SGLT2 Inhibitors Show Lower Severe Infection Risk Than DPP-4 Drugs, Review Finds

A large systematic review and meta-analysis of real-world cohort studies has found that people with type 2 diabetes who took sodium-glucose cotransporter 2 (SGLT2) inhibitors did not have a higher average risk of severe infection than those who took dipeptidyl peptidase-4 (DPP-4) inhibitors. In fact, the pooled estimates pointed in the opposite direction: incident pneumonia was 35% lower among SGLT2 inhibitor users (pooled ratio 0.65, 95% confidence interval 0.55–0.75) and incident sepsis was also 35% lower (0.65, 0.53–0.80) in five active-comparator cohorts for each endpoint. Yet the authors of the review, published in Endocrinology, Diabetes & Metabolism, are emphatic that these lower numbers should not be read as proof that the drugs prevent pneumonia or sepsis.

The review, registered in PROSPERO before its final database search through 29 June 2026, pooled adjusted hazard ratios and incidence rate ratios from observational studies using random-effects models with a modified Hartung-Knapp adjustment. The researchers searched PubMed, Embase, Scopus and Cochrane CENTRAL from inception, screened 402 records after deduplication, assessed 39 full-text reports and included 18 studies in the qualitative synthesis. Eleven studies contributed estimates to quantitative analyses, ten of them to at least one pooled analysis. All studies compared SGLT2 inhibitors against an active comparator, most often DPP-4 inhibitors, because the two drug classes are typically initiated at a similar stage of type 2 diabetes treatment, making the contrast a clinically meaningful one rather than a comparison against untreated patients.

Severe infection is a clinically relevant outcome when glucose-lowering drugs are compared in routine care. Type 2 diabetes itself is associated with increased risks of pneumonia, sepsis, infection-related hospitalisation and death, driven by hyperglycaemia, impaired innate immunity, chronic kidney disease, frailty and cardiopulmonary comorbidity. The infection-related safety profile of SGLT2 inhibitors has usually been framed around genital mycotic and urinary tract infections, but several cohorts have reported lower estimates for pneumonia, sepsis or infection-related hospitalisation. Biological explanations such as reduced inflammation, oxidative stress and preserved cardiorenal function compete with treatment-selection explanations, including lower frailty, better vaccination, greater adherence and closer clinical follow-up among people prescribed SGLT2 inhibitors.

The new synthesis differs from a previous systematic review in several methodological respects. It restricts all principal quantitative pooling to DPP-4i active comparators, applies explicit rules for overlapping data sources, reports prediction intervals alongside confidence intervals, and grades certainty with GRADE. Incident pneumonia and incident sepsis were kept as separate co-primary outcomes rather than combined into a composite, and infection-related hospitalisation was added as a secondary outcome. The included studies, published between 2021 and 2026, drew on claims, insurance, electronic health record, registry and hospital-linked data from eight jurisdictions and multinational federated networks.

Heterogeneity was substantial for both co-primary endpoints. For incident pneumonia, the between-study statistic I² reached 89.8%, with a 95% prediction interval of 0.45–0.93. For incident sepsis, I² was 90.6%, and the prediction interval of 0.38–1.13 crossed the null value of 1.0, meaning the range of associations expected in a future study included both lower and higher risk. Because heterogeneity persisted after every study was omitted in turn and no single study was influential, the authors interpret the sepsis pooled mean as exploratory, describing an average across heterogeneous settings rather than a common effect. Certainty of evidence was rated very low for both incident pneumonia and sepsis under GRADE.

Leave-one-out analyses revealed a striking asymmetry between the two endpoints. Removing one multinational federated-network cohort, which used an index date two years after drug initiation, collapsed pneumonia heterogeneity from I² of 89.8% to 0.0% and yielded a pooled estimate of 0.61 (0.56–0.67). Omitting any other cohort left pneumonia I² between 87.6% and 92.0%. Sepsis heterogeneity, by contrast, remained between 88.1% and 92.4% after each omission, indicating that no single study explained the dispersion. The authors note that the exceptionally small within-study variance of the federated-network estimate made I² highly sensitive to modest differences among estimates that all remained below 1.0, whereas sepsis showed a larger τ² and greater genuine dispersion in the underlying associations.

