The incidence of prediabetes reversion to normoglycemia

The analysis included nine trials with seven publications [26,27,28,29, 39, 42, 43] with a total of 2817 patients in the GLP-1RAs arm and 1348 patients in the control arm. Evidence with a low degree of certainty demonstrated that GLP-1RAs significantly revert prediabetic patients to a normoglycemia state compared to the control group [RR = 1.76, 95% CI (1.45, 2.13), P < 0.00001] (Fig. 3a). High heterogeneity was observed [I2 = 79%, P < 0.00001], which was partially resolved after the exclusion of Perreault et al. 2022c [I2 = 51%, P = 0.05] without significant change in the pooled analysis [RR = 1.83, 95% CI (1.57, 2.12), P < 0.00001].

Fig. 3

A forest plot shows the effect of GLP-1RAs on the incidence of prediabetes reversion to normoglycemic state. a Overall effect. b subgroup analysis based on the treatment duration

Subgroup analysis revealed that the effect of GLP-1RAs on prediabetes reversion was accomplished as early as 14–24 weeks [RR = 2.65, 95% CI (1.76, 3.99), P < 0.00001] and lasted up to 48–56 weeks [RR = 1.82, 95% CI (1.03, 3.21), P = 0.04] and 68–104 weeks [RR = 1.78, 95% CI (1.66, 1.91), P < 0.00001] (Fig. 3b). The pooled analysis was homogenous for 68–104 weeks [I2 = 0%, P = 0.57], but heterogeneous for 14–24 weeks [I2 = 53%, P = 0.06] and 48–56 weeks [I2 = 96%, P < 0.00001], which was resolved after the exclusion of Rosenstock et al.2010 [I2 = 31%, P = 0.22], and Perreault et al. 2022c [I2 = 44%, P = 0.18], respectively, without significant change in the pooled analysis [RR = 2.44, 95% CI (2.18, 2.74), P < 0.00001], [RR = 2.21, 95% CI (1.98, 2.46), P < 0.00001].

Subgroup analysis revealed that semaglutide 2.4mg once weekly (QW) significantly outperformed the control group [RR = 1.54, 95% CI (1.24, 1.91), P < 0.00001]. The pooled analysis showed heterogeneity [I2 = 85%, P = 0.001]. Heterogeneity was eliminated by excluding Perreault et al. 2022c without significant change in the pooled analysis [RR = 1.71, 95% CI (1.54, 1.90), P < 0.00001]. Furthermore, Liraglutide 1.8mg once daily (QD) and 3mg were significantly effective [RR = 5.68, 95% CI (2.52, 12.82), P < 0.00001], [RR = 1.85, 95% CI (1.67, 2.05), P < 0.00001], and the pooled analyses were homogenous [I2 = 33%, P = 0.22], [I2 = 52%, P = 0.15] (Figure S1).

The incidence of new-onset diabetes

The analysis included six trials with four publications [26,27,28,29] with a total of 2557 patients in the GLP-1RAs arm and 1258 patients in the control arm. Evidence with a moderate degree of certainty demonstrated that prediabetic patients who progressed to a diabetes state were significantly fewer in the GLP-1RAs group compared to the control [RR = 0.28, 95% CI (0.19, 0.43), P < 0.00001]. Semaglutide 2.4mg QW significantly outperformed the control group in the subgroup analysis [RR = 0.17, 95% CI (0.05, 0.52), P = 0.002]. Both pooled analyses were homogenous [I2 = 0%, P = 0.62], and [I2 = 0%, P = 0.99] (Fig. 4a).

Fig. 4

A forest plot shows the effect of GLP-1RAs on the incidence of new-onset diabetes. a Overall effect and subgroup analysis based on the type and dose of GLP-1RAs. b subgroup analysis based on the treatment duration

The protective effect of GLP-1RAs was noticed after 48–56 weeks and 68–104 weeks of the treatment [RR = 0.15, 95% CI (0.05, 0.44), P = 0.0005], [RR = 0.21, 95% CI (0.12, 0.38), P < 0.00001], but not as early as 14–24 weeks [RR = 0.44, 95% CI (0.17, 1.14), P = 0.09]. All the results were homogenous (P > 0.1) (Fig. 4b).

