How New Diabetes Medications Affect Kidney Health

This systematic review and network meta‐analysis followed the Preferred Reporting Items for Systematic Reviews and Meta‐analyses (PRISMA‐NMA, Table). The protocol for this study was registered in the International Prospective Register of Systematic Reviews (PROSPERO; registration number CRD42024464258). S1

We systematically searched the electronic literature of PubMed, Embase, the Cochrane Central Register of Controlled Trials, and ClinicalTrials.gov↗ according to a predefined protocol to identify relevant randomized controlled trials (RCTs) from inception up to July 2025 without language restriction but limited to adult humans. Details of the search terms are provided in the Data S1. Two additional studies were identified by manually searching the references of the included studies and other meta‐analyses.

After removing duplicate studies, two reviewers (Rong Lin and Chia‐Li Hsu) independently screened the titles, abstracts, and full texts of the remaining studies to select possible trials based on our inclusion criteria: (1) randomized controlled trials (RCTs, parallel design but not cross‐over); (2) two treatment arms, in which any DPP‐4 inhibitor, GLP‐1 receptor agonist, or SGLT‐2 inhibitor was compared with placebo (or usual treatment) to assess kidney disease outcomes in patients with type 2 diabetes and CKD; (3) reporting outcomes in diabetic CKD subgroups; and (4) a study duration of at least 24 weeks. Exclusion criteria included: (1) RCTs on other types of diabetes (e.g., type 1 diabetes mellitus or gestational diabetes); and (2) post‐hoc analyses of RCTs, which were excluded because they are not pre‐specified in the trial protocols, are exploratory in nature, and may increase the risk of selective reporting bias.

Three outcomes were assessed in our study: (1) the change in estimated glomerular filtration rate (eGFR) is reported as the mean difference with a 95% confidence interval (CI), (2) the change in urine albumin‐creatinine ratio (UACR) is reported as the ratio of fold change with a 95% CI, and (3) differences in composite renal outcomes (doubling of creatinine levels, 30% reduction in GFR, acute kidney injury, end‐stage kidney disease, or renal death) are reported as odds ratios (ORs) with 95% CIs.

Two reviewers (Rong Lin and Chia‐Li Hsu) independently extracted the data according to a predefined protocol using data collection software (Covidence, Veritas Health Innovation, Melbourne, Australia; available at www.covidence.org↗). The data extracted included study characteristics (first author, year of publication, country of origin, inclusion criteria for each trial, and sample size), population (sex, age, duration of diabetes, glycated haemoglobin [HbA1c], eGFR, type of intervention, and outcomes of interest).

The risk of bias was assessed by two reviewers (Rong Lin and Chia‐Li Hsu) using version 2 of the Cochrane risk of bias tool for randomized trials (RoB 2),15 which included random sequence generation, allocation concealment, blinding, missing outcome data, and selective reporting of results. Each domain was considered to have a low risk, some concerns, or high risk of bias.16

We also used the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) method to evaluate the level of confidence in the evidence from direct, indirect, and network estimates for all outcomes. We also used an Excel sheet provided by the GRADE Working Group to simplify the complicated steps.17 According to this method, the certainty of evidence from RCTs initially starts as high but may be downgraded to moderate, low, or very low owing to factors such as risk of bias, indirectness, inconsistency, or intransitivity. Regarding indirectness, we evaluated whether there were significant differences between the ethnic groups, intervention measures, control measures, and expected results reported in the studies and the actual application between different papers; if there were differences, they were downgraded. The Population, Intervention, Comparator, and Outcome (PICO) elements of the included studies were compared using our study protocol.18 We used a side‐splitting approach to analyse inconsistency.19 Publication bias was detected using funnel plots and Egger's test.20 When evaluating RCTs, these factors must be considered.21, 22, 23 To judge the intransitivity, we evaluated the indirect part of patient characteristics, differing co‐interventions, differing extents to which interventions of interest were optimally administered, differing comparators, and differences in outcome measurements, and considered downgrading according to severity.

All evaluations were performed by two reviewers (Rong Lin and Chia‐Li Hsu); if there was any disagreement, a third reviewer (Hon‐Yen Wu) was involved.

The PRISMA flow diagram shows the study selection process (Figure S1). We searched 1436 articles initially, and two articles were identified through references from other sources. A total of 208 articles were removed as duplicate publications, excluding studies based on the title, abstract, and full‐text review. Finally, 20 RCTs5, 10, 12, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44 were eligible for inclusion, which enrolled 80,670 participants with both T2DM and CKD.

The baseline characteristics of included RCTs are summarized in Table 1. The sample size ranged from 40 to 14 752, with the follow‐up durations between 24 and 282 weeks (median 104 weeks). The age of participants ranged from 61.2 to 68.5 years, with 21% to 48% being women. The duration of T2DM ranged from 10.6 to 19.6 years, baseline HbA1c ranged from 7.6% to 8.9%, and baseline eGFR ranged from 42.0 to 56.1 mL/min/1.73 m². Continuous variables are presented as mean ± SD or median (interquartile range, IQR), according to the original report of each trial. When dispersion measures were not available, only central tendency values are shown.

