Source of data

Nine trials were (predominantly) published as journal articles (ABC 2017; ACE 2017; DAISI 2008; Fang 2004; Kawamori 2009; Koyasu 2010; STOP‐NIDDM 2002; Wang 2000; Yun 2016). For STOP‐NIDDM 2002, we also considered additional data that we received from the trial authors in reply to specific questions from us, the STOP‐NIDDM website, additional PowerPoint presentations, and debate articles as a result of the main publications. For DAISI 2008, we also used the statistical report. For the ABC 2017 and ACE 2017, we used additional data that we received from the trial authors. One trial was published on a website and as abstracts only (EDIT 1997).

Most trials were two‐arm trials (ABC 2017; ACE 2017; DAISI 2008; Kawamori 2009; Koyasu 2010; STOP‐NIDDM 2002; Wang 2000; Yun 2016). Two trials had multiple trial arms (EDIT 1997; Fang 2004).

Comparisons

Three trials compared the AGI acarbose to placebo (ACE 2017; DAISI 2008; STOP‐NIDDM 2002). A further three trials compared acarbose to no intervention (Koyasu 2010; Wang 2000; Yun 2016). One trial compared acarbose to three other groups, namely placebo, metformin, and acarbose plus metformin (EDIT 1997). Another trial compared acarbose to three control groups, namely no intervention, metformin, and diet plus exercise (Fang 2004). Finally, one trial compared the AGI voglibose to placebo (Kawamori 2009) and one trial compared voglibose to diet and exercise (ABC 2017). We did not find any trials investigating miglitol.

Overview of trial populations

The 10 trials randomised 11,814 participants (Table 1). Of these, 5721 were randomised to an intervention group and 6093 to a comparator group. The percentage of participants that finished the trial in the intervention groups ranged between 50% and 99%. In the comparator groups, this ranged between 62% and 100%. Individual trial sample size ranged from 61 to 6522 participants.

Trial design

One trial had a 2x2 factorial design (EDIT 1997). The participants were first randomised to either acarbose or a matching placebo, and then also randomised to either metformin or a matching placebo. This resulted in four groups: acarbose plus metformin, acarbose plus placebo, metformin plus placebo, and placebo plus placebo. The other nine trials had a parallel design. All 10 trials had a superiority design.

Five trials compared the intervention to placebo (ACE 2017; DAISI 2008; EDIT 1997; Kawamori 2009; STOP‐NIDDM 2002). Two trials compared the intervention to metformin (EDIT 1997; Fang 2004). Another two trials compared the intervention to diet and exercise (ABC 2017; Fang 2004). Four trials compared the intervention to no intervention (Fang 2004; Koyasu 2010; Wang 2000; Yun 2016).

Six trials were multicentre trials, with the number of centres raging from 2 to 176 (ABC 2017; ACE 2017; EDIT 1997; Kawamori 2009; STOP‐NIDDM 2002; Yun 2016).

Five trials were double‐blinded for participants and personnel (ACE 2017; DAISI 2008; EDIT 1997; Kawamori 2009; STOP‐NIDDM 2002). Four trials were not blinded (ABC 2017; Koyasu 2010; Wang 2000; Yun 2016). One trial did not describe the method of blinding (Fang 2004).

Five trials blinded the outcome assessors (ABC 2017; ACE 2017; Koyasu 2010; DAISI 2008; STOP‐NIDDM 2002). It was unclear whether the other trials blinded outcome assessors.

The trials were performed between 1995 and 2016. The duration of the follow‐up ranged from one to six years. In two trials the intervention period was followed by a wash‐out period. This was three months for STOP‐NIDDM 2002 and four weeks for Kawamori 2009. In all other trials the intervention period and follow‐up period were identical.

Four trials had a run‐in period (ACE 2017; DAISI 2008; Kawamori 2009; STOP‐NIDDM 2002). One trial was terminated earlier than planned because the interim analysis already showed the efficacy of the intervention (Kawamori 2009). Another trial was terminated early because the interim futility analysis showed an absence of beneficial treatment effect of voglibose compared to the control (ABC 2017).

