Sodium–Glucose Cotransporter 2 Inhibitors and the Short-term Risk of Bladder Cancer: An International Multisite Cohort Study

Sodium–glucose cotransporter 2 (SGLT2) inhibitors are the newest drug class for the treatment of type 2 diabetes. First introduced in 2013, these drugs have become increasingly popular due to their low risk of hypoglycemia (1–3) and favorable or neutral effects on body weight compared with older antidiabetes drugs (1,3). Most notably, SGLT2 inhibitors have been shown to significantly reduce the risk of cardiorenal and mortality events in randomized controlled trials (RCTs), and thus, in addition to glucose control, they have been recently approved for the prevention of cardiovascular and renal disease (4–9).

Despite their cardiovascular benefits, their safety profile was initially a contentious issue. In premarketing RCTs of the SGLT2 inhibitor, dapagliflozin, there was a numerical imbalance in bladder cancer events in patients randomized to dapagliflozin versus placebo, with events occurring shortly after randomization (range 43–727 days) (10–12). This concern prompted the U.S. Food and Drug Administration to delay the approval of dapagliflozin in 2011 (10). Subsequent postmarketing RCTs of SGLT2 inhibitors have generated mixed findings, ranging from a numerical imbalance of bladder cancer events with empagliflozin (5,12), a neutral effect with canagliflozin (4), to a decreased incidence with dapagliflozin (6).

While the numerical imbalances in bladder cancer events observed in some of these RCTs relatively shortly after randomization could be due to chance, there are alternative hypotheses suggesting that the rapid timing of these events could also be explained by either an overdetection of prevalent bladder cancers in the intervention arm or through tumor promotion of existing cancer cells (13). To date, only one observational study has been conducted to address this safety signal in the real-world setting, but focused on bladder cancer events occurring after 1 year of treatment initiation (14).

Given the severity of bladder cancer, a disease with a high recurrence rate and relatively poor survival (15,16), and the new recommendations to preferentially prescribe SGLT2 inhibitors in patients with type 2 diabetes and established cardiovascular disease, chronic kidney disease, or at high cardiovascular risk (17,18), addressing this safety signal is of importance. Thus, the objective of this multisite population-based cohort study is to determine whether the use of SGLT2 inhibitors, compared with the use of glucagon-like peptide 1 receptor agonists (GLP-1RAs) or dipeptidyl peptidase 4 (DPP-4) inhibitors, is associated with an increase in the short-term risk of bladder cancer among patients with type 2 diabetes.

This international, multisite, cohort study used four databases from the U.K. and the U.S. The databases consisted of one primary care database, U.K. Clinical Practice Research Datalink (CPRD), and three U.S. administrative databases, including Medicare fee-for-service and two commercial health insurance databases (Optum’s de-identifed Clinformatics Data Mart Database [CDM] and MarketScan Health [IBM Corp]). The U.K. CPRD included the GOLD and Aurum databases, which contain the complete primary care medical records of >60 million patients across the U.K. (19). The CPRD was linked to the Hospital Episode Statistics Admitted Patient Care database, which contains records of inpatient encounters in National Health Services hospitals (20), and to the Office for National Statistics, a database of electronic death certificates (21). Medicare fee-for-service is a U.S. federal health insurance program that provides health care to American citizens at least 65 years of age and patients with disabilities, including records of ∼50 million adults, whereas MarketScan and CDM include data from U.S. employer-sponsored health plans, with nationwide commercial coverage including Medicare Advantage plans for >70 million individuals. For each enrolled individual, the three U.S. data sources contain demographic information, health plan enrollment status, longitudinal patient-level information on all reimbursed medical services, including inpatient and outpatient diagnoses and procedures, and pharmacy dispensing records, with information on medication start and refill, strength, quantity, and days’ supply.

The study protocol was approved by the Research Data Governance of the CPRD (protocol number 21_000718) and by the Jewish General Hospital (Montreal, Quebec, Canada) Research Ethics Board and the Brigham and Women's Hospital Institutional Review Board (Boston, MA).