The included RCTs were published between 2014 and 2025, and three main categories of novel antidiabetic agents were identified: SGLT2 inhibitors (14 RCTs), GLP‐1 receptor agonists (4 RCTs), and DPP‐4 inhibitors (2 RCTs). Table 1 presents the baseline design and population characteristics of these studies, while Table S2 provides detailed information on the interventions, renal outcomes, and main findings of each trial. Together, these tables offer a comprehensive overview of both the study populations and the renal effects of novel antidiabetic agents, serving as the foundation for the subsequent pooled and network meta‐analyses.

Quality assessment conducted using the ROB 2 is presented in Figure. Each study is presented with the corresponding drug and trial designation. Overall, eight studies were judged to have a low risk of bias, two at high risk, while the remaining studies were classified as having some concerns of bias. The most common sources of bias were related to selective reporting, incomplete outcome data, and measurement of outcomes. S2

The certainty of evidence, assessed using the GRADE framework, varied across outcomes (Tables). For composite renal outcomes, certainty ranged from high to very low, with several placebo‐controlled comparisons (e.g., dapagliflozin, canagliflozin, ertugliflozin, semaglutide vs. placebo) rated as high certainty, while indirect and head‐to‐head comparisons were frequently downgraded. For eGFR change, most placebo‐controlled comparisons were of moderate certainty, whereas head‐to‐head estimates were downgraded to low or very low certainty. For UACR, evidence was more variable, with dapagliflozin versus placebo supported by high certainty but many other comparisons downgraded to low or very low. Overall, placebo‐controlled comparisons provided the most robust and reliable evidence, whereas indirect and head‐to‐head comparisons remained limited in certainty. S3–S5

Nine studies reported changes in eGFR.12, 27, 30, 31, 32, 36, 40, 42, 43 These studies formed three disconnected subnetworks (Figure S3). Among them, one subnetwork contained all SGLT2 inhibitors from five studies, whereas DPP4 inhibitors and GLP1‐RA had one article each. To address this issue, we conducted a secondary analysis focusing on SGLT2 inhibitors, defined as an analysis restricted to one connected subnetwork to allow meaningful comparisons. As shown in the forest plot (Figure S4), none of the SGLT2 inhibitors demonstrated a statistically significant effect on the change in eGFR compared with placebo in patients with T2DM and CKD. The estimated mean differences (MD) in eGFR compared with placebo were −4.98 mL/min/1.73 m² (95% CI –10.36 to 0.40) for canagliflozin 300 mg, −0.89 (95% CI –2.28 to 0.51) for sotagliflozin 400 mg, and −0.56 (95% CI –1.95 to 0.84) for sotagliflozin 200 mg. Positive but nonsignificant effects were observed for empagliflozin (0.90, 95% CI –0.42 to 2.22), semaglutide (1.17, 95% CI –0.16 to 2.50), and canagliflozin 100 mg (1.71, 95% CI –0.31 to 3.72).

Six studies reported the outcomes of changes in UACR.27, 29, 30, 31, 34, 40 After the analysis, two disconnected subnetworks were identified (Figure S5). One subnetwork contained only one DDP‐4 inhibitor (linagliptin), and another contained three SGLT‐2 inhibitors. Therefore, we performed a secondary analysis focusing on SGLT2 inhibitors (canagliflozin 300 mg, canagliflozin 100 mg, and dapagliflozin 10 mg). The forest plot (Figure S6) showed that canagliflozin 300 mg (ratio of means 0.57, 95% CI 0.42–0.78) and canagliflozin 100 mg (0.59, 95% CI 0.48–0.73) significantly reduced UACR compared with placebo, whereas dapagliflozin 10 mg (0.80, 95% CI 0.61–1.05) was not statistically significant.

To assess potential publication bias, funnel plots were generated for all three outcomes (Figures S7–S9). For composite renal outcomes and change in eGFR, Egger's regression tests did not indicate small‐study effects (p > 0.05). For UACR, only seven studies were available; therefore, Egger's regression test was not performed due to limited statistical power. The funnel plot for UACR appeared symmetrical.

This systematic review and network meta‐analysis provides the most updated and comprehensive drug‐specific evaluation of the renal effects of novel antidiabetic agents in patients with T2DM and CKD. Unlike previous analyses that mainly reported class‐level effects,14, 45 our study revealed clinically relevant heterogeneity across individual agents. Dapagliflozin showed the largest reduction in composite renal outcomes, canagliflozin produced the most pronounced reduction in albuminuria, and both dulaglutide 1.5 mg and semaglutide demonstrated significant renal benefits, whereas exenatide showed only a nonsignificant trend. DPP‐4 inhibitors provided no renal protection. These findings highlight that renal benefits are not uniform within drug classes and underscore the importance of drug‐level evaluation.