Settings

STOP‐NIDDM 2002 and Kawamori 2009, mainly recruited participants through screening of high‐risk people (particularly first‐degree relatives of people with diabetes), and followed them up as outpatients in trial centres.

DAISI 2008 recruited participants from the population register of the city of Hoorn and followed them up as outpatients in the trial centre. Koyasu 2010, recruited participants from patients admitted to the hospital for elective coronary angiography and followed them up as outpatients. Yun 2016 recruited participants from patients who were hospitalised for acute coronary syndrome and followed them up as outpatients. ACE 2017 invited patients at cardiovascular and endocrinology centres with coronary heart disease to attend a screening. In EDIT 1997, the participants were ‘self‐referred’ but the exact setting is unclear. Setting and recruitment for the other three trials are also unclear (ABC 2017; Fang 2004; Wang 2000).

Participants

Four trials were performed in China (ACE 2017; Fang 2004; Wang 2000; Yun 2016), three in Japan (ABC 2017; Kawamori 2009; Koyasu 2010), one in the Netherlands (DAISI 2008), one in the UK (EDIT 1997), and one in multiple countries, namely Canada, Germany, Austria, the Nordic countries, Israel, and Spain (STOP‐NIDDM 2002).

Three trials reported the ethnicity of the participants. In two of these trials, the participants were almost exclusively white (EDIT 1997; STOP‐NIDDM 2002). In the other trial, the participants were almost entirely Han Chinese (ACE 2017). The percentage of female participants ranged from 8% to 52%. The mean age ranged from 47 years to 66 years. At baseline, the mean HbA1c ranged from 5.4% to 7.5%, the mean fasting plasma glucose (FPG) ranged from 5.5 to 6.6 mmol/L, and the mean two‐hour post glucose (2hPG) ranged from 6.3 to 10.7 mmol/L. Lastly, the mean BMI at baseline ranged from 21 to 31 kg/m².

Five trials reported both comorbidities and comedications (ABC 2017; ACE 2017; Koyasu 2010; STOP‐NIDDM 2002; Yun 2016). One trial reported only comorbidities (Kawamori 2009). The most commonly reported comorbidities were hypertension and previous myocardial infarction. The most commonly used medications were beta‐blockers, angiotensin‐converting enzyme (ACE) inhibitors, calcium channel blockers, and statins. In the ABC 2017 trial, considerably more people in the voglibose group used anti‐platelet therapy at baseline than in the control group (96.5% and 92.5%, respectively).

In most trials, IGT was newly diagnosed (ACE 2017; Kawamori 2009; Koyasu 2010; STOP‐NIDDM 2002; Wang 2000; Yun 2016). In three trials, this was not clearly stated (ABC 2017; EDIT 1997; Fang 2004).

The most common exclusion criterion was liver and/or kidney impairment (ABC 2017; ACE 2017; DAISI 2008; Koyasu 2010; STOP‐NIDDM 2002; Yun 2016). Four trials excluded people with gastrointestinal problems (ACE 2017; DAISI 2008; Koyasu 2010; Yun 2016). Three trials excluded people with a history of diabetes (ACE 2017; Koyasu 2010; Yun 2016). Furthermore, three trials excluded women who were pregnant or planned to become pregnant, and people with an intolerance or sensitivity to AGI (ACE 2017; DAISI 2008; Koyasu 2010). Finally, five trials excluded participants with a recent cardiovascular event (ABC 2017; ACE 2017; DAISI 2008; Koyasu 2010; STOP‐NIDDM 2002). There were also four trials that specifically included participants with a history of cardiovascular disease (CVD; ABC 2017; ACE 2017; Yun 2016; Koyasu 2010). This includes a history of myocardial infarction (ABC 2017), acute coronary syndrome, myocardial infarction, or (un)stable angina pectoris (ACE 2017), coronary artery disease or stable angina pectoris (Koyasu 2010), and acute coronary syndrome (Yun 2016). Three of these trials also excluded participants with a recent cardiovascular event (ABC 2017; ACE 2017; Koyasu 2010), as these trials wanted to include participants with a history of CVD, but who had not had an event recently. Because certain trials excluded participants with a history of CVD and other trials specifically included participants with a history of CVD, there is a different baseline risk for CVD between these trials.

