AGIs more than 100 mg t.i.d. In addition, the study of Iwamoto 18 , which intervention was voglibose, showed no any changes in hepatic enzymes abnormalities ( ≥ 3-fold ULN range) of AST and ALT. In addition, some studies 4,13-17 reported that these elevations were asymptomatic and became normalized after discontinuation of the study medication. Combined the result of subgroup analysis by dose of AGIs using, it probably indicated that there exists dose-response relationship between the hepatotoxicity and AGIs exposure dose.
Little research was available to explore the hepatotoxicity induced by AGIs. One review 7 about hepatotoxicity have been published in 2007. As part of this review regarding safety profile of acarbose, the content just described that acarbose may lead to hepatotoxicity, while not performing a rigorous and thorough analysis to evaluate the relationship between acarbose and risk of hepatotoxicity. In our study, we also searched the FDA Adverse Event Reporting System 46 . The search yielded several cases of fatal hepatitis event in those using acarbose; however, additional information was unbailable. We also searched the US LiverTox website 47 , and found no any information regarding voglibose and absence of clinical acute liver injury in those using miglitol. The website reported a case of acute liver cell damage in a patient administered with acarbose. Nevertheless, all of above information was based on case reports, which was unable to offer causality.
Our study has a few strengthens. First, it offers an up-to-date and complete overview of randomized controlled trials and observational studies concerning AGIs treatment. Though we included a number of RCTs, the data are relatively consistent and heterogeneity is acceptable. In addition to published reports, we searched ClinicalTrials. gov for completeness of data 48 .In the meanwhile, one should interpret the findings cautiously because of limitations. First, many patients were withdrawal from these trials, possibly because of the side effect of gastrointestinal tract caused by AGIs, for which we were unable to account for 7 . Second, the reporting of hepatic AEs was lacking in many studies, leading to their exclusion from analysis. Adverse events, unlike efficacy outcomes, are rarely predetermined for systematic data collection in clinical trials. Therefore, reporting of adverse events depends highly on the investigators. Third, the type of reported hepatic AEs were highly variable, which made the collection and analysis of data challenging. Last, some studies [14][15][16] have the potential overlap of study population across publications. We try to contact the author, but we can't get the contact information. Meanwhile, we have carefully checked the studies for potential overlap of study population across publications, and found that the study designs of these articles differed with each other (including total number of patients, number of groups, tested group, control group, and usage of medication). We also conducted a sensitivity analysis by removing these studies with potential overlap participants; the pooled estimates showed no significant changes, suggesting robustness of results to this potential issue.
In conclusion, our meta-analysis suggested that patients taking acarbose could have higher risk of liver damage compared to patients without AGIs. Although not definitive, the findings may suggest caution in the use of AGIs for those who are at high risk for hepatic dysfunction. In summary, although the effects of AGIs on hepatic AEs remain uncertain, the randomized evidence consistently suggests increased risk of liver enzymes elevation in general. Dose-response relationship may exist between the hepatotoxicity and AGIs dose.

Methods
Eligibility criteria. We included randomized controlled trials (RCTs) and observational studies (cohort studies and case-control studies) of patients with type 2 diabetes mellitus without any liver disease or abnormal liver transaminase that compared alpha-glucosidase inhibitors (acarbose, miglitol, voglibose) with placebo, lifestyle modification, or active antidiabetic agents. An eligible study should also follow up patients for at least 12 weeks (not applicable to case-control studies), and explicitly report outcome data regarding any hepatic AEs (e.g. hepatitis, death, liver transplantation, hospitalization for hepatotoxicity or withdrawal due to any liver damage), or highgrade alanine transaminase (ALT) and aspartate transaminase (AST). High-grade ALT and AST elevations were defined an elevation of more than 1.8-fold of the upper limit of normal (ULN) [13][14][15][16]39,42,49,50 . These liver transaminases (AST or ALT) are useful biomarkers of liver injury in a patient with some degree of intact liver function 50-52 . Literature search. We searched Medline, Embase, and the Cochrane Central Register of Controlled Trials (CENTRAL) for reports published in English language from inception to July 2015. We combined both Medical Subject Headings (MeSH) and free text terms for identifying relevant articles. An information expert (JJY) helped develop the search strategy (Appendix 1). In our search, we included search terms defining AGIs and T2DM only, as we had planned to evaluate all the potential adverse events of AGIs, including -but not limited to -hepatotoxicity. We also searched ClinicalTrials.gov for additional information, which provides important data on hepatotoxicity.
Study process. Two reviewers, trained in health research methods, independently screened titles/abstracts and full texts for eligibility, assessed risk of bias, and collected data from each eligible study, using standardized, pilot-tested forms, together with detailed instructions. Reviewers resolved disagreement through discussion or, if required, adjudication by a third reviewer.

Risk of bias assessment.
We assessed the risk of bias of RCTs using the Cochrane Collaboration's tool 53 .
The items included random sequence generation, allocation concealment, blinding of participants, caregivers, and assessors of outcomes (i.e. hepatitis or changes in liver enzymes), incomplete outcome data, prognostic balance between treatment groups, selective reporting. We planned to assessed the risk of bias of observational studies using the Newcastle-Ottawa Quality Assessment Scale 54 .
Data collection. We collected the following information from each study: study characteristics (authors' name, year of publication, total number of patients randomization, number of treatment groups, length of follow-up, funding source, countries involved, and number of study sites); patient characteristics (gender, age, diabetes duration, body mass index (BMI), baseline HbA1c level, and fasting plasma glucose); interventions (medications common to all groups (baseline treatment, details of AGIs therapy and control group); and outcomes (any hepatic adverse events occurred during the course of study). For RCTs, if the initial treatment assignment was switched (e.g. patients in placebo group started receiving AGIs agents after 24 weeks), we collected the data prior to that point. If a trial had multiple reports, we collated all data into a single study 55 . If the outcome data were reported in multiple follow up points, we used data with the longest follow-up. Data analysis. We planned to analyze RCTs and observational studies separately. However, no observational studies were eligible. For randomized trials, we pooled the data using Peto's methods because of the very low event rate 55 , and reported pooled Peto ORs and associated 95% CIs. We examined the heterogeneity among studies by the Cochran chi-square test and the I-squared statistic. We explored sources of heterogeneity with three priori subgroup hypotheses: type of AGIs agent (acarbose; miglitol and voglibose); type of control (AGIs vs placebo, AGIs vs active treatment); dose of AGIs using (AGIs ≤ 100 mg t.i.d. and > 100 mg t.i.d.); length of follow-up ( ≤ 26 weeks and > 26weeks) and mode of treatment (AGIs monotherapy, AGIs add-on/combination treatment). We carried out sensitivity analyses by using alternative effect measures (odds ratio (OR) vs. risk ratio (RR)), pooling methods (Peto vs. Mantel-Hanszel method), statistical models regarding heterogeneity (random vs. fixed effects), and removal of studies with potential overlap of study populations across publications.
We planned to examine publication bias by the funnel plot or other methods (such as Egger's and Begg's). Because of the low power of test associated with studies of low events rate, we were unable to examine this. We reported the results according to preferred reporting items for systematic reviews and meta-analyses (PRISMA) 56 .