Explore the Causal Effect of SGLT2 Inhibition on Venous Thromboembolism Events: A Drug-Target Mendelian Randomization Study With Pharmacovigilance Analysis

2.1 Selection of Genetic Instruments for SGLT2 Inhibition

After consulting the Drugbank database, SLC5A2 gene was selected as the target gene (Table S1). As illustrated in Figure 1, the process of identifying genetic variants relevant to SGLT2 inhibition involved four distinct phases. The initial phase relied on publicly accessible data from the eQTLGen Consortium ²⁵ to screen for genetic variants influencing the expression of SLC5A2. Moving to the second phase, we examined how each of these variants was correlated with glycated hemoglobin (HbA1c) levels, given that HbA1c serves as a key indicator of glycemic control. ²⁶ The genome-wide association study (GWAS) summary statistics of HbA1c were derived from 344,182 diabetes-free individuals of European ancestry participating in the UK Biobank (Table S2). Only those variants exhibiting a robust association with HbA1c (P < 5 × 10^-5) proceeded to the next stage. The third phase entailed colocalization testing to determine whether the same causal variant underlies both SLC5A2 expression and HbA1c variation, with a posterior probability exceeding 60% interpreted as positive evidence for a shared signal. ²⁶ Moreover, we applied LD-based clumping to the prioritized variants, referencing the 1,000 Genomes European panel and setting parameters at r² < 0.8 over a distance of 250 kb⁷. The strength of the genetic instruments was rigorously evaluated using the F-statistic; single nucleotide polymorphisms (SNPs) with an F-statistic below the conventional threshold of 10 were removed from the analysis to prevent weak instrument bias. ²⁷ OPEN IN VIEWER

2.2 Outcome Sources

As for the positive control data sets, we extracted type 2 diabetes from the IEU Open GWAS database, with OpenGWAS IDs named “ebi-a-GCST90029024”, “finn-b-T2D″, and “ebi-a-GCST90093109”. GWAS summary statistics for VTE, DVT, and PE were also sourced from the IEU Open GWAS database. The dataset for VTE comprised ebi-a-GCST90038607, finn-b-I9_VTE, and ukb-d-I9_VTE. The dataset for DVT comprised ukb-d-I9_PHLETHROMBDVTLOW and finn-b-I9_PHLETHROMBDVTLOW. Moreover, finn-b-I9_PULMEMB, ebi-a-GCST90038614, ukb-b-18366 and ukb-d-I26 were selected to serve as the GWAS data set for PE. The details of each outcome can be found in Table S2. The present study is based on publicly available statistics. Ethical approval and informed consent have been waived off.

2.3 MR and Summary-Data-Based Mendelian Randomization (SMR) Analysis

To reveal the causal effects of SGLT2 inhibitor with VTE-related traits, we performed MR analysis utilizing the ‘TwoSampleMR’ R package (version 0.6.22). The associations between cis-eQTLs and the outcomes were calculated in the form of odds ratios (ORs) and corresponding CIs by applying the inverse-variance weighted (IVW) method as the primary analysis. MR-Egger, weighted median, and weighted mode methods were additionally conducted. ²⁸ Moreover, positive control analyses were conducted to validate the genetic instruments for SGLT2 inhibitor, which was already approved by the FDA to control blood glucose and weight. Subsequently, we performed meta-analysis to pool the MR findings for VTE-, DVT-, and PE-related traits. ²⁹ As for the SMR analysis, we assessed the effect of the expression level of SLC5A2 using cis-expression quantitative trait loci (eQTL) dataset (eQTLGen) on VTE-related traits (finn-b-I9_VTE, finn-b-I9_PHLETHROMBDVTLOW, and finn-b-I9_PULMEMB) managed by the FinnGen consortium as the outcome using the default parameters via SMR portal.