Within each database, we assembled two new-user, active comparator cohorts of patients newly treated with 1) SGLT2 inhibitors or GLP-1RAs and 2) SGLT2 inhibitors or DPP-4 inhibitors from the time the first SGLT2 inhibitor entered the market at each site (U.K.: January 2013, U.S.: April 2013) until the end of data availability (CPRD: December 2019; Medicare: September 2018; MarketScan: September 2018; CDM: June 2020). Cohort entry was defined as the date of the first prescription of either study drug, within each cohort, during the study period. Patients concurrently exposed to SGLT2 inhibitors and a comparator drug on cohort entry were excluded from the analysis. We selected GLP-1RAs and DPP-4 inhibitors as the comparators because they are clinically relevant comparators used at similar disease stage as SGLT2 inhibitors (22) and have not been associated with an increased risk of bladder cancer (23,24).

In each cohort, patients were required to have at least 1 year of medical history or coverage in the database before cohort entry; this served as a minimum washout period to ensure incident drug use. We used the entirety of data availability as the look-back period for all covariates. We excluded patients younger than the age of 40 (and 66 in Medicare) at cohort entry, patients previously diagnosed with bladder cancer (either malignant or in situ), and those diagnosed with end-stage renal disease or undergoing dialysis, as these are contraindications to SGLT2 inhibitor use. Finally, we excluded organ transplant recipients and patients with a history of pioglitazone use; the latter is now infrequently used and has been associated with an increased risk of bladder cancer (25). In the U.S. databases, we required patients to have a diagnosis of type 2 diabetes on or before cohort entry and excluded patients with a nursing home admission before cohort entry, as exposures cannot be accurately ascertained during nursing home admissions.

Patients in the cohorts were considered continuously exposed to their cohort entry drug starting 1 day after cohort entry until an incident diagnosis of bladder cancer (defined in detail below), death from any cause, end of coverage or registration, or end of the study period, whichever occurred first. This exposure definition does not consider treatment terminations, as it assumes that patients may remain at risk long after treatment discontinuation, which is possible for an outcome such as cancer.

In the linked CPRD data set, malignant and in situ bladder cancer events were defined by the first of either a primary care diagnosis (identified using Read codes) (Supplementary Table 1) or an inpatient diagnosis (ICD-10 codes) (Supplementary Table 2) during the follow-up period. For the U.S. databases, bladder cancer events were defined using two outpatient or inpatient diagnoses (identified using International Classification of Diseases Ninth Revision code [Supplementary Table 3] or ICD-10 codes) occurring within 60 days, with the outcome date defined by the date of the first code. Bladder cancer codes have been shown to be well recorded in the CPRD, with sensitivities of at least 83% and specificities as high as 99% (26–28). Similarly, solid tumors such as bladder cancer have been shown to be well recorded in claims databases, with sensitivities above 85.0% and specificities as high as 98.0% (29,30).

To use the strengths of each database, we considered slightly different confounders in each database, rather than a universal set, measured at any time on or before cohort entry. All databases included age, sex, year of cohort entry, alcohol-related disorders, smoking history, bladder conditions (cystitis, urogenital infections, bladder stones), history of cancer (other than nonmelanoma skin cancer), and chronic obstructive pulmonary disease or emphysema (as proxies for heavy smoking). We also included measures of health care use and markers of health-seeking behaviors such as the uptake of cancer screening (fecal occult blood testing or colonoscopy, mammography, prostate-specific antigen testing) and vaccinations (including influenza and pneumococcal vaccinations). All databases included proxies of diabetes severity, including microvascular (nephropathy, neuropathy, retinopathy) and macrovascular (myocardial infarction, stroke, peripheral arterial disease) complications of diabetes, and previous use of antidiabetes drugs, including combination drugs at baseline (drug classes entered as nonmutually exclusive variables in the model). The CPRD also included duration of diabetes and hemoglobin A_1c levels, while the U.S. databases included race, geographic region, frailty, and other measures of health care utilization, including the number of physician and urologist visits, and out of pocket pharmacy costs. A complete list of confounders included in each database can be found in Supplementary Tables 4–11.

