A total of 47 published studies met the selection criteria and 6 of which were further excluded from the analysis due to absence of data on cardiovascular events. Five studies that included two comparator arms were treated as two separate trials, and one study that includes three comparator arms was treated as three separate trials. Eight studies from the clinicaltrial.gov website were included. Two studies that included two comparator arms were treated as two separate trials each. Therefore, 58 trials were included in the final analysis (Fig. 1). The characteristics of the included trials are summarized in Table S1 (online only). Of the 58 trials, a total of 17 604 patients were included, where 10 466 patients received GLP-1 receptor agonists, whereas 7138 patients received other comparators. Exenatide and liraglutide were studied in 33 and 18 studies, respectively. Of the 58 trials, there were 29 placebo-controlled trials and 29 active comparator trials.
According to JADAD scale, the methods used for randomization and allocation concealment were not clearly stated in some instances (74% and 52%, respectively). Conversely, the number of withdrawals and methods used for blinding were appropriately described in all cases (98% and 100%, respectively). There were 24 trials with low bias (Table S2; online only).
Risk of cardiovascular events
Table S2 lists the cardiovascular events reported in the 58 included trials, with 49 events in the GLP-1 receptor agonist group and 42 events in the control groups.
As shown in Table 1, the overall OR for cardiovascular events associated with GLP-1 receptor agonists was 0·52 (95% confidence interval [CI], 0·27–0·99, P = 0·047) compared with placebo (N = 29). After the exclusion of short-term trials, the OR changed to 0·55 (95% CI: 0·20–1·52, P = 0·249) (N = 9). For the placebo-controlled trials (Table 1), the risk for cardiovascular events with GLP-1 receptor agonists was 0·51 (95% CI: 0·22–1·15, P = 0·105) in trials of short-term duration (N = 20), whereas the risk was 0·37 (95% CI:0·12–1·16,P = 0·088) and 2·62 (95% CI:0·24–29·13,P = 0·433) in trials with a medium-term duration (N = 7) and a long-term duration (N = 2), respectively. For trials with monotherapy (N = 5) and add-on therapy (N = 24), the ORs were 0·30 (95% CI:0·06–1·62,P = 0·163) and 0·57 (95% CI: 0·28–1·15, P = 0·116), respectively. The ORs, 95% CIs and P values of all subgroup analyses are listed in Table 1. Results showed no heterogeneity of effects across trials (P > 0·05).
Table 1. Risk of cardiovascular events in subgroup analyses of placebo-controlled trials
| ||No. of study(N)||Q-statistics value||Odds ratio (95% CI)||P value|
|Patterns of treatment|
Compared with placebo, TSA indicated that the heterogeneity-adjusted information size required to demonstrate or reject a 33% relative risk reduction by the low-bias trials was 65 212 patients. This is much more than the number of patients who were actually accrued (trial sequential analysis adjusted 95% CI: 0·04–7·21). None of the boundaries for benefit or harm was crossed, and futility boundaries could not be produced, indicating that there was too little evidence to conclude whether GLP-1 receptor agonists were beneficial, harmful or without any effect with respect to cardiovascular events when compared with placebo (Fig. 2).
Figure 2. Trial sequential analysis of cardiovascular events compared with placebo. Heterogeneity-adjusted required information size of 65, 212 participants calculated on the basis of proportion of cardiovascular events of 0·36% in control group, and relative risk reduction in 33%, α = 5%, β = 20%, and I2 = 0%. The actual accrued number of participants was 7445, 11% of the required information size. The full cumulative Z curve does not cross the trial sequential monitoring boundaries and futility boundaries for benefit or harm. Horizontal dotted lines illustrate the traditional level of statistical significance (P = 0·05).
Download figure to PowerPoint
When only active comparator trials were analysed (N = 29), the summary OR for cardiovascular events was 0·84 (95% CI: 0·52–1·36, P = 0·473) in the GLP-1 receptor agonists group. After the exclusion of short-term trials, the OR changed to 0·87 (95% CI: 0·50–1·49, P = 0·604) (N = 20). In subgroup analyses of the active comparator trials (Table 2), the risk associated with GLP-1 receptor agonists was 0·54 (95% CI: 0·22–1·32, P = 0·174) and 1·00 (95% CI: 0·57–1·77, P = 0·989) for trials with monotherapy (N = 10) and add-on therapy (N = 19), respectively. In trials of long duration (N = 6), the OR was 0·75 (95% CI: 0·37–1·53, P = 0·435). No heterogeneity effect was detected between these trials (P > 0·05).
Table 2. Risk of cardiovascular events in subgroup analyses of active comparator trials
| ||No. of study (N)||Q-statistics value||Odds ratio (95% CI)||P value|
|≥52 weeks||6||3·23||0·75 (0·37,1·53)||0·435|
|Patterns of treatment|
Trial sequential analysis showed a lack of sufficient evidence of benefit, harm or without any effect on cardiovascular events when compared with active controls (trial sequential analysis adjusted 95% CI: 0·12–5·95). Only 10, 157 (13%) of the required heterogeneity-adjusted 79, 198 patients required to detect a 22% relative risk reduction were actually accrued (Fig. 3).
Figure 3. Trial sequential analysis for cardiovascular events compared with active comparators. Heterogeneity-adjusted required information size of 79,198 participants calculated on the basis of proportion of cardiovascular events of 0·73% in control group, relative risk reduction of 22%, α = 5%, β = 20% and I2 = 0%. The actual accrued number of participants was 10 157, 13% of required information size. Full cumulative Z curve does not cross the trial sequential monitoring boundaries and futility boundaries for benefit or harm. Horizontal dotted lines illustrate the traditional level of statistical significance (P = 0·05).
Download figure to PowerPoint
Separate analyses for exenatide showed that the risk of cardiovascular events was 0·46 (95% CI: 0·19–1·10, P = 0·819) compared with placebo (N = 15) and 0·91 (95% CI: 0·43–1·93, P = 0·802) with active control (N = 18). For liraglutide, the OR was 0·58 (95% CI: 0·17–2·02, P = 0·390) and 0·79 (95% CI: 0·42–1·48, P = 0·464) for placebo-controlled (N = 8) and active comparator trials (N = 10), respectively.