Glucagon-Like Peptide 1 Receptor Agonists and 13 Obesity-Associated Cancers in Patients With Type 2 Diabetes | Oncology | JAMA Network Open

Key Points

Thirteen human malignant neoplasms have been identified as obesity-associated cancers (OAC), ie, the presence of excess body fat is associated with increased risk of developing cancer and worse prognosis in patients with these specific tumors.1 Obesity also contributes to insulin resistance and type 2 diabetes (T2D), which may further increase the risk and worsen the prognosis of the OACs.^2,3

The glucagon-like peptide 1 receptor agonist (GLP-1RA) class of pharmaceuticals are highly effective agents for the treatment of T2D and for achieving weight loss.4^–9 GLP-1RAs have further been shown to reduce the risk of adverse cardiovascular outcomes in patients with obesity¹⁰ and to contribute to the resolution of nonalcoholic steatohepatitis.¹¹ Because of their efficacy in controlling T2D, obesity, and related comorbidities, we hypothesized that these agents might reduce the risk of the OACs. We recently reported that GLP-1RAs were associated with lower risks for colorectal cancer,¹² an OAC. Otherwise, clinical evidence of the potential clinical benefits of GLP-1RA in preventing OAC has not been systematically assessed. Here we conducted a nationwide multicenter retrospective cohort study in patients with T2D who were prescribed GLP-1RAs vs insulins or metformin to determine whether GLP-1RAs were associated with changes in the risk of each of 13 OACs, including esophageal, breast, colorectal, endometrial, gallbladder, stomach, kidney, ovarian, pancreatic, and thyroid cancer as well as hepatocellular carcinoma, meningioma, and multiple myeloma.¹

We used the TriNetX platform to access deidentified electronic health records (EHRs) of 113 million patients from 64 health care organizations across 50 states, covering diverse age, racial and ethnic, income, and insurance groups and clinical settings.13^,14 The platform’s built-in analytic functions allow patient-level analyses, while only reporting population-level data. The platform has been used for retrospective cohort studies.^15–26 Similar to this study, we have examined the association of GLP-1RAs with colorectal cancer incidence in patients with T2D¹² and the associations of GLP-1RA (semaglutide) with suicidal ideations²⁷ and with cannabis use in patients with obesity and those with T2D.²⁸ The MetroHealth System institutional review board determined that the research as described in this study was not human participant research and institutional review board approval and informed consent were not required. This cohort study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.

Available data elements of EHRs include extensive information on demographics, diagnoses (International Statistical Classification of Diseases and Related Health Problems, Tenth Revision), medications (Anatomical Therapeutic Chemical and medical prescription normalized medical prescription or RxNorm), procedures (Current Procedural Terminology), laboratory tests (Logical Observation Identifiers Names and Codes), genomics, visits, and socioeconomic and lifestyle information. The data on the analytic platform have been expanded to include oncology-specific data from cancer registry data from North American Association of Central Cancer Registries (NAACCR) records and other data resources.14

The study population comprised 1 651 452 patients with a diagnosis of T2D who had medical encounters with health care organizations and were prescribed GLP-1RAs vs insulin or metformin between March 2005 and November 2018 and had no history of any of the 13 OACs. The study population was divided into exposure and comparison groups. For comparing GLP-1RAs with insulins, the study population was divided into a GLP-1RA/no insulin group (48 983 patients prescribed a GLP-1RA but not insulins) and a insulin/no GLP-1RA group (1 044 745 patients prescribed insulins but not GLP-1RAs). For comparing GLP-1RAs with metformin, the study population was divided into a GLP-1RA/no metformin group (32 365 patients prescribed a GLP-1RA but not metformin) and a metformin/no GLP-1RA group (856 160 patients prescribed metformin but not GLP-1RAs) (Figure 1).

The 13 OACs are esophageal, breast, colorectal, endometrial, gallbladder, stomach, kidney, ovarian, pancreatic, and thyroid cancer as well as hepatocellular carcinoma, meningioma, and multiple myeloma.1 Each of the 13 OACs was examined as a separate outcome in groups that were propensity-score matched for covariates related to the specific OAC. For each OAC outcome, the exposure and comparison groups (ie, GLP-1RA/no insulin vs insulin/no GLP-1RA groups and GLP-1RA/no metformin vs metformin/no GLP-1RA groups) were propensity-score matched (1:1 using nearest neighbor greedy matching) for baseline covariates related to the specific OAC, including demographic characteristics (age, sex, race, and ethnicity); adverse socioeconomic determinants of health; family and personal history of cancer; genetic susceptibility to cancer; preexisting medical conditions, including obesity and overweight; and medical procedures, including cancer screening, bariatric surgery, and prior prescription of antidiabetes medications. Each eligible individual was followed up from the index event (the first prescription of GLP-1RAs, insulins, or metformin during March 2005 to November 2018) until the occurrence of the outcomes, death, loss to follow-up, or 15 years after the index event, whichever occurred first. Cox proportional hazard analyses were used to compare rates of time to events on a daily basis during the follow-up time after the index event. Hazard ratios (HRs) and 95% CIs were calculated. Cumulative incidences were estimated using the Kaplan-Meier survival analysis. All models are adjusted for confounders at baseline by propensity-score matching baseline covariates.