2.4 FAERS Database Analysis

We conducted a retrospective pharmacovigilance study using data from the FDA Adverse Event Reporting System (FAERS). All reports from 2004Q1 to 2025Q4 were extracted and processed. The drug of interest was SGLT2 inhibitors, including canagliflozin, dapagliflozin, empagliflozin, ertugliflozin, and ipragliflozin, which were recorded as primary suspect (PS). Adverse events were identified using the Medical Dictionary for Regulatory Activities (MedDRA) preferred terms (PTs): “PULMONARY EMBOLISM” and “PULMONARY THROMBOSIS” for PTE and “DEEP VEIN THROMBOSIS” for DVT. We performed case/non-case analyses to estimate the association between SGLT2 inhibitors and PE/DVT. Adverse event reports for SGLT2 inhibitors were compared with all other drugs in the database. Disproportionality was assessed using the Reporting Odds Ratio (ROR) and the Information Component (IC) calculated as the previous published study. ²⁴ For clinicians to better understand the above terms, a Supplementary Glossary (Table S3) was provided.

2.5 Statistical Analysis

For the MR analyses, the MR-PRESSO was used to detect outlier genetic variants. In addition, heterogeneity and horizontal pleiotropy were assessed using Cochrane’s Q test (P < 0.05 indicating significant heterogeneity) and the MR-Egger intercept test (P < 0.05 indicating evidence of pleiotropy), respectively. For the meta-analysis, we used estimated ORs and 95% CI. I² was used to evaluate heterogeneity across included studies, and if I² > 50%, it is considered significant heterogeneity. To decrease the influence of heterogeneity on the accuracy, all the analyses are conducted using random effect model. Probability values of < 0.05 were considered statistically significant. For SMR analysis, a significant association was defined as P< 0.05 with no evidence that the association is driven by linkage disequilibrium (P_HEIDI > 0.05). The HEIDI (Heterogeneity in Dependent Instruments) test is used in SMR analysis to distinguish whether the observed association between gene expression and the outcome is due to a shared causal variant (i.e., the gene itself) or to linkage disequilibrium (i.e., nearby but distinct genetic variants). A P_HEIDI > 0.05 indicates no evidence of confounding by linkage disequilibrium, supporting that the SLC5A2 gene expression, rather than a neighboring gene, is likely responsible for the observed effect. All SMR and HEIDI analyses were performed using the SMR portal (https://yanglab.westlake.edu.cn/smr-portal/). As for the disproportionality analysis of FAERS, a signal was considered statistically significant if the lower limit of the 95% confidence interval for ROR > 1, and the lower limit of the 95% credible interval for IC > 0. All the analyses except SMR analysis were done with R software (V4.4.1).

3.1 Results of MR Analyses

We selected 6 significant cis-eQTL SNPs from the eQTLGen consortium as genetic instruments for the SLC5A2 gene expression. All instruments had F-statistics exceeding 10, indicating negligible weak instrument bias (Table S4). As Table 1 shows, genetically proxied SGLT2 inhibition was associated with a significantly reduced risk of type 2 diabetes (P < 0.05), confirming the well-established hypoglycemic effect of SGLT2 inhibition. The colocalization testing showed a posterior probability exceeding 60% representing positive evidence for a shared signal between SLC5A2 gene expression and HbA1c level. This protective causal association was consistently replicated across another 2 external GWAS validation datasets (all P < 0.05), validating the reliability of the selected genetic instruments and the MR analysis framework.

OPEN IN VIEWER

Table 1.

Positive Control Analyses of the Genetically Proxied SGLT2 Inhibition on Type 2 Diabetes

Exposure	Trait id	Trait names	Method	nsnp	b	Pval
SGLT2 inhibitor	ebi-a-GCST90029024	T2D	IVW	6	-0.01	2.67E-05
SGLT2 inhibitor	finn-b-T2D	T2D	IVW	6	-0.31	2.68E-05
SGLT2 inhibitor	ebi-a-GCST90093109	T2D	IVW	5	-0.29	1.68E-02