EDIT 1997 intended to include people ‘at risk of developing diabetes’ with a FPG of 5.5 to 7.7 mmol/L. Based on the HbA1c and 2hPG baseline values, 36.9% of the participants had normal glucose tolerance, 8.1% had IFG, 25.5% had IGT, 14% had both IFG and IGT, and 15.5% had T2DM.

Diagnostic criteria

Four trials used the WHO 1985 criteria (WHO 1985), for IGT (a FPG of < 7.8 mmol/L and a 2hPG of 7.8 to 11.1 mmol/L; DAISI 2008; Fang 2004; STOP‐NIDDM 2002; Wang 2000). However, DAISI 2008 increased the lower limit of the 2hPG measurement to 8.6 mmol/L, because of the higher incidence of conversion to diabetes. And STOP‐NIDDM 2002 added a lower limit to the criteria of FPG, namely 5.6 mmol/L.

In 1998, the WHO changed the criteria for IGT to a FPG of less than 7.0 mmol/L and a 2hPG of 7.8 to 11.1 mmol/L (WHO 1998). Five trials included participants with IGT according to these criteria (ABC 2017; ACE 2017; Kawamori 2009; Koyasu 2010; Yun 2016). But again, some trials altered the criteria. Two trials lowered the FPG; Kawamori 2009 lowered it to 6.9 mmol/L and Yun 2016 lowered it to 6.1 mmol/L.

We excluded trials with participants that had T2DM. However, Koyasu 2010 also included participants with ‘mild T2DM’, which they defined as a FPG of less than 7.0 mmol/L, a 2hPG of more than 11.1 mmol/L, and a HbA1c of less than 6.5%. The current criteria for T2DM are a FPG of at least 7.0 mmol/L , or a 2hPG of at least 11.1 mmol/L (WHO/IDF 2006), or a HbA1c of at least 6.5% (ADA 2010). The ‘mild T2DM’ participants had a lower FPG and HbA1c than needed for a T2DM diagnosis, but did have a 2hPG of 11.1 or higher mmol/L. Therefore, the participants could be said to have T2DM. However, as the average 2hPG of the entire trial population was around 10.7 mmol/L in the acarbose group and 10.4 mmol/L in the control group, which is below the 11.1 mmol/L needed for T2DM diagnosis, and the participants had a FPG of 7.0 mmol/L or less and HbA1c of 6.5% or less, we decided to include this trial.

EDIT 1997 included participants at increased risk for T2DM with a FPG of 5.5 to 7.7 mmol/L.

Interventions

The alpha‐glucosidase inhibitor interventions were all administered orally.

In two trials, the acarbose dose started at 25 mg three times a day and increased to 50 mg three times a day during the first weeks (Fang 2004; Yun 2016). In four trials, the acarbose dose was 50 mg three times a day from the start (ACE 2017; EDIT 1997; Koyasu 2010; Wang 2000). In DAISI 2008, the acarbose dose started at 50 mg once daily and increased to 50 mg three times a day. In STOP‐NIDDM 2002, the acarbose dose started at 50 mg three times a day and ended at 100 mg three times a day or the maximum tolerated dose.

Finally, the two trials that investigated the effect of voglibose used a dose of 0.2 mg three times a day (ABC 2017; Kawamori 2009). ABC 2017 reduced the dose to a quarter or half of the original dose in case of gastrointestinal adverse effects.

In three trials, the participants did not take the trial medication on the days the glycaemic tests were performed (ACE 2017; DAISI 2008; Fang 2004). In one trial, the trialists describe that at the end of the intervention period the participants in the acarbose group took the final oral glucose tolerance test (OGTT) after a week‐long wash‐out period (Wang 2000). The other trials did not explicitly state if the participants were on trial medication or not when they performed the glycaemic tests (ABC 2017; EDIT 1997; Kawamori 2009; Koyasu 2010; STOP‐NIDDM 2002; Yun 2016). However, for the FPG test the participants would have to have been fasting, and because the medications have to be taken with a meal, it can be assumed that they would not have taken their medication for these tests. Additionally, the medication would not have had an effect on the 2hPG OGTT tests. Alpha‐glucosidase inhibitors inhibit an enzyme that breaks down complex carbohydrates, but the OGTT test uses glucose, a monosaccharide. The medication would only have had an effect if any of the trials had used a full‐meal tolerance test, but this is not the case.