The data were collected and analyzed on April 26, 2024, within the analytics platform. All statistical analyses in this study, including propensity-score matching, Kaplan-Meier survival analysis, and Cox proportional hazard analysis were done using built-in functions within the platform that are implemented using Survival version 3.2-3 in R version 4.0.2 (R Project for Statistical Computing) and libraries and utilities for data science and statistics in Python version 3.7 (Python Software Foundation) and Java version 11.0.16 (Oracle). Details of clinical codes for eligibility criteria, treatment strategies, outcomes, and baseline covariates are in eTable 1 in Supplement 1.

The study population included 1 651 452 patients with T2D (mean [SD] age, 59.8 [15.1] years; 827 873 [50.1%] male and 775 687 [47.0%] female participants; 5780 [0.4%] American Indian or Alaska Native, 65 893 [4.0%] Asian, 281 242 [17.0%] Black, 13 707 [0.8%] Native Hawaiian or Other Pacific Islander, and 1 000 780 [60.6%] White participants). For comparing GLP-1RAs with insulins in patients with T2D, the study population included 1 093 728 patients with T2D who had no prior diagnosis of any OAC and were prescribed GLP-1RAs or insulins but not both between March 2005 and November 2018. The GLP-1RA/no insulin group (n = 48 983) compared with the insulin/no GLP-1RA group (n = 1 044 475) was younger; included more women and White participants; had a higher prevalence of family history of cancer, obesity or overweight, medical encounters for cancer screening, and prior prescriptions of other antidiabetic agents, including insulins, metformin, dipeptidyl peptidase 4 (DPP-4) inhibitors, sodium-glucose cotransporter 2 (SGLT2) inhibitors, sulfonylureas, thiazolidinediones, and α-glucosidase inhibitors. For each OAC outcome, the GLP-1RA/no insulin and the insulin/no GLP-1RA groups were separately matched for covariates associated with the OAC. The Table shows the characteristics of the GLP-1RA/no insulin and insulin/no GLP-1RA groups before and after propensity-score matching for covariates related to colorectal cancer. The characteristics of the exposure and comparison groups before and after matching for each of the other 12 OACs are in eTables 2 to 13 in Supplement 1.

Compared with insulins, GLP-1RAs were associated with a significantly lower risk of 10 of the 13 OACs, including gallbladder cancer (HR, 0.35; 95% CI, 0.15-0.83), meningioma (HR, 0.37; 95% CI, 0.18-0.74), pancreatic cancer (HR, 0.41; 95% CI, 0.33-0.50), hepatocellular carcinoma (HR, 0.47; 95% CI, 0.36-0.61), ovarian cancer (HR, 0.52; 95% CI, 0.03-0.74), colorectal cancer (HR, 0.54; 95% CI, 0.46-0.64), multiple myeloma (HR, 0.59; 95% CI, 0.44-0.77), esophageal cancer (HR, 0.60; 95% CI, 0.42-0.86), endometrial cancer (HR, 0.74; 95% CI, 0.60-0.91), and kidney cancer (HR, 0.76; 95% CI, 0.64-0.91). The HR for stomach cancer among patients taking GLP-1RAs vs those taking insulin was less than 1, but it was not statistically significant (HR, 0.73; 95% CI, 0.51-1.03). GLP-1RAs were not associated with risk of postmenopausal breast cancer or thyroid cancer (Figure 2). Figure 3 shows the cumulative incidences of colorectal cancer and liver cancer comparing GLP-1RAs with insulins. The mean (SD) follow-up time for the outcome of colorectal cancer was 2074.7 (435.3) days for the GLP-1RA/no insulin group and 1981.8 (471.1) days for the insulin/no GLP-1RA group. The mean (SD) follow-up time for the outcome of liver cancer was 2023.1 (1112.6) days for the GLP-1RA/no insulin group and 2037.9 (766.4) days for the insulin/no GLP-1RA group.

