Literature search results

Our comprehensive search identified a total of 3021 records. The titles and abstracts of all these records were screened; however, only 30 articles seemed eligible for full-text screening. Of them, 12 RCTs were included in our systematic review and meta-analysis. No further RCTs were identified after performing citation analyses. We provide the PRISMA flow diagram of the study selection process in Fig. 1.

Fig. 1

The PRISMA flow diagram of the study selection process

Characteristics of individual studies

Our meta-analysis included 12 RCTs, comprising 2720 individuals with prediabetes. The largest sample size was in Zhang et al., which included 1678 participants [18], while the smallest sample size was in Malin et al., which included 16 participants [31]. Three studies were conducted in India [8, 25, 30] and three were conducted in the United States [28, 31, 32]. Figure 2 shows the geographical distribution of all RCTs included in our meta-analysis [18, 24, 26, 27, 34, 35]. The follow-up duration ranged from three months in both Viskochil et al. [28] and Malin et al. [31] studies to 24 months in Zhang et al. [18]. We summarized all included studies and their patients’ baseline characteristics in Table 2 and Supplementary Table 2, respectively. In addition, Supplementary Table 3 summarizes the lifestyle interventions used by each of the included RCTs.

According to the ROB-2 tool, three RCTs showed a low risk of bias; six RCTs demonstrated some concerns regarding their risk of bias; and the remaining three RCTs showed a high risk of bias. A summary and graph for risk of bias assessment are shown in Supplementary Fig. S1a and S1b, respectively.

Fig. 2

Geographical distribution of the randomized controlled trials included in our meta-analysis

Table 2 Summary of the included studiesOutcome measuresIncidence of type 2 diabetes

Our pooled analysis revealed a significant reduction in the incidence of type 2 diabetes when metformin was combined with lifestyle interventions compared to lifestyle interventions alone (RR = 0.85, 95% CI [0.75, 0.97], P = 0.01). The pooled studies were homogenous (I2 = 0%, P = 0.70) (Fig. 3). However, sensitivity analysis omitting Zhang et al. revealed no significant difference between the two groups (RR = 1, 95% CI [0.75, 1,34], I2 = 0%) (Table 3 and Supplementary Fig. S2).

Fig. 3

Pooled results for the incidence of type 2 diabetes

Glycemic control measurementsHbA1c

Our pooled analysis for the change in HbA1c levels at the endpoints of included studies showed that the combination of metformin and lifestyle interventions exhibited a significant reduction in HbA1c levels, compared to lifestyle interventions alone (SMD = -0.10, 95% CI [-0.19, -0.01], P = 0.03). The pooled studies were homogenous (I2 = 0%, P = 0.89) (Fig. 4a). However, sensitivity analysis omitting Zhang et al. showed no significant difference between the two groups (RR = -0.21, 95% CI [-0.44, 0.02], I2 = 0%) (Table 3 and Supplementary Fig. S3).

Furthermore, our subgroup analysis based on different time points, revealed that the combined approach resulted in a significant decrease in HbA1c levels at 3 and 6 months (SMD = -0.37, 95% CI [-0.68, -0.05], P = 0.02; SMD = -0.35, 95% CI [-0.62, -0.07], P = 0.01, respectively). However, the pooled studies at 12 months indicated no significant difference between the two groups (SMD = -0.20, 95% CI [-0.49, 0.09], P = 0.17). The studies within each subgroup exhibited homogeneity (I2 = 0%, P = 0.97; I2 = 0%, P = 0.58; I2 = 0%, P = 0.86, respectively) (Fig. 4b).

