What this is
- This pilot network meta-analysis evaluates the risk of hematologic malignancies linked to GLP-1 receptor agonists and SGLT2 inhibitors.
- The analysis included 55 randomized controlled trials with over 200,000 participants.
- Findings indicate that dulaglutide may increase the risk of hematologic malignancies, while tirzepatide appears protective, especially against lymphoma.
Essence
- Dulaglutide is associated with a higher risk of hematologic malignancies, with an odds ratio (OR) of 2.18. In contrast, tirzepatide shows a protective effect, with an OR of 0.14, particularly for lymphoma.
Key takeaways
- Dulaglutide is linked to an increased risk of overall hematologic malignancies, with an OR of 2.18 (95% CI = 1.14–4.19). This suggests that patients using dulaglutide should be monitored for potential hematologic cancer development.
- Tirzepatide demonstrates a significantly lower risk of hematologic malignancies, with an OR of 0.14 (95% CI = 0.03–0.60). This indicates a potential protective effect, particularly against lymphoma.
- No significant associations were found between SGLT2 inhibitors and hematologic malignancy risk, suggesting these agents may not contribute to the development of blood cancers.
Caveats
- The average trial duration of approximately 129.6 weeks may not be sufficient to capture long-term malignancy risks, particularly for hematologic cancers.
- The focus on randomized controlled trials may exclude valuable data from observational studies, limiting the comprehensiveness of the findings.
- Heterogeneity in diagnostic methods across trials could affect the accuracy of pooled estimates for hematologic malignancies.
Simplified
1. Introduction
Glucagon-like peptide-1 (GLP-1) receptor agonists and sodium–glucose cotransporter 2 (SGLT2) inhibitors have emerged as innovative classes of antihyperglycemic agents, acting through mechanisms distinct from traditional therapies [1]. As the clinical application of these agents continues to grow, attention has also turned to their long-term safety profiles regarding cancer risk. While the relationship between these agents and solid malignancies [2] has been moderately explored, their influence on hematologic cancers remains largely uncharted. Previous reports had demonstrated that diabetes mellitus was associated with a higher risk of hematological malignancies and is an independent risk factor of all-cause and cause-specific mortality [3,4]. To date, only a few investigations have focused specifically on hematologic malignancies in the context of GLP-1 receptor agonist [5] or SGLT2 inhibitor [6] therapy. This knowledge gap is particularly significant given the generally poor prognosis associated with hematologic malignancies [7]. Fortunately, many large-scale randomized controlled trials (RCTs) have systematically captured adverse events, including hematologic cancers, within their safety assessments [8,9,10,11].
When RCTs are limited in duration or statistical power to assess rare adverse outcomes, meta-analytic methods can serve as a valuable tool to aggregate evidence. These methods harness data from extensive clinical trials and observational studies, offering insights that may reflect real-world clinical trends [12,13]. A well-executed network meta-analysis (NMA) is particularly advantageous, as it enables indirect comparisons among multiple interventions and facilitates the identification of superior regimens or doses [14]. Such approaches enhance evidence-based decision-making in clinical pharmacology.
Given the above, we undertook a pilot NMA to examine the comparative risks of hematologic malignancies associated with various GLP-1 receptor agonists and SGLT2 inhibitors. Our analytical framework builds upon previous studies examining neurological [15], colorectal cancer [2], metastatic cancer [16], and audiovestibular outcomes [17] related to these agents. This study represents the first NMA, to our knowledge, focusing specifically on hematologic malignancy risk in this drug class.
2. Methods
2.1. Study Design and Analytical Framework
This pilot NMA employed a hypothesis-driven design to preliminarily evaluate the hematologic malignancy risks linked to GLP-1 receptor agonists and SGLT2 inhibitors. Following the Cochrane Collaboration's guidelines for adverse event-focused synthesis [18], we concentrated on predefined outcomes rather than general safety profiles. The review was conducted in accordance with the PRISMA extension statement for network meta-analyses (PRISMA-NMA) [19] (refer to Table S1). Our study was prospectively registered on PROSPERO (CRD42024622310) and approved by the Tri-Service General Hospital's Institutional Review Board (TSGHIRB E202516007).