Our findings are consistent with large‐scale outcome trials confirming the renoprotective effects of SGLT2 inhibitors, including DAPA‐CKD,11 CREDENCE,10 and EMPA‐KIDNEY.12 These studies demonstrated significant reductions in kidney failure, albuminuria progression, and cardiovascular mortality, supporting the robust efficacy of this drug class. Earlier network meta‐analyses, such as those by Cao et al.45 and Yang et al.,14 also confirmed class‐wide benefits of SGLT2 inhibitors, but did not incorporate more recent evidence.

For GLP‐1 receptor agonists, our analysis aligns with evidence from the REWIND trial,46 where dulaglutide reduced albuminuria progression and improved renal outcomes. More recently, the FLOW trial demonstrated significant renal benefits of semaglutide, reinforcing the role of GLP‐1 receptor agonists in kidney protection.43 However, our results also highlight heterogeneity within this class: dulaglutide and semaglutide showed significant benefits, whereas exenatide exhibited only a nonsignificant trend.

By focusing on individual agents rather than drug classes, our study extends previous findings and identifies meaningful drug‐level differences, such as the stronger albuminuria reduction observed with canagliflozin compared with dapagliflozin, and the differential renal efficacy among GLP‐1 receptor agonists.

The renoprotective effects of SGLT2 inhibitors are likely mediated by mechanisms beyond glucose lowering. These include restoration of tubuloglomerular feedback, reduction of intraglomerular pressure, improved renal oxygenation, and attenuation of oxidative stress.49, 50 Additional systemic effects, such as natriuresis, lower blood pressure, and weight reduction, also contribute to kidney protection.47, 48, 49, 50, 51 GLP‐1 receptor agonists may act through complementary pathways, including anti‐inflammatory and antifibrotic effects, improvement in blood pressure and body weight, and possible direct renal protection observed in preclinical studies.6, 51, 52, 53 The variable efficacy across individual GLP‐1 receptor agonists observed in our analysis may reflect differences in receptor affinity, pharmacokinetics, or trial populations. In contrast, the absence of benefit from DPP‐4 inhibitors is consistent with their lack of significant hemodynamic or renal‐specific effects.4, 5

Our findings have direct clinical relevance for the management of patients with T2DM and CKD. SGLT2 inhibitors should remain the first‐line therapy given their consistent and robust renal benefits across trials.7, 8, 9, 10, 11, 12 Notably, dapagliflozin showed the greatest effect on composite renal outcomes, whereas canagliflozin demonstrated stronger reductions in albuminuria, suggesting that patient characteristics, such as baseline albuminuria, may guide optimal drug selection within this class. Among GLP‐1 receptor agonists, both dulaglutide 1.5 mg and semaglutide were associated with significant renal benefits, whereas exenatide showed only a nonsignificant trend, emphasizing that not all agents within this class are equivalent. These findings highlight the importance of individualized therapy and suggest that treatment decisions should consider not only glycemic control and cardiovascular outcomes but also the specific renal effects of individual drugs. In contrast, DPP‐4 inhibitors conferred no renal protection and therefore should not be prioritized for kidney outcomes in this population.

This study has several limitations. First, most treatment comparisons relied on indirect evidence, as few head‐to‐head RCTs were available, which lowers the certainty of some estimates. Second, although the risk of bias was generally low to moderate, common concerns included selective reporting, incomplete outcome data, and outcome measurement. The certainty of evidence also varied: placebo‐controlled comparisons, particularly for dapagliflozin, canagliflozin, and semaglutide, were often rated high, whereas indirect and head‐to‐head comparisons were frequently downgraded to low or very low due to imprecision or inconsistency. Publication bias could not be fully excluded; although funnel plots and Egger's tests did not suggest small‐study effects for composite renal outcomes and eGFR, the limited number of studies for UACR reduced statistical power. Third, heterogeneity in study design, patient populations, and outcome definitions may have influenced the pooled results. Fourth, the analysis was based on study‐level aggregate data, preventing stratified analyses by baseline renal function or albuminuria. Fifth, the median follow‐up duration was relatively short, potentially underestimating long‐term renal effects. In addition, we excluded post‐hoc analyses, such as the SURPASS‐4 renal outcome study,54 because of their exploratory design and higher risk of bias. Finally, newer agents, including tirzepatide, were not included; future large‐scale, head‐to‐head RCTs are needed to define their comparative renal efficacy.

In conclusion, this network meta‐analysis reinforces SGLT2 inhibitors as the most reliable therapy for renal protection in patients with T2DM and CKD, while GLP‐1 receptor agonists provide additional but variable benefits, and DPP‐4 inhibitors show no effect. By evaluating agents at the drug level, our study highlights clinically meaningful heterogeneity and underscores the need for further head‐to‐head trials to refine treatment strategies and assess newer therapies such as tirzepatide.

This study was based on data extracted from previously published randomized controlled trials and trial registries. PubMed, EMBASE, the Cochrane Central Register of Controlled Trials (CENTRAL), and ClinicalTrials.gov↗ were systematically searched from inception to July 2025. All data analyzed are included in this article and its supplementary materials.