We used propensity score-based fine stratification weighting to control for confounding, an analytical approach that efficiently deals with confounding with infrequent exposures (31). Weighting methods are preferred to matching when there is an unequal distribution of patients by exposure status. Multivariable logistic regression was used to estimate propensity scores conditional on the covariates listed above, and patients in nonoverlapping regions were trimmed. We created 50 strata using the distribution in the SGLT2 inhibitor group, and patients in each comparator were assigned to these strata using their propensity score values (31). Patients exposed to SGLT2 inhibitors were assigned a weight of one, and comparator patients were weighted in proportion to the number of SGLT2 patients in their assigned stratum (31). Thus, comparator patients were weighted to represent SGLT2 inhibitor–exposed patients. Hazard ratios (HRs) generated from propensity score-based fine stratification describe the average treatment effect in the treated, that is, the average causal effect of treatment in the SGLT2 inhibitor cohort. Covariate balance was assessed using standardized differences, with differences of <0.10 considered acceptable (32).

For each exposure group, we calculated weighted incident rates of bladder cancer, with 95% CIs based on the Poisson distribution. Weighted Kaplan-Meier curves were plotted to display the cumulative incidence of bladder cancer over the follow-up period for each exposure group. Finally, weighted Cox proportional hazards models were fit to estimate marginal HRs of bladder cancer with 95% CIs, using robust variance estimators, comparing the use of SGLT2 inhibitors with the use of GLP-1RAs and DPP-4 inhibitors, separately.

We performed four secondary analyses. In the first, we used an on-treatment exposure approach, whereby patients were followed while continuously exposed to their cohort entry drug, allowing for a 1-year grace period between nonoverlapping consecutive prescriptions to account for diagnostic delays. In the second, we considered time since treatment initiation by stratifying the follow-up into three predefined categories (≤1 year, 1.1–2 years, and >2 years). Third, we stratified the exposure by type of SGLT2 inhibitor (dapagliflozin, empagliflozin, or canagliflozin) to determine whether there were any drug-specific effects. Finally, we considered whether there is effect measure modification by sex, age (<50, 50–74, and ≥75 years), and history of urogenital infection by including an interaction term in the primary model between exposure status and these variables.

We conducted two sensitivity analyses to assess the robustness of our findings. First, to examine the impact of residual heterogeneity across data sets, we repeated the analysis using a universal propensity score adjustment restricted to covariates that are present in all databases. Lastly, we repeated the analyses without in situ cancers to investigate the impact of increased detection.

All analyses were conducted using a common analytical protocol in each of the four databases. To account for the inherent heterogeneity of the different data sources, results were pooled using random-effects models with inverse variance weighting. As an additional sensitivity analysis, results were pooled using fixed-effects models. Heterogeneity was assessed using the I² statistic. All analyses were conducted with SAS version 9.4 (SAS Institute, Cary, NC), R version 4.1.3 (R Foundation for Statistical Computing, Vienna, Austria), and the Aetion Evidence Platform version 4.48.

Part of this study is based on data from the CRPD obtained under license from the U.K. Medicines and Healthcare products Regulatory Agency. The data are provided by patients and collected by the U.K. National Health Service as part of their care and support. The interpretation and conclusions contained in this study are those of the author(s) alone. Because electronic health records are classified as “sensitive data” by the U.K. Data Protection Act, information governance restrictions (to protect patient confidentiality) prevent data sharing via public deposition. Data are available with approval through the individual constituent entities controlling access to the data. Specifically, the primary care data can be requested via application to the CPRD (https://www.cprd.com). No additional data are available.

The first cohort included 453,560 and 375,997 SGLT2 inhibitor and GLP-1RA users, respectively (Supplementary Figs. 1–4). After pooling, the cohorts generated 1,978 incident bladder cancer events during 1,684,049 person-years of follow-up (crude incidence rate: 117.5, 95% CI 112.3–122.7 per 100,000 person-years). The median follow-up for this cohort across the databases ranged from 1.5 to 2.2 years.