For comparing GLP-1RAs with metformin in patients with T2D, the study population included 888 525 patients with T2D who had no prior diagnosis of any OAC and were prescribed GLP-1RAs or metformin but not both between March 2005 and November 2018. For each OAC outcome, the GLP-1RA/no metformin group (n = 32 365) and the metformin/no GLP-1RA group (n = 856 160) were separately matched for covariates related to the OAC (eTables 14-26 in Supplement 1). Compared with metformin, GLP-1RAs were not associated with a lower risk of colorectal cancer, gallbladder cancer, and meningioma but were associated with an increased risk of kidney cancer (Figure 4). Figure 3 shows the cumulative incidences of colorectal cancer and liver cancer by comparing GLP-1RAs with metformin. The mean (SD) follow-up time for the outcome of colorectal cancer was 1967.2 (592.2) days for the GLP-1RA/no metformin group and 2101.6 (576.0) days for metformin/no GLP-1RA group. The mean (SD) follow-up time for the outcome of liver cancer was 1970.9 (426.0) days for the GLP-1RA/no metformin group and 2129.8 (514.7) days for metformin/no GLP-1RA group.

Using a data platform29 to analyze more than 15 years of longitudinal EHRs of a US population-based cohort of more than 100 million individuals, we found that in patients with T2D who had no history of any OAC, GLP-1RAs compared with insulins were associated with a significant risk reduction in 10 of 13 OACs, including esophageal, colorectal, kidney, pancreatic, gallbladder, ovarian, endometrial, and liver cancers as well as meningioma and multiple myeloma. Decreased risk reduction that did not reach statistical significance was also noted for stomach cancer. Of those cancers that showed decreased risk of GLP-1RAs compared with insulin, risk reduction was also noted for GLP-1RAs relative to metformin for colorectal cancer, gallbladder, and meningiomas, although these findings were not statistically significant.

Our observations on the reduction in the incidence of OACs in patients with T2D treated with GLP-1RAs compare favorably with the OAC-reducing effects of intensive lifestyle intervention (ILI) observed in the Look AHEAD trial (Action for Health in Diabetes)30 and with the results of metabolic-bariatric surgery as recently reported in the SPLENDID (Surgical Procedure and Long-term Effectiveness In Neoplastic Disease Incidence and Death) trial.³¹ The Look AHEAD study, a randomized clinical trial in which 4859 patients with T2D and overweight or obesity (age, 45-76 years; median follow-up, 11 years) were randomized to an ILI or diabetes support and education group, found a 16% reduction in risk for OAC (HR, 0.84; 95% CI, 0.68-1.04).³⁰ The SPLENDID trial, a matched cohort study, compared 5053 patients with obesity with 25 265 nonsurgical matched controls, with a median age of 46 years and median follow-up of 6.1 years, showed an OAC risk reduction of 32%, (HR, 0.68; 95% CI, 0.53-0.87).³¹

A recent 9-year follow-up population-based historical cohort study32 conducted in Israel reported a decrease (although not statistically significant) in incidence of pancreatic cancer (HR, 0.50; 95% CI, 0.15-1.71) in patients with T2D treated with GLP-1RAs compared with insulin.³² Our US population-based study, with 15 years of follow-up and a larger sample size, now extends these observations, suggesting that treatment of patients with T2D with GLP-1RAs vs insulin is associated with a significantly decreased incidence of pancreatic cancer (HR, 0.41; 95% CI, 0.33-0.50).

In contrast to the risk reduction shown for most of the OACs, thyroid cancer showed no statistically different risk in patients treated with GLP-1RAs compared with insulins. Studies in rodents indicate that GLP-1RAs promote thyroid C-cell hyperplasia and medullary thyroid carcinoma (MTC) by a GLP-1R mediated increase in calcitonin synthesis.33 High levels of fasting serum insulin and insulin resistance are associated with an increased risk of thyroid cancer.³⁴ Although clinical evidence for an association of thyroid cancer with the use of GLP-1RAs has been reported as inconclusive,³⁵ the findings from our study together with previous reports of insulins promoting cancer growth suggest that GLP-1RAs might be associated with increased risk of thyroid cancer. Our results are further supported by a recent report³⁶ by the French National Health Cancer Data System showing that the use of GLP-1RAs for 1 to 3 years was associated with increased risk of all thyroid cancers (adjusted HR, 1.78; 95% CI 1.04-3.05).³⁶ These studies support the package warnings included with GLP-1RAs that these agents are contraindicated in patients with multiple endocrine neoplasia syndrome type 2 and that patients should be counseled regarding the potential risk of MTC and symptoms of thyroid tumors.

Kidney cancers showed an increased risk with GLP-1RA treatment relative to that with metformin (HR, 1.54; 95% CI 1.27-1.87) but a decrease relative to insulin (HR, 0.76; 95% CI 0.64-0.91). GLP-1RAs have direct effects on kidney function mediated by GLP-1Rs in renal vasculature; however, these are not associated with increased mitogenesis,37 and to our knowledge, there have been no previous reports of kidney cancers with the use of GLP1-RAs. These divergent risks require further clinical and mechanistic studies for full evaluation. Nonetheless, they suggest the need for continued monitoring in patients being treated with GLP-1RAs.

