SGLT2 inhibitor oncology research expands as clinical proof remains uneven

A bibliometric and semantic analysis of 346 publications found that research on SGLT2 inhibitors in oncology has expanded rapidly since 2014, while the amount of indexed activity has grown faster than the strength of evidence for clinical benefit. The study combined records from Web of Science Core Collection and PubMed, finding that total output rose from four records in 2014 to 101 in 2025. Original articles increased from three to 67 over the same period. After accounting for the expanding volume of English-language biomedical articles indexed in PubMed, the authors estimated an annual incidence-rate ratio of 1.35 for article activity, with a 95% confidence interval of 1.28 to 1.43 and a nominal P value below 0.001. The authors emphasized that this index measures research attention, not treatment efficacy, causal benefit or evidence maturity.

The analysis, published in Frontiers in Oncology on October 21, 2026, was a study of bibliographic metadata rather than a clinical trial or patient-level outcome review. Records were retrieved from January 1, 2012, through July 26, 2026, with the first eligible publication appearing in 2014; the 2026 data were incomplete and were excluded from the main complete-year inference. The final audited corpus included 258 articles and 88 reviews. The article records were used for the primary BERTopic model, while a sensitivity model added reviews. Researchers also conducted publication, citation, co-citation, keyword and controlled-concept analyses, and built an exploratory evidence map assigning one label each for drug, oncologic context, research role and evidence tier.

The article-only BERTopic model assigned 222 of 258 records to 10 topics and retained 36 as outliers. Its largest topic, represented by terms associated with canagliflozin and cells, contained 82 records. Other topics involved patients and cancer, doxorubicin, cancer and SGLT2 inhibitors with DPP4 inhibitors, bladder cancer and renal-cell-carcinoma risk, nanoparticles, cisplatin and hepatocellular carcinoma. The Article-plus-Review sensitivity model identified 13 topics. After alignment of topic centroids, assignments for articles showed an adjusted Rand index of 0.634 and an adjusted mutual information score of 0.698, with a mean matched-centroid cosine similarity of 0.938. These results indicated substantial but imperfect stability after reviews were included. Topic labels were summaries of high-frequency terms, not validated biological groups or clinical categories.

The literature has developed along several parallel tracks rather than through a single replacement of one question by another. These tracks include cancer safety, mechanistic proposals for antitumor activity, real-world cancer outcomes and cardio-oncology supportive care. Citation patterns reflected this diversification. In the Web of Science subset of 283 publications, the cited-reference co-citation network contained 511 nodes and 1,323 links, with 11 clusters extending from early diabetes, hypoxic-tumor and glucose-lowering-agent themes through signaling pathways and hepatocellular-carcinoma models to cancer therapy-related cardiac dysfunction, antitumor potential and target-trial emulation. Citation bursts shifted toward cardiovascular outcomes and cardiotoxicity, indicating where scholarly attention was moving rather than proving that the newer questions had been settled.

Publication geography and journal networks showed where the field had become most active. China had the highest publication output, followed by the United States and Taiwan, while the United States, United Kingdom, Canada and Saudi Arabia had higher proportions of international collaboration. China Medical University ranked first among the listed institutions with 15 publications, whereas Massachusetts General Hospital had the highest citation count among the institutions shown, with six publications and 434 citations. The source journals spanned oncology, pharmacology, endocrinology and cardiovascular medicine. International Journal of Molecular Sciences ranked first in publication output with 12 papers, followed by Cancers and Scientific Reports with eight each, while Biomedicine & Pharmacotherapy had fewer papers but a higher citation count than several more productive journals.

Direct antitumor effects remain the least clinically mature part of the landscape. Experimental studies have proposed mechanisms involving glucose transport, mitochondrial respiration, AMPK–mTOR signaling, oxidative stress, apoptosis, autophagy, angiogenesis and immune modulation. In liver-cancer models, canagliflozin was associated with reduced glucose uptake and angiogenic activity, and experiments have also examined possible effects on chemoresistance. A mouse renal-cancer–sunitinib model reported that dapagliflozin improved cardiotoxicity phenotypes while enhancing tumor sensitivity to sunitinib. These findings generate testable biological hypotheses, but they do not establish a common anticancer mechanism or clinical efficacy. The authors also noted that concentrations required to inhibit tumor growth in laboratory systems may exceed clinically achievable free exposure at approved doses, and that expression of SGLT2 or SLC5A2 in a tumor does not itself demonstrate dependence on the transporter.