Table 1 presents selected pooled characteristics of the exposure groups before and after weighting (a complete list of the characteristics by site can be found in Supplementary Tables 4–8). Before weighting, the exposure groups had similar ages and a history of bladder conditions. SGLT2 inhibitor users had a shorter duration of diabetes and were less likely to have microvascular complications of diabetes and use insulin compared with GLP-1RA users. After weighting, the covariates were balanced between the exposure groups with standardized differences <0.10 (Supplementary Tables 4–8).

Table 2 summarizes the results of the primary analysis. Overall, SGLT2 inhibitors were not associated with an increased risk of bladder cancer compared with GLP-1RAs (weighted incidence rate [95% CI] 95.9 [58.1–158.2] vs. 94.4 [54.2–164.3] per 100,000 person-years, respectively; pooled HR 0.90, 95% CI 0.81–1.00). The cumulative incidence of bladder cancer was similar across the four databases (Supplementary Figs. 5–8).

In secondary analyses, the HR was consistent when an on-treatment exposure definition was used (HR 0.86, 95% CI 0.74–1.00) (Fig. 1). There were no meaningful differences in the HRs by SGLT2 inhibitor drug type, and there was no evidence of a duration-response relation (Fig. 1 and Supplementary Tables 12 and 13). The association was not modified by sex, age, or history of urogenital infection (Fig. 1 and Supplementary Fig. 9). Finally, the results were highly consistent when the primary analysis was repeated using the universal propensity score, restricting to malignant cancer events, and pooling using a fixed-effects model (Fig. 1 and Supplementary Table 14).

In the second cohort, there were 347,059 and 853,186 SGLT2 inhibitor and DPP-4 inhibitor users, respectively (Supplementary Figs. 10–13); these patients were followed for a median of 1.6 to 2.6 years. After pooling, there were 4,164 incident bladder cancer events during 2,755,807 person-years of follow-up, yielding a crude incidence rate of 151.1 (95% CI 146.5–155.8) per 100,000 person-years.

Selected pooled characteristics of the SGLT2 inhibitor and DPP-4 inhibitor cohorts are presented in Table 3. Before weighting, the exposure groups were similar on sex, bladder conditions, cancer, and history of myocardial infarction. SGLT2 inhibitor users were younger than DPP-4 inhibitor users and less likely to have a history of peripheral arterial disease and stroke. SGLT2 inhibitor users had a higher prevalence of obesity and were more likely to have used GLP-1RAs and insulin compared with DPP-4 inhibitor users. Site-specific baseline characteristics are presented in Supplementary Tables 8–11, with all covariates achieving balance after weighting.

The results of the primary analysis are illustrated in Table 4. Overall, the use of SGLT2 inhibitors was not associated with the incidence of bladder cancer compared with the use of DPP-4 inhibitors (weighted incidence rate [95% CI] 92.7 [52.3–164.4] vs. 91.4 [46.8–178.3] per 100,000 person-years, respectively; pooled HR 0.99, 95% CI 0.91–1.09). There was no difference in the cumulative incidence of bladder cancer in SGLT2 inhibitor users compared with DPP-4 inhibitor users (Fig. 1 and Supplementary Figs. 14–17).

Similar results were obtained using an on-treatment exposure definition, duration categories, and SGLT2 inhibitor drug type (Fig. 1 and Supplementary Tables 15 and 16). Similarly, the association was not modified by sex, age, or history of urogenital infection (Fig. 1 and Supplementary Fig. 18). Finally, the results of the sensitivity analyses were highly consistent with the primary analysis (Fig. 1 and Supplementary Table 17).

In this international, multisite cohort study, the use of SGLT2 inhibitors was not associated with a short-term increased risk of bladder cancer compared with GLP-1RAs or DPP-4 inhibitors. The study findings were consistent across a range of predefined subgroup and sensitivity analyses. Overall, contrary to some RCTs, the findings of this large real-world study suggest that SGLT2 inhibitors are not associated with an early increased risk of bladder cancer.