Genetically proxied SGLT2 inhibition showed no statistically significant causal association with the risk of VTE, DVT, or PE. The results are presented in Figure 2. For PE, analysis of the ebi-a-GCST90038614 dataset revealed a weak but statistically significant increased risk (OR = 1.004, 95% CI: 1.002–1.007, P = 0.001), whereas the finn-b-I9_PULMEMB, ukb-b-18366, and ukb-d-I26 datasets all yielded non-significant associations (ORs ranging from 0.795 to 1.000, all P > 0.05). In the VTE outcome, divergent results were also observed. The ebi-a-GCST90038607 dataset showed a significantly elevated risk (OR = 1.012, 95% CI: 1.008–1.016, P < 0.001), while the finn-b-I9_VTE dataset demonstrated a significant protective effect (OR = 0.658, 95% CI: 0.531–0.815, P < 0.001), and the ukb-d-I9_VTE dataset was non-significant (OR = 1.000, 95% CI: 0.997–1.004, P = 0.894). For lower-extremity deep vein thrombosis (DVT), the finn-b-I9_PHLETHROMBDVTLOW dataset identified a significant protective effect (OR = 0.642, 95% CI: 0.477–0.866, P = 0.004), while the ukb-d-I9_PHLETHROMBDVTLOW dataset was non-significant (OR = 1.000, 95% CI: 0.997–1.002, P = 0.733). No outlier genetic variants were detected via MR-PRESSO, and MR-Egger intercept tests indicated no horizontal pleiotropy across all analyses (all P > 0.05). Cochrane’s Q value assessment found no significant heterogeneity in IVW estimators for SGLT2 inhibition and VTE, DVT, or PE (all P > 0.05). The details of the above results can be found in Table S5-7.OPEN IN VIEWER

3.2 Meta-Analysis of MR Results and SMR Analyses

As displayed in Figure 2, in the meta-analysis of the above MR findings for VTE-related traits, we discovered significant inter-trait heterogeneity (all I²>50%). Thus, we selected the random-effects model for further estimations. The pooled ORs with 95% CIs further displayed the absence of a causal link between SGLT2 inhibition and VTE, DVT, or PE (all P > 0.05). During the SMR analyses, no significant association was observed between the SLC5A2 gene expression level and the risk of VTE, DVT, or PE (all P_SMR > 0.05) with no evidence of linkage disequilibrium confounding in all analyses (all P_HEIDI > 0.05). The details of the results can be found in Figure 3 and Table S8. The SMR analyses further supported the lack of a causal relationship between SGLT2 inhibition and VTE-related traits.OPEN IN VIEWER

3.3 FAERS Signal Mining Analysis

After screening for adverse event reports where SGLT2 inhibitors were listed as the PS, a total of 35,118 reports were identified (Figure 4). Among these, there were 83 cases of PE and 53 cases of DVT. Disproportionality analysis yielded a Reporting Odds Ratio (ROR) of 0.51 (95% CI: 0.42–0.64) for PE and 0.54 (95% CI: 0.45–0.64) for DVT, both of which were statistically significant (p < 0.001). The Information Component (IC) values were -0.95 (IC025: -1.26) for PE and -0.89 (IC025: -1.13) for DVT. The above results indicated no significant signals of SGLT2 inhibitors for either PE or VTE.OPEN IN VIEWER

4.1 Main Findings of the Study

In this study, we employed a multifaceted approach combining drug-target MR, SMR, and real-world pharmacovigilance analysis using the FAERS database to investigate the causal relationship between SGLT2 inhibition and the risk of VTE, DVT, and PE. The positive control analysis confirmed that SGLT2 inhibition was significantly associated with reduced risk of type 2 diabetes, consistent with established clinical evidence. This validation supports the reliability of our genetic instruments and MR framework. Our findings suggest that genetically proxied SGLT2 inhibition is not causally associated with these thrombotic outcomes after meta-analysis, despite some heterogeneity across individual GWAS datasets. The absence of a consistent causal effect is further supported by the SMR analyses. The FAERS disproportionality analysis revealed SGLT2 inhibitors did not emerge as significant signals for PE or DVT, which is consistent with the MR findings and supports the safety profile of SGLT2 inhibitors with respect to thrombotic risk.