Clinical observations likewise require careful separation from causal effects. Observational studies have reported associations between SGLT2 inhibitor use and lower incidence of hepatocellular or renal-cell carcinoma, but the authors warned that these findings could be affected by confounding by indication, diabetes severity, treatment selection, healthcare utilization and competing mortality. Drug-target Mendelian randomization and observational analyses have also suggested a prostate-cancer risk signal, yet genetic instruments represent lifelong target perturbation whereas clinical cohorts reflect shorter-term exposure and do not estimate the same effect. In a hormone-therapy setting, SGLT2 inhibitor use was associated with later treatment failure, but overall and cancer-specific survival did not show clear differences. These results are hypothesis-generating and do not demonstrate cancer prevention or tumor-directed treatment.

Cardio-oncology has advanced most visibly toward prospective evaluation, although its immediate purpose is to support cardiovascular safety and continuity of cancer care rather than to establish SGLT2 inhibitors as anticancer drugs. Observational studies among patients receiving anthracyclines have reported fewer cardiac events with SGLT2 inhibitor use, while real-world comparisons in people with diabetes and cancer have suggested a lower risk of cancer therapy-related cardiac dysfunction. Treatment selection, differences in baseline cardiovascular risk and unequal access to care could still contribute to those associations. The EMPACARD-PILOT study observed a lower six-month incidence of cancer therapy-related cardiac dysfunction with empagliflozin in selected high-risk patients receiving anthracyclines, but its size and design were insufficient to replace a randomized trial.

The exploratory evidence map quantified the imbalance between the volume and maturity of the literature. Across all 346 publications, 125 records were classified as clinical observational studies, 107 as preclinical or mechanistic studies, 107 as evidence syntheses and seven as clinical interventional studies. All 88 audit-confirmed reviews were assigned to the evidence-synthesis tier. By research role, antitumor repurposing was the largest category, with 101 records, followed by clinical outcomes and management with 95, supportive or toxicity protection with 93, and risk or safety with 57. Two coders independently classified the records, achieving Cohen’s kappa values of 1.000 for agent, 0.972 for context, 0.965 for research role and 0.987 for evidence tier. The map described research composition; it did not estimate treatment effects or assess risk of bias.

Randomized-trial syntheses considered by the authors had not confirmed an overall class-wide increase in cancer risk, but the authors said uncertainty remained for rare events, site-specific risks, individual drugs and long latency periods. They identified several needs for stronger translation: drug-specific analyses, exposure levels that can be achieved clinically, evidence of target engagement, cancer-specific endpoints, appropriate latency periods and competing-risk methods. Cardio-oncology studies also need consistent definitions and imaging or biomarker endpoints, with concurrent reporting of cancer outcomes, completion of anticancer treatment, kidney outcomes and safety events. If a drug is proposed to have both cardioprotective and tumor-sensitizing effects, trials must demonstrate that one does not compromise the other. The analysis itself was limited by English-language restrictions, differences between databases, incomplete 2026 coverage, sensitivity to topic-modeling parameters and the tendency of citation counts to favor older publications. Its central finding is therefore a map of an expanding, increasingly diverse field whose clinical conclusions remain limited and uneven.

Subject of Research: Bibliometric and semantic analysis of SGLT2 inhibitor research in oncology from 2014 to 2026.

Article Title: From cancer safety signals to cardio-oncology translation: a dual-database bibliometric and BERTopic semantic analysis of SGLT2 inhibitor research in oncology (2014–2026)

Article References: Tang, X., Niu, W., Lan, Z., & Zhong, Z. (2026). From cancer safety signals to cardio-oncology translation: a dual-database bibliometric and BERTopic semantic analysis of SGLT2 inhibitor research in oncology (2014–2026). Frontiers in Oncology, 16, Article 1969057. https://doi.org/10.3389/fonc.2026.1969057

Image Credits: AI Generated

DOI: 10.3389/fonc.2026.1969057

Keywords: SGLT2 inhibitors, cancer safety, cardio-oncology, drug repurposing, bibliometrics, BERTopic, cancer metabolism, real-world evidence, anthracyclines, clinical translation, evidence mapping, oncology