The existing evidence on the association between SGLT2 inhibitors and bladder cancer has been primarily generated from RCTs. In premarketing RCTs of dapagliflozin, there was a numerical imbalance in bladder cancer events in patients randomized to dapagliflozin versus placebo (9 of 6,045 vs. 1 of 3,512, respectively; rate ratio: 5.17, 95% CI 0.68–233.55), with all events occurring relatively soon after randomization (range 43–727 days) (10). Following the release of other SGLT2 inhibitors, postmarketing RCTs have generated mixed findings with respect to bladder cancer incidence. Indeed, there were more bladder cancer events in patients randomized to empagliflozin 25 mg versus placebo (9 of 2,342 and 5 of 2,333, respectively) (12), although removing early events lessened this imbalance in a post hoc analysis (33). Moreover, there was no imbalance of bladder cancer events in the canagliflozin trial (canagliflozin: 1.0 per 1,000 vs. placebo: 1.1 per 1,000 person-years) (4). Finally, in the Dapagliflozin Effect on Cardiovascular Events–Thrombolysis in Myocardial Infarction 58 (DECLARE-TIMI 58) trial, dapagliflozin was associated with a 43% decreased risk of bladder cancer compared with placebo (HR 0.57, 95% CI 0.35–0.93) (6). This decreased risk may be explained by the selective patient population of this trial, which excluded patients with a history of bladder cancer or at a high risk of bladder cancer in response to the concerns from premarketing RCTs (6).

A meta-analysis of 46 RCTs found that SGLT2 inhibitors were associated with an almost fourfold increased risk of bladder cancer (SGLT2 inhibitors: 18 events, comparator: 1 event; odds ratio 3.87, 95% CI 1.48–10.08) (34). However, this estimate was based on very few events and did not include results from two of the three large cardiovascular outcome trials (5,6). Most recently, an observational study using Scandinavian databases assessed the risk of bladder and renal cancer among SGLT-2 inhibitor users and GLP-1RA users (14). After lagging the outcomes by 1 year, there was no increased risk of bladder cancer among SGLT-2 inhibitors (HR 0.88, 95% CI 0.59–1.31) (14). However, this study was limited by its small sample size (73 bladder cancer events in 154,999 patients) and was not designed to specifically address the early bladder cancer signal that was generated in RCTs (14).

Contrary to some RCTs, our study did not find an increased risk of bladder cancer with the use of SGLT2 inhibitors compared with GLP1-RAs or DPP-4 inhibitors. This may be explained through differences in study design, as RCTs are not designed or powered to address cancer end points. Thus, it is possible that the imbalances observed in RCTs were due to chance, as there were very few bladder cancer cases in each trial. An alternative hypothesis for the apparent imbalances observed in RCTs is an overdetection of prevalent bladder cancer (13); this could be through increased urinalysis testing to diagnose suspected urinary and genital tract infections in SGLT-2 inhibitor users, a known adverse effect of this drug class (35–40). However, the lack of an association in our study argues against this hypothesis or a potential tumor promoter effect. Moreover, the pooled incidence rate of bladder cancer in our cohorts is consistent with what is to be expected among patients with type 2 diabetes (25,41), which provides additional reassurance on the safety of SGLT2 inhibitors.

This study has several strengths. First, our multisite approach provided us with large cohorts to sufficiently power all primary and most secondary analyses, while increasing the generalizability of our findings. We also observed consistent results across data sources and different populations. Second, we used two active comparators for our analyses, which reduced confounding by indication and present clinically meaningful comparisons. Third, we used a new-user study design, which eliminated biases common in pharmacoepidemiologic studies (42,43). Finally, the use of propensity score fine stratification weighting maximized the size of the cohorts, while achieving good balance between the exposure groups.

This study has some limitations. First, the different databases captured prescriptions that were either dispensed (U.S. Medicare, CDM, MarketScan) or written by general practitioners (U.K. CPRD). As such, exposure misclassification is possible if patients did not adhere to treatment or if patients switched between drug classes during follow-up, although this possible misclassification is expected to be nondifferential between the exposure groups.