Before performing the meta-analysis, we observed notable divergence across individual MR datasets. For VTE, the ebi-a-GCST90038607 dataset suggested a significantly elevated risk (OR = 1.012, P < 0.001), while the finn-b-I9_VTE dataset demonstrated a significant protective effect (OR = 0.658, P < 0.001), and the ukb-d-I9_VTE dataset was not significant. Similar inconsistencies were seen for DVT and PE. Several factors may drive these discrepancies. First, differences in population demographics across GWAS consortia may influence allele frequencies and LD patterns. Second, the inconsistency in VTE case definitions and ascertainment methods (e.g., hospital inpatient records vs. registry-based validated diagnoses) may introduce heterogeneity. Third, the statistical power varies considerably across datasets; datasets with smaller sample sizes or lower event rates may produce unstable estimates. Fourth, differences in genetic architecture, including the specific SLC5A2 cis-eQTL variants captured by each GWAS, could affect the precision of the genetic instruments. To address this heterogeneity, we employed a random-effects model for all meta-analyses, which accounts for between-study variance and yields more conservative pooled estimates. The pooled ORs remained non-significant across all thrombotic outcomes (all P > 0.05), supporting the overall safety profile.

4.2 Where we Stand

Due to their metabolic and cardiovascular benefits, SGLT2 inhibitors have become a cornerstone therapy for patients with diabetes mellitus. However, concerns regarding thromboembolic risk have emerged. The mild diuresis induced by SGLT2 inhibitors may lead to volume depletion and increased blood viscosity, potentially elevating VTE risk.^11,17 Conversely, SGLT2 inhibitors may also reduce VTE risk given their roles of reducing inflammation.^15,16

FDA safety signals are derived from systematic analysis of spontaneous adverse event reports (e.g., FAERS). Disproportionality methods, such as the ROR or IC, compare observed versus expected reporting frequencies for a given drug–event pair. A signal is declared when the lower bound of the 95% CI for ROR exceeds 1 (or IC > 0), indicating a disproportionately higher reporting rate that may warrant further investigation. Correspondingly, thromboembolic events have been reflected in a potential safety signal identified by the U.S. Food and Drug Administration (FDA) between April–June 2015 (https://www.fda.gov/Drugs/GuidanceComplianceRegulatoryInformation/Surveillance/AdverseDrugEffects/ucm484292.htm). During the past years, though there were several studies pharmacovigilance analysis of SGLT2 inhibitors, significant signals haven’t been reported from FAERS in terms of VTE risk.^30-33 Thus, pharmacovigilance specifically addressing VTE using the most recently published data is still lacking.

Several clinical studies have explored the efficacy and safety of SGLT2 inhibitors in terms of VTE, though there is a heterogeneity across the results of these studies without significant conclusions. For example, Patil et al ³⁴ demonstrated the use of SGLT2 inhibitors owned a HR of 1.03 when compared with glucagon-like peptide-1 receptor agonist (GLP-1RA). Another two trials indicated the incidence of VTE was low and did not differ between SGLT2 inhibitors and placebo.^35,36 Moreover, several meta-analyses also focused on such issue. For example, the potential for SGLT2 inhibitors to confer additional protective effects against VTE than dipeptidyl peptidase-4 (DPP4) inhibitors has been suggested by prior epidemiological evidence based on a nationwide population-based study and meta-analysis. ³⁷ A previous meta-analysis demonstrated no increased risk of VTE with SGLT2 inhibitors. ¹⁷ In contrast to these prior studies, our investigation offers several methodological advances. Unlike observational studies susceptible to confounding (obesity, advanced diabetes, and cardiovascular comorbidities are themselves independent risk factors for VTE), our drug-target MR design provides unconfounded estimates of lifelong SGLT2 inhibition on VTE risk.