Overlapping data sources posed a further challenge. Three DPP-4i-comparator cohorts came from the same TriNetX federated network: an unselected multinational type 2 diabetes cohort, a cohort restricted to end-stage kidney disease and one restricted to psoriasis. Only the unselected cohort entered the primary synthesis; the selected-population cohorts were retained for sensitivity analyses and qualitative triangulation, because findings from clinically selected populations are not directly transportable to the broader type 2 diabetes population. Two Australian cohorts were both retained because patient-level overlap could not be quantified, and each was removed separately in sensitivity analyses. Adding the end-stage kidney disease cohort to pneumonia moved the estimate to 0.68 (0.55–0.85) and pushed the prediction interval across the null (0.39–1.21); adding the end-stage kidney disease and psoriasis cohorts to sepsis gave 0.68 (0.58–0.81) with a prediction interval of 0.43–1.08. Similar pooled means can therefore coexist with materially different predictive uncertainty.

Infection-related hospitalisation, a secondary outcome, was reported by two studies: a US Veterans Affairs cohort (hazard ratio 0.74, 0.67–0.81, measuring time to first event) and an Australian post-discharge cohort (weighted incidence rate ratio 0.74, 0.71–0.77, capturing event rates including repeat admissions). Because fewer than three studies were available under the registered protocol, these estimates were not pooled, and certainty was rated low. Mortality evidence was also not pooled because the contributing reports differed in cause specificity, clinical time origin and effect measure. One Hong Kong cohort reported lower infection-related mortality (0.43, 0.32–0.57), pneumonia-related mortality (0.41, 0.29–0.58) and sepsis-related mortality (0.39, 0.18–0.84) with SGLT2 inhibitors, while a Japanese cohort of older adults already hospitalised with pneumonia found no statistically significant difference in in-hospital mortality (0.64, 0.40–1.05). These distinct outcomes and time origins were interpreted narratively.

Comparator choice offered an indirect check on the findings. In two general type 2 diabetes cohorts that also compared SGLT2 inhibitors with GLP-1 receptor agonists, the lower-risk pattern was not consistent or statistically clear: one reported a hospitalised community-acquired pneumonia hazard ratio of 0.94 (0.44–1.89), and the other reported higher estimates for incident pneumonia (1.27, 1.23–1.32) and severe sepsis (1.27, 1.22–1.31) with SGLT2 inhibitors. In clinically selected populations with psoriasis or advanced chronic kidney disease, SGLT2 inhibitors and GLP-1 receptor agonists did not differ. Because GLP-1 receptor agonists may themselves be associated with lower infection risk, these contrasts cannot separate residual treatment channelling against DPP-4 inhibitors from a specific association with GLP-1 receptor agonists. Randomised evidence likewise does not support a class-wide preventive effect: a meta-analysis of trials directly comparing the two drug classes did not report pneumonia or sepsis, and a network meta-analysis of 105 trials found no consistent class-level reduction in infection outcomes.

The authors conclude that the evidence does not justify prescribing SGLT2 inhibitors to prevent pneumonia or sepsis, nor does it modify established cardiorenal indications, contraindications or sick-day guidance. What it does indicate is that choosing an SGLT2 inhibitor rather than a DPP-4 inhibitor was not associated with an increased average risk of severe infection in the populations studied. The findings also argue against treating infection as a single class-wide safety outcome: the Australian post-discharge cohort observed higher hospitalisation for genital mycotic infection, foot infection and osteomyelitis despite lower estimates for several severe infection endpoints. All eleven studies contributing to quantitative synthesis were judged to have moderate risk of bias, with residual confounding by frailty, kidney function, diabetes severity, prior infection, vaccination and healthcare access the principal concern. Future studies, the authors suggest, should emulate target trials with prespecified eligibility and negative-control outcomes, include more than one active comparator, and report validated severe-infection algorithms before any preventive claim could be entertained.

Subject of Research: Comparative severe infection outcomes of SGLT2 inhibitors versus DPP-4 inhibitors in type 2 diabetes

Article Title: Severe Infection Outcomes With SGLT2 Versus DPP‐4 Inhibitors in Type 2 Diabetes: A Systematic Review and Active‐Comparator Meta‐Analysis of Real‐World Cohorts

Article References: Huang, Y.-M., Wang, J.-H., Huang, W.-T., Hsu, J.-Z., & Teh, S.-H. (2026). Severe Infection Outcomes With SGLT2 Versus DPP ‐4 Inhibitors in Type 2 Diabetes: A Systematic Review and Active‐Comparator Meta‐Analysis of Real‐World Cohorts. Endocrinology, Diabetes & Metabolism, 9(6), Article e70381. https://doi.org/10.1002/edm2.70381

Image Credits: AI Generated

DOI: 10.1002/edm2.70381

Keywords: SGLT2 inhibitors, DPP-4 inhibitors, type 2 diabetes, pneumonia, sepsis, meta-analysis, systematic review, real-world evidence, heterogeneity, drug safety, observational cohorts, GRADE certainty