Study selection
After an initial search, 1993 articles were downloaded from 4 databases. Then after reading the titles and abstracts, 94 articles were obtained by excluding articles with duplicate content and those not relevant to this study. After reading the full text in detail, 59 articles were excluded for the following reasons, such as the study design (n=2), insufficient information for a meta-analysis (n=15), fewer than 20 participants (n=29), lack of baseline data or the baseline did not meet the inclusion criteria (n=5), conference abstract (n=8). Thirty-five studies were subsequently included in this network meta-analysis (Fig.1).
Study characteristics
In this network meta-analysis, 35 studies were included, comprising a total sample size of 4322 participants. The 35 studies included 31 RCTs and 4 cohort studies. The summary data of each included study are shown in Additional fle 2: Table S1, and the network plot is shown in Fig. 2.
Risk of bias within studies
We evaluated the quality of the included studies for assessing the risk of bias. 6 studies were designed as open-label[22, 24, 29, 39, 42, 43]. Among those open-label studies, 2 studies were judged to have high risk of selective reporting bias. One article did not specify the method of randomization[42], and the other article reported more than 20% of total withdrawals during follow-up, but did not mention the specific reasons[29]. Four studies were cohort studies and were rated high quality according to NOS scores. The details of risk of bias quality assessment for each RCT and cohort study was shown in Additional fle 3: Figure S1-2, Table S2.
Synthesis of results
LVEF
(1) Evidence Network. Twenty-eight literatures reported LVEF involved three antidiabetic regimens. The size of the sample containing the intervention determines the size of the points, and the number of RCTs of the treatment intervention determines the thickness of the line. There is four closed-loop formation. (Additional fle 4: Figure S3(a)).
(2) Publication Bias. Most of the scatter points in the funnel plot were located on either side of the vertical line. They were fundamentally symmetric and may have some level of publication bias (Additional fle 5: Figure S4(a)).
(3) Network Meta-analysis. Overall response rates for 3 antidiabetic drugs were reported in 28 publications. The network comparison was conducted in 3 antidiabetic drugs, with 12 pairwise comparisons, one of which was significantly different from placebo. Compared with placebo, MD and 95% CI of DPP-4i was -0.89% and (-1.76, -0.03). The specific results are shown in Table 1 and Additional fle 6: FigureS5(a).
(4) SUCRA Probability Ranking. SUCRA scores in descending order of probability are as follows: Placebo(72.4%)>GLP-1RA(59.3%)>SGLT-2i(42.3%)>DPP-4i(26%). (Additional fle 7: TableS3 and Additional fle 8: FigureS6(a)).
Results of subgroup analyses
To identify subgroups of patients who may benefit more from antidiabetes therapy, patients were divided into two groups according to whether they had CVD with T2DM or CVD alone. Treatment with GLP-1RA significantly improved the LVEF in patients with CVD alone [MD=1.65%, 95% CI (0.49, 2.81)], while no significant difference was identified between 3 interventions and placebo among patients with CVD and T2DM. (Additional fle 9: TableS3(a),(b))
LVEDD
(1) Evidence Network. Ten literatures reported LVEDD involved three antidiabetic regimens. The sample size using the intervention is indicated by the size of the dot, and the number of RCTs for the treatment intervention is indicated by the thickness of the line, and all 3 cis indicate direct comparisons that did not form a closed loop. (Additional fle 4: Figure S3(b)).
(2) Publication Bias. In the funnel plot, most of the scattered points converge on either side of the vertical line. They were essentially symmetric, and a certain degree of publication bias was possible. The funnel plot of the total effective rates of the three antidiabetic drugs is shown in Additional fle 5: Figure S4(b).
(3) Network Meta-analysis. 10 studies provided changes in LVEDD before and after drug intervention. 12 pairwise comparisons were generated from the network analysis of 3 antidiabetic drugs, The difference in mean pre- and post-treatment change in SGLT-2i for LVEDD compared to placebo was greater than zero [MD -0.72 mm 95% CI (-1.30, -0.14)], suggesting that SGLT-2i was more associated with improvement in LVEDD than placebo. (Table 2 and Additional fle 6: FigureS5(b)).