Fig. 4

Pooled results for the change in HbA1c levels at (a) endpoints and (b) different time points

Fasting blood glucose

Our pooled analysis at the endpoints of included studies did not reveal any statistically significant difference between metformin plus lifestyle interventions versus lifestyle interventions alone (SMD = -0.12, 95% CI [-0.33, 0.09], P = 0.26) (Fig. 5a). The pooled studies were heterogenous (I2 = 50%, P = 0.04). To address this heterogeneity, we conducted a sensitivity analysis using the leave-one-out model, which showed that heterogeneity was best resolved by excluding Zhang et al., and the results became statistically significant, favoring the combination of metformin and lifestyle interventions (SMD = -0.21, 95% CI [-0.39, -0.02], I2 = 0%) (Supplementary Fig. S4). In contrast, sensitivity analysis omitting Wiegand et al. showed that the two groups were comparable (SMD = -0.13, 95% CI [-0.35, 0.10], I2 = 55%) (Supplementary Fig. S4).

Interestingly, our subgroup analysis based on different time points revealed that there was no significant difference between the two groups at both 3 and 6 months (SMD = -0.13, 95% CI [-0.42, 0.15], P = 0.37; SMD = -0.07, 95% CI [-0.31, 0.17], P = 0.58, respectively). However, adding metformin to lifestyle interventions significantly decreased FPG at 12 months, compared to lifestyle interventions alone (SMD = -0.34, 95% CI [-0.59, -0.08], P = 0.01). The studies within each subgroup exhibited homogeneity (I2 = 0%, P = 0.76; I2 = 0%, P = 0.74; I2 = 3%, P = 0.37, respectively) (Fig. 5b). Subgroup analysis based on the geographical distribution showed that the two groups were comparable in studies conducted in both Asia and America (SMD = -0.10, 95% CI [-0.35, 0.15], P = 0.45; SMD = -0.44, 95% CI [-1.11, 0.24], P = 0.20, respectively) (Fig. 5c).

Fig. 5

Pooled studies for fasting blood glucose levels (a) at endpoints, (b) with subgrouping based on the follow-up duration, (c) with subgrouping based on the geographical distribution

Blood pressure measurementsDiastolic blood pressure

There was no statistically significant difference in DBP changes between the two groups (SMD = 0.04, 95% CI [-0.05, 0.13], P = 0.42). The pooled studies demonstrated homogeneity (I2 = 0%, P = 0.95) (Supplementary Fig. S5). The results remained non-significant after omitting Zhang et al. and Wiegand et al. (SMD = -0.04, 95% CI [-0.32, 0.24]; SMD = 0.04, 95% CI [-0.05, 0.13], respectively) (Table 3 and Supplementary Fig. S6).

Systolic blood pressure

Similarly, there was no statistically significant difference in SBP changes between the two groups (SMD = -0.04, 95% CI [-0.13, 0.05], P = 0.34). The pooled studies demonstrated homogeneity (I2 = 0%, P = 0.75) (Supplementary Fig. S7). The results remained non-significant after omitting Zhang et al. (SMD = -0.16, 95%CI [-0.41, 0.08], Table 3) and Supplementary Fig. S8) and Wiegand et al. (SMD = -0.04, 95% CI [-0.13, 0.05], Supplementary Fig. S8).

Body weight measurementsBody mass index

Our pooled analysis at the studies’ endpoints revealed that there was no statistically significant difference between the two groups in terms of BMI changes (SMD = -0.02, 95% CI [-0.10, 0.07], P = 0.67). The pooled studies demonstrated homogeneity (I2 = 0%, P = 0.80) (Supplementary Fig. S9). The results remained non-significant after omitting Zhang et al. (SMD = -0.03, 95% CI [-0.22, 0.17], Table 3 and Supplementary Fig. S10).

In addition, our subgroup analysis at 6 and 12 months showed similar results (SMD = -0.12, 95% CI [-0.42, 0.19], P = 0.45; SMD = 0.04, 95% CI [-0.22, 0.30], P = 0.76, respectively). The pooled studies within these subgroups displayed homogeneity (I2 = 0%, P = 0.41; I2 = 0%, P = 0.71, respectively) (Supplementary Fig. S11). Moreover, our subgroup analysis based on the age group revealed that adding metformin to lifestyle interventions was comparable to lifestyle interventions alone in both adults and adolescents (SMD = -0.01, 95% CI [-0.10, 0.08], P = 0.81; SMD = -0.17, 95% CI [-0.66, 0.31], P = 0.51, respectively) (Supplementary Fig. S12).