Second, while solid tumor diagnoses have been shown to be well recorded in the different databases, outcome misclassification is possible, but this is likely nondifferential between the groups and should not bias a relative risk given the high specificity of bladder cancer (44).

Third, given the observational nature of this study, residual confounding from unmeasured, unknown, or mismeasured variables (i.e., smoking, chronic kidney disease staging) is possible. We attempted to reduce the impact of residual confounding by using an active comparator and a wide variety of potential confounders and confounder proxies across the different databases. Reassuringly, the results were highly consistent across the U.S. and the U.K.

Finally, our study was limited by its short median follow-up (range 1.5–2.6 years). However, the rationale for conducting this study was based on bladder cancer imbalances from RCTs of very short duration (range 43–727 days) (10). Thus, our study was specifically designed to address whether SGLT2 inhibitors are associated with a short-term risk of bladder cancer. Future studies will be needed to assess whether these drugs are associated with an increased risk of this outcome in the long-term.

In this international, multisite cohort study, the use of SGLT2 inhibitors was not associated with an increased risk of bladder cancer, compared with the use of GLP-1RAs or DPP-4 inhibitors. This study adds to the growing safety profile of SGLT2 inhibitors and provides reassurance on their short-term bladder cancer safety.

Funding. This study was funded by a Canadian Institutes of Health Research Project Grant (no. 103520). D.A. is the recipient of a Canadian Institutes of Health Research fellowship award. O.Y. holds a salary support from the Fonds de recherche du Québec – Santé (FRQS) Junior 1. B.H. receives funding from Cancer Research UK (PPRCPJT\100017 & 22185). R.W.P. holds the Albert Boehringer Ist Chair in Pharmacoepidemiology at McGill University. E.P. was supported by a National Institutes of Health National Institute on Aging Career Development Grant (K08AG055670) and by a Patient Centered Outcomes Research Institute (PCORI) research grant (DB-2020C2-20326). L.A. holds a Fonds de Recherche du Québec – Santé Distinguished Research Scholar Award and is the recipient of a McGill University William Dawson Scholar Award.

The sponsor had no influence on the design and conduct of the study, collection, management, analysis, or interpretation of the data, and preparation, review, or approval of the manuscript.

Duality of Interest. R.W.P. has received consulting fees from Biogen, Merck, Nant Pharma, and Pfizer for work unrelated to this study. S.S. is an investigator of investigator-initiated grants to Brigham and Women’s Hospital from Boehringer Ingelheim and is a consultant to Aetion Inc., a software manufacturer of which he owns equity, unrelated to the topic of this study. E.P. is co-investigator of a research grant to the Brigham and Women’s Hospital from Boehringer Ingelheim, outside the submitted work. L.A. has received consulting fees from Janssen and Pfizer for work unrelated to this study. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. D.A. wrote the manuscript. D.A., H.T., H.Y., S.V., E.P., and L.A. did the statistical analyses. O.H.Y.Y. and L.C. provided clinical expertise. B.H., R.W.P., and S.S. provided statistical expertise. E.P. and L.A. acquired the data. E.P. and L.A. supervised the study. All authors critically revised the manuscript. All authors analyzed and interpreted the data. All authors approved the final version of the manuscript and agree to be accountable for the accuracy of the work. All authors conceived and designed the study. The guarantors (E.P. and L.A.) affirm that this manuscript is an honest, accurate, and transparent account of the study being reported, that no important aspects of the study have been omitted, and that any discrepancies from the study as planned (and, if relevant, registered) have been explained. E.P. and L.A. and are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Parts of this study were presented as a poster at the 82nd Scientific Sessions of the American Diabetes Association, virtual and at New Orleans, LA, 3–7 June 2022, and as an oral presentation at the 38th International Conference on Pharmacoepidemiology & Therapeutic Risk Management, Copenhagen, Denmark, 24–28 August 2022.