Secondary outcomesFPG (mg/dl)

The analysis included eight trials with 6 publications [26, 28, 39,40,41, 43] with a total of 2636 patients in the GLP-1RAs arm and 1297 patients in the control arm. Evidence with a high degree of certainty demonstrated that the GLP-1 group significantly decreased FPG compared to the control group [MD = -8.00, 95% CI (-8.76, -7.23), P < 0.00001]. The pooled analysis was homogenous (P > 0.1). The liraglutide 1.8mg and semaglutide 2.4mg QW significantly outperformed the control group in the subgroup analysis [MD = − 8.24, 95% CI (− 11.31, − 5.16), P < 0.00001], and [MD = − 8.47, 95% CI (− 9.73, − 7.22), P < 0.00001], respectively, and the pooled analysis was homogenous for both (P > 0.1) (Fig. 5a).

Fig. 5

A forest plot shows the overall effect and subgroup analysis based on the type and dose of GLP-1RAs on a FPG and b HbA1c

HbA1c (%)

The analysis included six trials with four publications [26, 28, 38, 39] with a total of 2568 patients in the GLP-1RAs arm and 1274 patients in the control arm. Evidence with a low degree of certainty demonstrated that the GLP-1 group significantly reduced HbA1c levels when compared to the control group [MD = − 0.29, 95% CI (− 0.36, − 0.23), P < 0.00001] (Fig. 5b). High heterogeneity was observed [I2 = 81%, P = 0.0001]. The heterogeneity was eliminated by excluding Le Roux et al. 2017 [26] [I2 = 0%, P = 0.74] without significant change in the pooled analysis [MD = − 0.32, 95% CI (− 0.35, − 0.29), P < 0.00001]. Semaglutide 2.4mg QW significantly outperformed the control group in the subgroup analysis [MD = − 0.32, 95% CI (− 0.35, − 0.28), P < 0.00001], and the pooled analysis was homogenous (P > 0.1) (Fig. 5b).

Weight loss (kg)

The analysis included eight trials with six publications [26, 28, 39,40,41, 43] with a total of 2636 patients in the GLP-1RAs arm and 1297 patients in the control arm. Evidence with a very low degree of certainty demonstrated that the GLP-1 group significantly reduced body weight when compared to the control group [MD = − 6.38, 95% CI (− 9.64, − 3.12), P = 0.0001] (Fig. 6a). There was significant heterogeneity [I2 = 98%, P < 0.00001], which could not be resolved with the statistical analysis. Semaglutide 2.4mg QW significantly outperformed the control group in the subgroup analysis [MD = − 11.60, 95% CI (− 12.95, − 10.26), P < 0.00001] with the pooled analysis being homogenous (P > 0.1) while, liraglutide 1.8mg showed no significant difference compared to the control group in the subgroup analysis [MD = -1.60, 95% CI (− 5.63, 2.42), P = 0.43] (Fig. 6a). The pooled analysis was heterogeneous (P < 0.00001), which was resolved by the exclusion of Mashayekhi et al. 2022 (P > 0.01), after which the pooled analysis significantly favored the liraglutide 1.8mg group [MD = − 3.52, 95% CI (− 4.58, − 2.47), P = 0.43].

Fig. 6

A forest plot shows the overall effect and subgroup analysis based on the type and dose of GLP-1RAs on a body weight and b waist circumference

Waist circumference (cm)

The analysis included four trials [26, 39, 41, 43] with a total of 1543 patients in the GLP-1RAs arm and 822 patients in the control arm. Evidence with a moderate degree of certainty demonstrated that t00he GLP-1 group significantly reduced waist circumference compared to the control group [MD = − 3.41, 95% CI (− 4.03, − 2.80), P < 0.00001]. The pooled analysis was homogenous (P > 0.1). The liraglutide 1.8mg group significantly outperformed the control group in the subgroup analysis [MD = − 2.95, 95% CI (− 4.46, − 1.44), P = 0.0001]. The pooled analysis was homogenous (P > 0.1) (Fig. 6b).

Lipid profile (mg/dl)

The analysis included five trials [26, 38, 39, 41, 43] with a total of 1564 patients in the GLP-1RAs arm and 843 patients in the control arm. The GLP-1 group significantly reduced TG and LDL compared to the control group [MD = − 9.28, 95% CI (− 12.77, − 5.78), P < 0.00001], [MD = − 3.21, 95% CI (− 5.29, − 1.13), P = 0.02] (Fig. 7a and b). The pooled analysis was homogeneous for both outcomes (P > 0.1). However, no significant difference was observed between the control group and the GLP-1 group regarding HDL [MD = 0.82, 95% CI (− 1.46, 3.10), P = 0.48] (Fig. 7c). The pooled analysis was heterogeneous (P = 0.02), which was resolved after the exclusion of Zhou et al. [38] (P > 0.1) without significant effect on the pooled analysis [MD = 0.27, 95% CI (− 0.65, 1.20), P = 0.56]. The evidence was of moderate, low, and very low degrees of certainty for TG, LDL, and HDL, respectively.