(4) SUCRA Probability Ranking. The SUCRA value of 10 studies indicates that SGLT-2i (SUCRA 91.7%) had the highest probability of LVEDD improvement compared to two other classes of antidiabetic drugs and placebo. The remaining SUCRA scores in descending order of probability are as follows:
GLP-1RA(39.6%)>DPP-4i(38.6%)>Placebo(30.1%). (Additional fle 7: TableS3 and Additional fle 8: FigureS6(b)).
LVEDV
18 pieces of literature reported LVESV involving 4 interventions (Additional fle 4: Figure S3(c)). Most of the scatter points in the funnel plot were located on either side of the vertical line. The points were essentially symmetric and may have some degree of publication bias. (Additional fle 5: Figure S4(c)). No significant differences emerged between the four interventions (Table 3 and Additional fle 6: FigureS5(c)). SUCRA scores in descending order of probability are as follows: GLP-1RA(64.9%)>Placebo(54.3%)>DPP-4i(46.8%)>SGLT-2i(34.0%). (Additional fle 7: TableS3 and Additional fle 8: FigureS6(c)).
LVESD
(1) Evidence Network. LVESD was reported in 6 studies, involving 2 antidiabetic therapies (GLP-1RA and SGLT-2i). The size of the sample containing the intervention determines the size of the points, and the number of RCTs of the treatment intervention determines the thickness of the line. All interventions represent direct comparison without closed-loop formation. (Additional fle: 4 Figure S3(d)).
(2) Publication Bias. In the funnel plot, most of the scattered points converge on either side of the vertical line. They were essentially symmetric, and a certain degree of publication bias was possible. (Additional fle 5: Figure S4(d)).
(3) Network Meta-analysis. 6 studies provided changes in LVESV before and after drug intervention. GLP-1RA significantly reduced LVESD compared with placebo [MD=-0.38 mm, 95% CI (-0.66, -0.10)]. There was no difference in the pairwise comparison of treatment effects between the two drugs. (Table 4 and Additional fle 6: FigureS5(d)).
(4) SUCRA Probability Ranking. SUCRA scores in descending order of probability are as follows: GLP-1RA(88.7%)>SGLT-2i(57.4%)>Placebo(3.9%). (Additional fle 7: TableS3 and Additional fle 8: FigureS5(d)).
LVESV
16 pieces of literature reported LVESV involving 3 interventions (Additional fle 4: Figure S3(e)). The scatterred points of funnel plot were fundamentally symmetrical and may have a certain degree of publication bias. (Additional fle 5: Figure S4(e)). No significant difference was identified in LVESV between the 3 interventions. (Table 5 and Additional fle 6: FigureS5(e)). SUCRA scores in descending order of probability are as follows: GLP-1RA(75.3%)>SGLT-2i(70.8%)>Placebo(4.0%). (Additional fle 7: TableS3 and Additional fle 8: FigureS6(e)).
LVMI
(1) Evidence Network. 10 literatures reported LVMI involved three antidiabetic regimens. The sample size using the intervention is indicated by the size of the dot, and the number of RCTs for the treatment intervention is indicated by the thickness of the line, and there is one closed-loop formation. (Additional fle 4: Figure S3(f)).
(2) Publication Bias. In the funnel plot, most of the scattered points converge on either side of the vertical line. They were essentially symmetric, and a certain degree of publication bias was possible (Additional fle 5 Figure S4(f)).
(3) Network Meta-analysis. In terms of the outcome of LVMI, two classes of drug showed significant benefits with regard to reducing LVMI in all patients compared to placebo: GLP-1RA [MD =-1.07g/m2, 95% CI (-1.71, -0.42)], SGLT-2i [MD =-0.28 g/m2, 95% CI (-0.43, -0.12)]. GLP-1RA showed efficacy compared to SGLT-2i [MD =-0.79g/m2, 95% CI (-1.46, -0.12)]. In addition, DPP-4i showed a negative impact compared to SGLT-2i [MD =1.34g/m2, 95% CI (0.93, 1.75)]. (Table 6 and Additional fle 6: FigureS5(f)).
(4) SUCRA Probability Ranking. SUCRA scores in descending order of probability are as follows: DPP-4i(95.2%)> GLP-1RA(71.1%)>SGLT-2i(33.7%)>Placebo(0.0%). (Additional fle 7 TableS3 and Additional fle 8 FigureS6(f)).
e’
(1) Evidence Network. e’ was reported in 6 studies, involving 3 antidiabetic therapies. The size of the sample containing the intervention determines the size of the points, and the number of RCTs of the treatment intervention determines the thickness of the line, and all interventions represent direct comparison without closed-loop formation. (Additional fle 4: Figure S3(g)).