Outcomes

Six trials explicitly stated a primary and secondary outcome in the publication (ABC 2017; ACE 2017; DAISI 2008; Kawamori 2009; Koyasu 2010; STOP‐NIDDM 2002). The most commonly defined primary outcome was the development of T2DM. ACE 2017 defined T2DM as two successive plasma glucose values of a FPG of 7.0 mmol/L, or more, or 2hPG of 11.1 mmol/L or more. Kawamori 2009 defined it as an HbA1c level of 6.5% or more and, on two separate occasions, a 2hPG of 11.1 mmol/L, or more, a FPG of 7.0 mmol/L or more, or a random plasma glucose concentration of 11.1 mmol/L or more. DAISI 2008, Fang 2004 and Wang 2000 defined T2DM as a FPG of 7.8 mmol/L or more, or a 2hPG of 11.1 mmol/L or more, or both. Finally, STOP‐NIDDM 2002 defined T2DM as a 2hPG of 11.1 mmol/L or more.

Trial registers or similar documents were not available for a number of trials (EDIT 1997; Fang 2004; Wang 2000; Yun 2016). For DAISI 2008 and Kawamori 2009, there were no differences between the trial registers and the publications. However, for STOP‐NIDDM 2002, the definition of the outcome cardiovascular events was different in the final publication compared with the trial register information. In ABC 2017, the outcome progression from IGT to T2DM, which was mentioned in the trial register, was not reported in the final publication. The trial register for Koyasu 2010 mentions both the incidence of cardiovascular events and carotid intima‐medial thickness (IMT) as primary outcome, but the publication only mentions the change in IMT.

Six trials reported all‐cause mortality (ABC 2017; ACE 2017; DAISI 2008; Kawamori 2009; Koyasu 2010; STOP‐NIDDM 2002). Four trials reported cardiovascular mortality, non‐fatal myocardial infarction, non‐fatal stroke, and congestive heart failure (ABC 2017; Koyasu 2010; STOP‐NIDDM 2002; Yun 2016). Three trials reported serious adverse events (ACE 2017; DAISI 2008; Kawamori 2009). Four trials reported non‐serious adverse events (DAISI 2008; Kawamori 2009; STOP‐NIDDM 2002; Wang 2000).

Of the secondary outcomes, seven trials reported measures of blood glucose control and lipids (ABC 2017; ACE 2017; DAISI 2008; Fang 2004; Koyasu 2010; STOP‐NIDDM 2002; Yun 2016). Three trials reported body weight (ACE 2017; Koyasu 2010; STOP‐NIDDM 2002), and five listed reported BMI (ACE 2017; Fang 2004; Koyasu 2010; STOP‐NIDDM 2002; Yun 2016). Six trials reported data on blood pressure (ABC 2017; ACE 2017; Fang 2004; Koyasu 2010; STOP‐NIDDM 2002; Yun 2016). Finally, one trial reported hypoglycaemic episodes (ACE 2017). EDIT 1997 reported that they had investigated a number of outcomes (e.g. plasma glucose and lipids), but these data have not been accessible to us thus far. We acquired some of the data on measures of blood glucose control, lipids, body weight, BMI, and blood pressure through contact with the trial authors (ABC 2017; ACE 2017; STOP‐NIDDM 2002), or from the trial protocol (DAISI 2008).

Subgroup analyses within the included trials

The subgroup analyses in ACE 2017 looked at sex (male/female), region in China (Beijing and Tainjin/Central/South and Southwest/West and East/Northeast/Hong Kong), coronary heart disease inclusion criteria (previous myocardial infarction/previous unstable angina/current stable angina/more than 1 of the above), history of heart failure (yes/no), age (≤ 63.5/> 63.5), HbA1c (≤ 5.9/> 5.9), FPG (≤ 5.47 mmol/L/> 5,47 mmol/L), 2hPG (≤ 9.12 mmol/L/> 9.12 mmol/L), systolic blood pressure (≤ 130 mmHg/> 130 mmHg), BMI (≤ 25 kg/m²/> 25 kg/m²), and estimated glomerular filtration rate (≤ 88.5 mL/min/1.73 m²/> 25.0 mL/min/1.73 m²). They found a HR of 0.70 (95% CI 0.50 to 0.99) for the 5‐point composite outcome for the subgroup of the northeast region of China. There were no statistically significant effects found in the other subgroups.

