GLP-1 RAs and Kidney Cancer Risk: Does Study Design Drive the Signal?

TL;DR: When GLP-1 receptor agonists are compared against SGLT2 inhibitors,  a clinically appropriate active comparator,  using a target trial emulation framework, the previously reported association between GLP-1 RA use and increased kidney cancer risk disappears entirely (HR 1.05, 95% CI 0.62–1.77).

Abstract

Study at a Glance

Study Snapshot

Study Facts Table

What Researchers Actually Did

Lopez and colleagues designed this study as an explicit methodological challenge to three prior observational analyses that each reported an elevated kidney cancer risk among GLP-1 RA users. Using the TriNetX global EHR database spanning January 2006 through October 2024, the investigators first specified the protocol of a hypothetical randomized trial, then mapped each component to available observational data in accordance with the target trial emulation framework. Eligible participants were adults with type 2 diabetes and overweight or obesity. Treatment was defined as initiation of a GLP-1 RA versus initiation of either an SGLT2 inhibitor or metformin, with time zero anchored to the moment eligibility criteria, treatment assignment, and follow-up start all coincided. The analysis was framed as the observational analog of an intention-to-treat estimate. Propensity score matching (1:1 nearest-neighbour, 0.1 calliper) was used to adjust for baseline confounders including age, sex, race, ethnicity, socioeconomic determinants, BMI, HbA1c, CKD stage, eGFR, proteinuria, nicotine dependence, antihypertensive use, NSAID use, and others. Covariate balance was confirmed using standardized mean differences below 0.10.

The two treatment comparisons were analyzed separately because they address different methodological questions. The GLP-1 RA versus SGLT2i arm (63,789 eligible, 26,670 matched) was designed to replicate and interrogate the Xie et al. finding using a rigorous trial emulation structure. The GLP-1 RA versus metformin arm (266,172 eligible, 102,490 matched) was designed to replicate and interrogate the Wang et al. finding, with additional stratification by diabetes duration (fewer than 4 years versus 4 or more years) to probe whether channelling bias driven by treatment sequencing explains the observed signal. Reporting adhered to both STROBE and TARGET guidelines.

Key Findings: Primary Outcomes

• GLP-1 RA vs. SGLT2i,  no association: In the fully matched cohort of 26,670 adults, 30 kidney cancer events occurred in the GLP-1 RA group versus 27 in the SGLT2i group. The overall HR was 1.05 (95% CI 0.62–1.77), the RR was 1.11 (95% CI 0.66–1.87), and the absolute risk difference was 0.0% (95% CI −0.1 to 0.1). No association was detected at 1-, 3-, or 5-year follow-up intervals.
• GLP-1 RA vs. Metformin, elevated HR overall, attenuated in early T2D subgroup: In the matched cohort of 102,490 adults, the overall HR was 1.90 (95% CI 1.39–2.60). The RR was 1.19 (95% CI 0.88–1.59), substantially lower than the HR, indicating that the hazard signal is temporally front-loaded and that cumulative incidence differences are modest. Absolute risk difference remained 0.0% (95% CI 0.0–0.1), reflecting the low baseline incidence of kidney cancer.
• Attenuation in T2D duration <4 years subgroup: Among the 52,104 matched individuals with fewer than four years of type 2 diabetes at treatment initiation, the HR fell to 1.26 (95% CI 0.77–2.05), no longer reaching statistical significance. The RR in this subgroup was 0.74 (95% CI 0.47–1.18). This subgroup represents the population in which GLP-1 RA and metformin initiators are most clinically comparable.
• 1–3 year follow-up window, GLP-1 RA vs. Metformin: HR was 2.86 (95% CI 1.65–4.94), with RR 2.16 (95% CI 1.25–3.72), suggesting the excess hazard is concentrated early in the follow-up, a pattern more consistent with detection bias or pre-existing subclinical disease than with a causal carcinogenic effect.

Key Findings: Secondary Outcomes and Subgroup Analyses

• GLP-1 RA vs. SGLT2i at 1 year: HR 0.54 (95% CI 0.27–1.11), RR 0.57 (95% CI 0.28–1.16), a numerically lower point estimate for the GLP-1 RA group, though confidence intervals cross 1.0.
• GLP-1 RA vs. Metformin, T2D duration 4 or more years: HR 1.68 (95% CI 0.66–4.23), based on 11 versus 10 events (n=7,535 per group). Wide confidence intervals preclude meaningful interpretation; this reflects low statistical power from sparse events.
• Stratified analyses, GLP-1 RA vs. SGLT2i by T2D duration: Models did not converge due to insufficient event counts in both the <4-year and 4-or-more-year subgroups. Results were not reported for these strata.
• Divergence between HR and RR in the metformin comparison: The overall HR of 1.90 versus an RR of 1.19 indicates temporal non-proportionality. Hazard excess is concentrated early in follow-up, and cumulative cancer burden over the study period is modest, a pattern the authors interpret as consistent with detection or surveillance differences rather than sustained carcinogenic exposure.

Results: Adverse Events and Safety Profile

This study was designed exclusively to evaluate kidney cancer incidence as an oncologic safety signal, not to characterize the broader adverse event profile of GLP-1 receptor agonists. No additional safety or tolerability outcomes were reported. The outcome of interest, incident kidney cancer (ICD-10 C64), was identified using TriNetX data harmonized with tumour registry data to enhance diagnostic validity. Residual misclassification was expected to be non-differential and to bias results toward the null.

Statistical Approach and Rigor

The analytic strategy was appropriate for its stated purpose. Propensity score matching on a broad set of clinically relevant confounders, including markers of renal risk (CKD stages 1–3, proteinuria, eGFR) and metabolic severity (HbA1c, BMI), was a methodological strength. Standardized mean differences below 0.10 confirmed acceptable post-match balance. Reporting HR alongside RR for each comparison was a deliberate and informative choice: divergence between these metrics is diagnostic of front-loaded risk, which the authors appropriately flag as a marker of possible detection bias. Stratification by diabetes duration was prespecified and directly addressed the channelling bias hypothesis. Conditional Cox proportional hazards models accounted for within-match clustering. Limitations of the analytic approach include the ITT framing, which estimates initiation effects and may attenuate associations dependent on sustained exposure; the lack of cumulative dose or persistence data; and the inherently limited power to detect associations for a rare outcome in subgroup strata, as evidenced by model non-convergence in two SGLT2i subgroups and wide confidence intervals throughout.

Clinical Takeaway

Three prior studies raised a signal that GLP-1 receptor agonists may increase kidney cancer risk, a biologically counterintuitive finding given that obesity is a primary driver of renal cell carcinoma and GLP-1 RAs are first-line therapies for obesity management. This target trial emulation, the most methodologically rigorous of the four studies now in this literature, shows that the signal dissolves when the comparator group is clinically appropriate. Comparing GLP-1 RAs to SGLT2 inhibitors,  drugs prescribed for overlapping indications at similar points in the treatment pathway, produced no kidney cancer association across any follow-up interval. The elevated HRs seen against metformin are substantially explained by channelling bias from treatment sequencing: GLP-1 RA initiators in real-world data tend to have longer disease duration and greater comorbidity burden than metformin initiators, and when that gap

Physician-Led, Whole-Person Care

A doctor who takes the time to truly understand you.

Personal care that starts with listening and is guided by experience and ingenuity.

Health, Longevity, Wellness

One-on-One Cannabis Guidance

Metabolic Balance

Leave a Message
Metabolic Care
Medical Consulting
Cannabis Care