(2) Publication Bias. In the funnel plot, most of the scattered points converge on either side of the vertical line. They were fundamentally symmetrical and may show a degree of publication bias. (Additional fle 5: Figure S4(g)).
(3) Network Meta-analysis. 6 studies provided changes in e’ before and after drug intervention. The results showed that DPP-4i was the only drug that significantly increased e' [MD=3.82cm/s, 95% CI (2.92, 4.7)]. Also compared to placebo, the difference in mean change in e’ with GLP-1RA treatment was less than zero [MD=-0.43cm/s, 95% CI (-0.81, -0.04)], indicating a negative effect of GLP-1RA on e’. In addition, SGLT-2i [MD=-2.94, 95% CI (-4.24, -1.68)] and GLP-1RA [MD=-4.24, 95% CI (-5.22, -3.27)] significantly reduced e' compared to DPP-4i. While GLP-1RA had a more significant negative effect compared to SGLT-2i [MD =-1.28 95% CI (-2.27, -0.3)]. (Additional fle 9: TableS3(i) and Additional fle 6: FigureS5(g)).
(4) SUCRA Probability Ranking. SUCRA scores in descending order of probability are as follows: DPP-4i(100.0%)>SGLT-2i(66.5%)>Placebo(33.9%)>GLP-1RA(0.6%). (Additional fle 7: TableS3 and Additional fle 8: FigureS6(g)).
E/e’
(1) Evidence Network. E/e’ was reported in 15 studies, involving 3 antidiabetic therapies. The sample size using the intervention is indicated by the size of the dot, and the number of RCTs for the treatment intervention is indicated by the thickness of the line with a closed-loop formation. (Additional fle 4: Figure S3(h)).
(2) Publication Bias. In the funnel plot, most of the scattered points converge on either side of the vertical line. They were fundamentally symmetrical and may show a degree of publication bias. ((Additional fle 5: Figure S4(h)).
(3) Network Meta-analysis. 15 studies provided changes in E/e’ before and after drug intervention. Compared with placebo, DPP-4i significantly improved E/e' [MD=-5.97 95%CI (-10.35,-1.59)], while GLP-1RA showed a negative impact compared to DPP-4i [MD=5.78 95% CI (0.60, 10.95)]. (Additional fle 9: Table S3(j) and Additional fle 6: FigureS5(h)).
(4) SUCRA Probability Ranking. SUCRA scores in descending order of probability are as follows: DPP-4i(79.0%)>SGLT-2i(77.8%)>GLP-1RA (9.7%)>Placebo (33.6%). (Additional fle 7: TableS3 and Additional fle 8: FigureS6(h)).
Results of subgroup analyses
The subgroup of patients with CVD and T2DM included 5 studies. In the analysis, SGLT-2i showed a more significant ability to reduce E/e' than placebo [MD=- 0.08; 95% CI (-0.79, -0.06)]. No significant differences were found for the rest of the results. (Additional fle 9 : TableS3(c)).
E/A
18 pieces of literature reported LVESV involving 4 interventions (Additional fle 4: Figure S3(i)). The scattered points of the funnel plot were essentially symmetric and may have some degree of publication bias. (Additional fle 5: Figure S4(i)). No significant difference was identified in the mean change of E/A between the 4 interventions. (Additional fle 9: Table S3(k) and Additional fle 6: FigureS5(i)). SUCRA scores in descending order of probability are as follows: GLP-1RA(64.9%)>Placebo(54.3%)>DPP-4i(46.8%)>SGLT-2i(34.0%). (Additional fle 7: TableS3 and Additional fle 8: FigureS6(i)).
SBP, NT-proBNP and 6MWT
A significant association between GLP-1RA therapy and improvement of 6-min walk distance was found in the overall population compared with placebo [MD=1.52m, 95% CI (0.29, 2.76)]. No significant difference was identified in the mean change of SBP and NT-proBNP between the 4 interventions. The detailed results are shown in Additional fle 10, figures are shown in Additional fle 4-6,8: Figure S3-6(j,k,l,q,s,t), SUCRA scores’ table is shown in Additional fle 7: TableS3.