2.2. Literature Search Strategy
A broad and systematic literature search was conducted across eight major databases—PubMed, Embase, Cochrane CENTRAL, ClinicalTrials.gov, ProQuest, ScienceDirect, Web of Science, and ClinicalKey—up to 4 December 2024 (). Two independent reviewers (PT Tseng and BY Zeng) performed the selection process, starting with screening of titles and abstracts. Full-text evaluation was used for eligibility determination. Conflicts were resolved through mutual discussion. Additionally, reference lists from eligible reviews and prior meta-analyses were manually examined to identify supplementary studies. No filters for language were applied. Table S2
2.3. Eligibility Criteria
Study selection was guided by the PICOS model:
Population: Individuals without a prior diagnosis of hematologic malignancy
Intervention: Any GLP-1 receptor agonist or SGLT2 inhibitor, regardless of dosage
Comparator: Placebo, standard of care, or active control group
Outcome: Hematologic malignancy incidence
Study design: Randomized controlled trials (RCTs)
RCTs were eligible if they enrolled patients free from hematologic malignancy at baseline and evaluated the use of GLP-1 receptor agonists or SGLT2 inhibitors in human subjects, with either systematic adverse event tracking or explicit reporting of hematologic outcomes. Studies were excluded if they (1) involved pre-existing hematologic malignancies, (2) lacked a direct comparison group, (3) did not report hematologic events, (4) used a non-randomized design, or (5) were based on animal models. To reduce selective outcome reporting bias, only trials with structured monitoring of adverse events were retained [20].
2.4. Risk of Bias Assessment
Risk of bias was independently assessed by two reviewers using the Cochrane Risk of Bias Tool version 1.0 [21]. Any disagreement was resolved by consensus with a third investigator.
2.5. Outcome Definitions and Dose Categorization
Hematologic malignancies included in this pilot NMA were categorized as lymphoma, leukemia, and myeloma, based on definitions from the Global Burden of Disease project [22] and the 5th edition of the WHO classification of hematolymphoid neoplasms [23]. Dropout rate from trials, representing discontinuation for any cause, was used as a proxy for overall tolerability.
Medication dosages were stratified in accordance with definitions from their respective source trials:
Canagliflozin: Low: 100 mg; High: 300 mg.
Efpeglenatide: Low: 2 mg; Medium: 4 mg; High: 6 mg.
Ertugliflozin: Low: 5 mg; High: 15 mg.
Dapagliflozin: Low: 2.5 mg; Medium: 5 mg; High: 10 mg.
Injectable Semaglutide: Low: 0.5 mg; Medium: 1.0 mg; High: 2.4 mg.
Empagliflozin: Low: 1–10 mg; High: 25–50 mg.
2.6. Data Extraction and Handling
Two authors (PT Tseng and BY Zeng) independently extracted data on study design, patient demographics, intervention characteristics, malignancy outcomes, and trial discontinuations. Any unclear or missing information was clarified through correspondence with study authors. The extraction process followed Cochrane Handbook standards for systematic reviews [24].
2.7. Statistical Methodology
Random-effects models were utilized for all treatment comparisons involving more than one active comparator [25], and analyses were conducted using MetaInsight version 4.0.2 (Complex Reviews Support Unit, National Institute for Health Research, UK), which employs the netmeta package in R for frequentist NMAs [26].
When encountering single-zero-event arms, we applied a continuity correction to stabilize variance estimates, whereas studies with zero events in both arms were excluded to avoid underestimation of treatment effects [27,28]. Results were presented as odds ratios (ORs) with 95% confidence intervals (CIs), displayed via forest plots [29]. A network graph was constructed to visualize direct and indirect treatment connections. To assess agreement between direct and indirect comparisons, we implemented the node-splitting approach [26,30]. Statistical significance was defined as a two-sided p-value < 0.05.
2.8. Sensitivity and Subgroup Analyses
We conducted stratified analyses based on hematologic malignancy subtype—lymphoma, leukemia, and myeloma—to examine treatment-specific patterns. Further, a Bayesian NMA framework was used to validate the frequentist findings and assess model stability. Surface under the cumulative ranking curve (SUCRA) values were calculated to determine treatment hierarchy using Litmus Rank-O-Gram and radial SUCRA plots [31]. A deviation-from-model-fit analysis was conducted to assess potential inconsistency in treatment effects [32]. Finally, we applied the GRADE framework to evaluate the certainty of evidence across network estimates [33].
2.9. Ethics
All procedures adhered to the ethical principles of the Declaration of Helsinki.
3. Results
3.1. Study Selection and Characteristics
The literature review and selection process are illustrated in Figure 1. After removing 90 articles for failing to meet inclusion criteria (see Table S3), 54 publications encompassing 55 randomized controlled trials (RCTs) were included for analysis (Table S4) [8,9,10,11,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,66,67,68,69,70,71,72,73,74,75,76,77,78,79,80,81,82,83]. These trials enrolled a total of 200,606 participants, with a mean age of 62.7 years (range: 41.2 to 71.9 years) and an average female proportion of 38.2% (range: 20.8–79.0%). The median follow-up period was 129.6 weeks, ranging from 12 to 281 weeks.