Waist circumference

Adding metformin to lifestyle interventions showed comparable results to lifestyle interventions alone in terms of changes in waist circumference (SMD = 0.00, 95% CI [-0.09, 0.10], P = 0.93). The pooled studies demonstrated homogeneity (I2 = 0%, P = 0.57) (Supplementary Fig. S13). The results remained non-significant after omitting Zhang et al. (Table 3 and Supplementary Fig. S14).

Weight

Our pooled studies at endpoints showed a trend towards higher weight with lifestyle interventions plus metformin; however, this did not reach statistical significance (SMD = 0.12, 95% CI [-0.03, 0.27], P = 0.11). The pooled studies demonstrated heterogeneity (I2 = 57%, P = 0.07) (Supplementary Fig. S15). To address this heterogeneity, we conducted a sensitivity analysis using the leave-one-out model, which showed that it was best resolved by excluding Zhang et al., and the results became significant favoring the lifestyle interventions only group (SMD = 0.22, 95% CI [0.10, 0.33], I2 = 0%) (Supplementary Fig. S16).

Insulin resistance

Our pooled analysis at the studies endpoints showed no statistically significant difference between the two groups, with a SMD − 0.06 (95% CI [-0.16, 0.03], P = 0.17). The pooled studies demonstrated homogeneity (I2 = 0%, P = 0.86) (Supplementary Fig. S17). The results remained non-significant after omitting Zhang et al. and Wiegand et al. (SMD = -0.15, 95% CI [-0.51, 0.22]; SMD = -0.06, 95% CI [-0.16, 0.03], respectively) (Table 3 and Supplementary Fig. S18).

Lipid profile measurementsSerum HDL

Our pooled analysis at 6 and 12 months showed no statistically significant difference between the two groups (SMD = 0.03, 95% CI [-0.27, 0.33], P = 0. 83; SMD = -0.02, 95% CI [-0.28, 0.23], P = 0.87, respectively). The pooled studies at 6 and 12 months were homogenous (I2 = 0%, P = 0.80; I2 = 0%, P = 0.95) (Supplementary Fig. S19). The results remained statistically non-significant after omitting the study by Wiegand et al. (SMD = 0.02, 95% CI [-0.20, 0.23], Supplementary Fig. S20).

Serum LDL

Our pooled analysis at 6 and 12 months showed no statistically significant difference between the two groups (SMD = 0.30, 95% CI [-0.01, 0.59], P = 0.05; SMD = -0.02, 95% CI [-0.27, 0.24], P = 0.89, respectively). The pooled studies within each subgroup displayed homogeneity (I2 = 0%, P = 0.58; I2 = 0%, P = 0.75, respectively) (Supplementary Fig. S21). The results remained non-significant after omitting Wiegand et al. (SMD = 0.06, 95% CI [-0.15, 0.27], Supplementary Fig. S22).

Total cholesterol

Our pooled analysis at 6 and 12 months showed that the changes in total cholesterol were comparable between the two groups (SMD = 0.10, 95% CI [-0.20, 0.40], P = 0.52; SMD = -0.02, 95% CI [-0.28, 0.23], P = 0.85, respectively). The pooled studies displayed homogeneity (I2 = 0%, P = 0.49; I2 = 0%, P = 0.57, respectively) (Supplementary Fig. S23). The results remained non-significant after omitting Wiegand et al. (SMD = 0.03, 95% CI [-0.18, 0.24], Supplementary Fig. S24).

Serum triglycerides

Our pooled analysis at 6 and 12 months showed that the changes in serum triglycerides were comparable between the two groups (SMD = -0.02, 95% CI [-0.22, 0.17], P = 0.80; SMD = 0.05, 95% CI [-0.21, 0.30], P = 0.71, respectively). The pooled studies displayed homogeneity (I2 = 17%, P = 0.30; I2 = 0%, P = 0.46, respectively) (Supplementary Fig. S25).

Table 3 Summary of our analysis