In a MR analysis, Huang et al ³⁸ demonstrated SGLT2 inhibitors significantly reduced the risk of VTE using 13 proxy SNPs by the method of MR. Nevertheless, they only utilized one GWAS data set to draw the conclusion, which may cause bias. In contrast, our study employed a more robust framework by integrating multiple independent GWAS datasets, followed by meta-analyses to pool estimates across heterogeneous populations. Furthermore, we complemented MR with SMR analysis to validate findings at the SLC5A2 gene expression level and with real-world FAERS disproportionality analysis, both of which supported the absence of a causal association. Hence, our multi-dataset, triangulation-based framework represents a major methodological advance over the single-dataset MR analysis by Huang et al. Above all, most of the recent studies tend to show no significant association between SGLT2 inhibitors and VTE.

4.3 Where we Are Heading

Corresponding to the most previous studies, our study further proved the low risk of VTE induced by the SGLT2 inhibitors, which offers distinct methodological advantages over conventional clinical studies in evaluating the causal relationship between SGLT2 inhibition and VTE risk. Drug-target MR has emerged as a valuable tool for both identifying novel therapeutic targets and investigating drug-related complications, with proven utility across a range of diseases. ²⁶ Traditional observational studies are inherently susceptible to confounding by indication ³⁹ —where the underlying reasons for prescribing SGLT2 inhibitors (e.g., obesity, advanced diabetes, cardiovascular comorbidities) are themselves independent risk factors for VTE. ⁴⁰ In contrast, our MR approach utilizes genetic variants that randomly assort at conception to proxy SGLT2 inhibition, mimicking the randomization process of clinical trials and providing unconfounded estimates of lifelong drug exposure on disease outcomes.

Our approach captures the lifetime effect of SGLT2 inhibition, overcoming the limited follow-up (1–5 years) of clinical studies that may miss slowly accumulating thrombotic effects. Genetic variants reflect lifelong exposure to SGLT2 inhibition, enabling a more comprehensive assessment. We enhanced robustness through triangulation of multiple genetic methods, integrating standard MR, SMR, colocalization, and meta-analysis across independent GWAS datasets. SMR validated findings at the gene expression level, while colocalization ensured that shared causal variants drove the associations. Multiple outcome datasets allowed us to detect and quantify heterogeneity—an advantage over single observational studies or clinical trials.

4.4 Limitations

Several limitations should be considered. First, while drug-target MR proxies the lifelong effect of SGLT2 inhibition, this approach has inherent assumptions. Although we conducted comprehensive sensitivity analyses and colocalization, residual horizontal pleiotropy cannot be entirely excluded. Moreover, genetic instruments were derived from European populations. This is a significant limitation because genetic architecture (e.g., allele frequencies, linkage disequilibrium patterns), lifestyle factors, and VTE prevalence vary considerably across ethnic groups. Therefore, our findings may not be directly generalizable to non-European populations, which warrants future validation across diverse ancestral backgrounds. Second, MR estimates may not be directly comparable to shorter-term pharmacological effects in trials or real-world practice. Third, although SMR validated our findings at the gene expression level, it cannot distinguish causality from pleiotropy if the variant affects outcomes independently of SLC5A2 expression, despite the HEIDI test. Fourth, the FAERS pharmacovigilance analysis has inherent biases, including under-reporting, selective reporting, and lack of detailed clinical information (dosage, comorbidities, concomitant medications), precluding confounder adjustment or subgroup analyses. FAERS also cannot provide incidence rates—signals reflect disproportionality, not true risk. Finally, given that our MR analysis captures the lifetime effect of SGLT2 inhibition, and the FAERS analysis reflects post-marketing spontaneous reports, the possibility of a small effect size or rare thrombotic events cannot be definitively ruled out. Future large-scale prospective cohort studies or randomized controlled trials with extended follow-up periods and detailed adverse event adjudication are warranted to further validate our findings.