Across the included studies, we identified 24 distinct treatment arms: one standard-of-care or placebo group, and 23 active comparator groups representing various GLP-1 receptor agonists and SGLT2 inhibitors across multiple dosing regimens. GLP-1 receptor agonists examined included tirzepatide, efpeglenatide, liraglutide, albiglutide, dulaglutide, exenatide, semaglutide, and lixisenatide. SGLT2 inhibitors included bexagliflozin, canagliflozin, empagliflozin, ertugliflozin, dapagliflozin, and sotagliflozin.
3.2. Primary Outcome: Incidence of Overall Hematologic Malignancies
In this pilot NMA preliminarily evaluating all hematologic malignancies combined, dulaglutide was associated with a significantly elevated risk compared to control (OR = 2.18; 95% CI = 1.14–4.19). In contrast, tirzepatide demonstrated a significantly protective association, with markedly lower odds of hematologic malignancy compared to the control group (OR = 0.14; 95% CI = 0.03–0.60). Among all agents assessed, tirzepatide ranked as the safest in this domain (see Figure 2 and Figure 3, Figure S3A, and Table 1).
3.3. Subgroup Analysis: Specific Hematologic Cancer Types
When broken down by specific cancer subtype, tirzepatide continued to demonstrate a statistically significant reduction in lymphoma risk (OR = 0.18; 95% CI = 0.03–0.93) compared to the control arm. No other agent showed a significant increase or decrease in lymphoma risk relative to control (see). Figures S1A, S2A and S3B, and Table S5A
As for the risk of leukemia and myeloma, none of the investigated interventions exhibited statistically significant associations when compared with control groups (). Figures S1B,C, S2B,C and S3C,D, and Table S5B,C
3.4. Safety Profile: Dropout Rate
Dropout rate was used as a surrogate marker of overall tolerability. Several agents were associated with significantly lower discontinuation rates than the control group, including liraglutide (OR = 0.50; 95% CI = 0.38–0.66), tirzepatide (OR = 0.56; 95% CI = 0.42–0.74), medium-dose injectable semaglutide (OR = 0.65; 95% CI = 0.43–0.98), high-dose injectable semaglutide (OR = 0.71; 95% CI = 0.52–0.96), and high-dose empagliflozin (OR = 0.74; 95% CI = 0.56–0.97). Among all agents, liraglutide ranked highest for treatment retention, followed by tirzepatide (D). Figures S1D, S2D and S3E, and Table S5
3.5. Robustness Evaluation: Bayesian-Based Sensitivity Analysis
To verify the consistency of findings, we performed parallel analyses using a Bayesian framework. The direction and magnitude of treatment effects in Bayesian NMA aligned closely with those obtained from frequentist models (A–E). Drug rankings by Bayesian SUCRA values are displayed inA–E andA–J. No significant deviation was detected in model fit across treatment comparisons as assessed by deviation-from-consistency modeling (A–O). Figure S4 Table S6 Figure S5 Figure S6
3.6. Risk of Bias and Inconsistency Assessment
Across all studies, 78.7% (303 of 385 items) were judged to have a low risk of bias, 15.8% (61 items) as unclear, and 5.5% (21 items) as high risk (A,B). Evaluation of consistency between direct and indirect evidence did not reveal significant discrepancies (A–E). The overall quality of evidence supporting the primary and secondary outcomes was rated as moderate to high using the GRADE approach (A–E). Figure S7 Table S7 Table S8
4. Discussion
This pilot NMA offers preliminary views into the distinct hematologic safety profiles of GLP-1 receptor agonists and SGLT2 inhibitors. Among the GLP-1 receptor agonists, tirzepatide stood out for its association with a significantly reduced incidence of hematologic malignancy—especially lymphoma—while dulaglutide was linked to an elevated risk. In contrast, no significant associations were detected between any SGLT2 inhibitor and the risk of blood cancers. The divergent trends observed between tirzepatide and dulaglutide are particularly noteworthy, considering their shared classification within the GLP-1 receptor agonist family. These findings suggest that potential hematologic effects may differ considerably across agents, even within the same drug class.
One of the most prominent observations in this pilot study was the elevated risk of hematologic malignancy associated with dulaglutide. A possible mechanistic explanation may involve dulaglutide's influence on apoptosis-related signaling cascades. Preclinical studies have suggested that dulaglutide can enhance anti-apoptotic activity by modulating the BCL2/BAX/c-caspase3 axis, a pathway implicated in tumor progression and cell survival [84]. Though a case linking dulaglutide to medullary thyroid carcinoma has been reported [85], evidence for its association with hematologic malignancy is still lacking in clinical settings. Another theoretical mechanism involves dulaglutide's impact on gene methylation. One study reported that dulaglutide induces demethylation of CDH1, the gene encoding E-cadherin—a cell adhesion molecule whose dysregulation has been implicated in certain leukemias [86,87]. While these molecular signals could hypothetically explain the observed associations, no direct clinical evidence has yet confirmed causality. Therefore, the potential linkage between hematologic malignancy and such medication still remained early speculation. Thus, further clinical and mechanistic investigations are required to establish a clearer link between dulaglutide use and hematologic cancer development.