Fig. 7

A forest plot shows the overall effect and subgroup analysis based on the type and dose of GLP-1RAs on a TG, b LDL, and c HDL

Safety outcomes

Any gastrointestinal disorders, nausea, vomiting, and diarrhea adverse events were reported for prediabetic patients by three trials [26, 39, 43] with a total of 1635 patients in the GLP-1RAs arm and 804 patients in the control arm, and GLP-1RAs revealed a higher incidence of the aforementioned adverse events compared to the control [RR = 1.45, 95% CI (1.03, 2.03), P < 0.00001], [RR = 2.47, 95% CI (2.10, 2.91), P < 0.00001], [RR = 3.54, 95% CI (2.61, 4.79), P < 0.00001], [RR = 1.69, 95% CI (1.40, 2.03), P < 0.00001], respectively (Fig. 8b–e). All the pooled analyses were homogenous (P > 0.1) except for any gastrointestinal disorders, where significant heterogeneity was present [I2 = 67%, P = 0.05]. The heterogeneity was resolved by excluding Le Roux et al. 2017 [I2 = 0%, P = 0.99] without significant change in the pooled analysis [RR = 1.78, 95% CI (1.27, 2.48), P = 0.0008].

Fig. 8

A forest plot shows the overall effect of GLP-1RAs on a Any adverse events, b Any gastrointestinal disorders, c nausea, d vomiting, e diarrhea, f hypoglycemia, g serious adverse event, h headache

Any adverse events, hypoglycemia, and any serious adverse events were reported by two trials [26, 39] with a total of 1611 patients in the GLP-1RAs arm and 777 patients in the control arm, and GLP-1RAs revealed a higher incidence of Any adverse events and hypoglycemia [RR = 1.06, 95% CI (1.03, 1.09), P < 0.00001], [RR = 4.36, 95% CI (3.08, 6.16), P < 0.00001], while no significant difference was found regarding any serious adverse events [RR = 1.19, 95% CI (0.95, 1.48), P = 0.13] (Fig. 8a, f, and g). The pooled analyses were homogenous (P > 0.1).

GLP-1RAs did not show significant difference in terms of headache compared to the control [RR = 1.08, 95% CI (0.89, 1.31), P = 0.45], as reported by 2 studies [26, 43]. The pooled analysis was homogenous (P > 0.1) (Fig. 8h).

Sensitivity analysis

Due to the differences in the included studies and patient characteristics, we performed a sensitivity analysis to see the robustness of our analysis. Lundkvist et al. [27] administered GLP-1RA combined with a sodium-glucose co-transporter 2 (SGLT2) inhibitor. Mashayekhi et al. [40] compared GLP-1RAs without lifestyle modification to the hypocaloric diet. Zhou et al. [38] included only patients aged 6 to 18 years old. As a result, these studies were excluded from the analysis, which revealed no significant change on the results.

Trial sequential analysis

The TSA demonstrates that the cumulative z-curve had crossed the trial sequential monitoring boundary for the beneficial effect of GLP-1RAs therapy on the incidence of prediabetes reversion to normoglycemia (Figure S2) and the incidence of new-onset diabetes (Figure S3), and the actual cumulative sample size exceeded the required information size. This showed that the sample size was sufficient to draw firm conclusions regarding the beneficial effect of GLP-1RAs therapy on prediabetes reversion to normoglycemia and reducing the incidence of new-onset diabetes.

Publication bias

The funnel plots are shown in figures S4 to S12. No publication bias was observed in any of the efficacy outcomes as indicated by Egger’s (P > 0.05) and Begg’s (P > 0.05) tests, except for LDL, where there was significant publication bias as indicated by Egger’s test (P = 0.023). After applying the trim-and-fill method, it was revealed that after trimming three studies, no significant effect was noticed on the pooled estimate [MD = − 2.39, 95% CI (− 4.35, − 0.44)] (Figure S9).