The subgroups that ABC 2017 analysed were sex (male/female), age (< 65/≥ 65), BMI (< 25 kg/m²/≥ 25 kg/m²), hypertension (no/yes), dyslipidaemia (no/yes), smoking (no/yes), arteriosclerosis obliterans (no/yes), and OGTT (no/yes). The only factor that statistically significantly affected the risk of cardiovascular events was age. The group of less than 65 years had statistically significantly fewer events in the control group than in the voglibose group (logrank P = 0.02).

In Kawamori 2009, the analysed subgroups were age (10‐year increase), sex (male/female), BMI (5 kg/m² increase), dyslipidaemia (yes/no), hypertension (yes/no), family history of diabetes (yes/no), 2hPG (0.55 mmol/L increase), insulinogenic index (0.2 decrease), homoeostasis model assessment for insulin resistance (1 increase), smoker (yes/no), intensity of daily activity (I/III to IV), intensity of daily activity (II/III to IV), and concomitant use of ACE inhibitor or angiotensin II receptor blocker (yes/no). According to these subgroup analyses, people with a higher BMI (HR 1.495, P = 0.0023), people with a lower insulinogenic index (HR 1.246, P < 0.0001), and people who had a higher intensity of daily activity (HR 1.992, P < 0.0001), had a higher risk of developing T2DM.

In STOP‐NIDDM 2002, the (univariate) subgroup analyses included FPG, 2hPG, fasting insulin, 2‐hour insulin, HbA1c, total cholesterol, LDL, HDL, total triglycerides, systolic blood pressure, diastolic blood pressure, weight, BMI, waist circumference, concomitant medications, and current smoker status. People with a higher FPG (HR 1.767, P = 0.03), people with a higher systolic blood pressure (HR 1.029, P < 0.001), people with a higher diastolic blood pressure (HR 1.043, P = 0.006), and people with concomitant medications (HR 2.071, P = 0.02) had a higher risk of cardiovascular events. They also performed multivariate analyses on the outcomes FPG (HR 1.830, P = 0.03) and systolic blood pressure (HR 1.031, P < 0.001).

Primary outcomes

Secondary outcomes

Adjustment for high discontinuation rate in the acarbose arm of STOP‐NIDDM 2002

The discontinuation rate in the acarbose group of STOP‐NIDDM 2002 was higher than in the placebo group (acarbose 31% versus placebo 19%). Despite the fact that discontinuing participants remained in the ITT analyses, it is possible that those participants were not followed up regularly every third month, and thus possible occurrence of T2DM or a cardiovascular event was less likely to be discovered (Sawicki 2004).

In order to investigate the possible influence of differences in the frequency of follow‐up, we re‐analysed the data with the following adjustments: first we requested the mean number of trial visits for both treatment groups. Next, we divided the number of visits of the placebo group by the number of visits in the acarbose group. We used this outcome as a correction factor for the number of events in the acarbose group (occurrence of cardiovascular morbidity and T2DM).

The authors of STOP‐NIDDM 2002 reported that the mean numbers of trial visits in the acarbose group (N = 682) and in the placebo group (N = 686) were 13.3 (SD = 5.4) and 14.6 (SD = 4.3), respectively. The calculated correction factor was: 14.6/13.3 = 1.1. The outcomes before and after this correction factor are listed in Appendix 19. The effects sizes for incidence of T2DM and occurrence of any cardiovascular disease became smaller after the correction but remained statistically significant. The effects size for myocardial infarctions did not change (due to the fact that there was only one case in the acarbose group). The other outcomes remained statistically not significant after correction.

Primary outcomes

Secondary outcomes

Primary outcomes

Secondary outcomes

Primary outcomes

Secondary outcomes