Conversely, tirzepatide was associated with a protective effect against overall hematologic malignancies, most notably in reducing lymphoma risk. In fact, the tirzepatide exerted a different reaction on GLP-1 receptor from those of traditional GLP-1 receptor agonists. Tirzepatide is a biased agonist of GLP-1 receptor with significantly weaker affinity to GLP-1 receptor than natural GLP-1 receptor agonists (such as dulaglutide) [88]. Specifically, tirzepatide could act more on enhancing cAMP expression but less on β arrestin 2 recruitment [89]. Previous reports had demonstrated the potential role of β arrestin 2 on carcinogenesis [90]. Although evidence from clinical trials on tirzepatide's role in oncology is still sparse [91], early preclinical studies suggest it may inhibit oncogenic pathways. For example, Kong et al. demonstrated that tirzepatide exerted weight-dependent anti-tumor effects in mice, modulating metabolic and signaling pathways based on the animals' baseline obesity status [91]. In obese mice, the compound suppressed glycolytic and ErbB signaling—two pathways linked to cancer proliferation. In leaner models, tirzepatide enhanced O-linked glycosylation and phospholipase D signaling. Overactivation of ErbB pathways is well-documented in multiple tumor types [92], and inhibition of this axis is an emerging strategy in cancer therapeutics [93]. Although these findings are promising, they remain largely theoretical and early speculated based on animal models. Human studies explicitly designed to assess tirzepatide's potential anti-cancer effects are still lacking and urgently needed. Future large-scale clinical trials and NMAs should be warranted to support or refute the result of our study.
Strengths and Limitations
Our analysis possesses several methodological advantages that enhance its clinical relevance. The use of NMA enabled us to perform indirect comparisons across a broad spectrum of GLP-1 receptor agonists and SGLT2 inhibitors, generating more granular insights than conventional meta-analyses. By restricting our inclusion criteria to RCTs, we ensured that the source data were of high methodological quality. Furthermore, we excluded patients with pre-existing hematologic malignancies, allowing for a more accurate assessment of drug-related causality or prophylaxis. We also carried out malignancy-type-specific subgroup analyses, providing more detailed results to inform targeted clinical strategies. Lastly, Bayesian sensitivity analysis reinforced the validity of our findings, as results remained consistent across modeling frameworks.
Nonetheless, several limitations merit discussion. First, the average trial duration of approximately 129.6 weeks (around 2.5 years) may be insufficient to fully capture the long-term risk of developing malignancy. Although the hematologic malignancy might have shorter period of latency (around 2 years) than solid tumors, the relatively shorter trial duration might still limit the property of hematologic malignancy detection []. Second, while our focus on RCTs enhances internal validity, it may exclude useful data from well-conducted observational studies. Although we recognized the fact that the numbers of included studies would matter a lot in clinical practice, we kept to only include RCTs due to avoidance of heterogeneity of study designation. Third, heterogeneity in diagnostic methodology across international trials could introduce variability in how hematologic cancers were identified, potentially affecting accuracy in pooled estimates. Fourth, given the small number of target events, the main results of this NMA should be interpreted with caution. Additionally, most included RCTs were not primarily designed to evaluate hematologic cancer outcomes, as they typically addressed glucose control in diabetes or weight reduction in obesity. Therefore, malignancy data may have been underreported or inconsistently monitored. Since adverse event reporting often receives less rigorous scrutiny than primary outcomes, there's a possibility that hematologic events may have been missed or incompletely recorded. In addition, they were not initially designed to exclude the potential confounding factors, such as drug interactions, tolerance of these antidiabetics and further risk factors of these malignancies. Therefore, although the patient characteristic data of the current study fell within the hematologic malignancy range (i.e., age, gender distribution), we could not make further exploration regarding the confounding effects related to patient characteristics. 94
5. Conclusions
In summary, this pilot NMA provides preliminary evidence suggesting agent-specific differences in hematologic malignancy risk among GLP-1 receptor agonists. Specifically, dulaglutide was linked to a significantly higher incidence of overall hematologic malignancies, whereas tirzepatide demonstrated a notable risk-reducing effect—particularly for lymphoma. In contrast, no meaningful associations were identified for SGLT2 inhibitors regarding hematologic cancer risk. However, given the nature of the preliminary findings of this pilot NMA, we could only view the results as an exploratory finding but not final conclusion. Future research should prioritize large-scale RCTs directly addressing the risk of hematologic malignancy related to specific GLP-1 receptor agonists prescription. Further, to support or refute the results of the current NMA, large-scale and comprehensive NMA addressing various new generation antidiabetic agents should be warranted.