Our study, with follow-up over 15 years, found no signs of increase or decrease in risk for breast cancer in postmenopausal women with T2D being treated with GLP-1RAs compared with those being treated with insulin or metformin. GLP-1RAs have been shown to reduce the growth of murine and human breast cancer cell lines in vitro and in vivo murine models.38 However, a meta-analysis of more than 50 randomized clinical trials, evaluating GLP-1RAs in women aged between 45 to 70 years and followed up from 24 weeks to 7.5 years, showed no differences in benign, premalignant, or malignant breast neoplasms in patients treated with GLP-1RAs compared with other antidiabetic agents or placebos.³⁹ A more recent population-based cohort study of 44 984 women 40 years and older treated with GLP-1RAs or other antidiabetic agents for a mean of 3.5 years showed no overall significant difference in the risk for breast cancer occurrence. However, an increased risk (HR, 2.66; 95% CI, 1.32-5.38) was noted for those treated between 2 to 3 years with a return to null after more than 3 years’ treatment.⁴⁰ Interestingly, the SPLENDID trial of bariatric surgery for weight reduction, which found an overall 32% risk reduction for OACs, showed no significant difference among women for incidence of overall or postmenopausal breast cancer.³¹ This lack of effect on breast cancer risk needs to be further investigated to determine the impact of longer duration of therapy as well as to more fully understand the relation between GLP-1RAs and estrogen metabolism. The lack of breast cancer risk reduction by GLP-1RAs and the similar lack of protection by bariatric surgery may also suggest the possibility that factors determining the incidence of breast cancer in patients with overweight or obesity may have been initiated long before intervention with GLP-1RAs and/or bariatric surgery and therefore require earlier intervention to affect risk reduction. The concept that early intervention might reduce breast cancer incidence is supported by the observation that both pregnancy and breastfeeding reduce the incidence of breast cancer.^41,42

Our study has several limitations. First, this is a retrospective observational study of patient EHRs, which has inherent limitations including overdiagnosis, underdiagnosis, and misdiagnosis; unmeasured or uncontrolled confounders; and biases. Although we controlled for an extensive list of variables, these limitations and biases could not be fully eliminated; therefore, no causal inferences can be drawn. Second, patients in our study represented those who had medical encounters with health care systems contributing to the data platform. Although both the exposure and comparison groups were drawn from the same EHR database and from the same time period, which should not significantly affect the HR calculations, results from the platform need to be validated in other EHR databases and analytics platforms. Third, the status of incident cancer was based on the presence of first-ever diagnosis codes of OACs documented in patient EHRs, which also included oncology-specific data from cancer registry data, such as NAACCR records. However, it is unknown how well cancer diagnoses are captured in patient EHRs. For this study, the main interest was the relative risk (or HR) of cancer diagnosis. Since all patients in the study population were drawn from the same health care organizations in the data platform, cancer underdiagnosis, misdiagnosis, or overdiagnosis should not have a substantial impact on the relative risk analysis. Fourth, the built-in functions did not allow us to control for variables (eg, weight loss) that occurred after the index event and to identify individual patient data, which precludes our ability to correlate risk reduction with a degree of weight loss, which was demonstrated to be particularly important in the SPLENDID bariatric study.25 In addition, we could not explicitly control for health care utilization and insurance type although the study population included patients who had medical encounters with health care organizations and were withdrawn from the same 64 health care organizations in the network. Finally, due to the lack of patients’ medication adherence information in EHRs, we used intention-to-treat (medication prescriptions) as a causal contrast of interest regardless of whether the individuals adhered to their medications and the duration of the medication use.

Open Access: This is an open access article distributed under the terms of the CC-BY License. © 2024 Wang L et al. JAMA Network Open.

Corresponding Authors: Nathan A. Berger, MD, Center for Science, Health, and Society, Case Comprehensive Cancer Center ([email protected]), and Rong Xu, PhD, Center for Artificial Intelligence in Drug Discovery, Case Western Reserve University School of Medicine, 10900 Euclid Ave, Cleveland, OH 44106 ([email protected]).

Author Contributions: Dr Xu and Ms Wang had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Concept and design: Xu, Berger.

Acquisition, analysis, or interpretation of data: All authors.

Drafting of the manuscript: Xu, Berger.

Critical review of the manuscript for important intellectual content: All authors.

Statistical analysis: Wang, Xu.

Obtained funding: Xu, Berger.

Administrative, technical, or material support: Xu, Kaelber, Berger.

Supervision: Xu, Kaelber, Berger.

Conflict of Interest Disclosures: Drs Kaelber and Berger reported receiving grants from the National Institutes of Health (NIH) during the conduct of the study. No other disclosures were reported.

Funding/Support: We acknowledge support from National Cancer Institute Case Comprehensive Cancer Center (grant Nos. CA221718 and CA043703), American Cancer Society (grant No. RSG-16-049-01–MPC), The Landon Foundation–American Association for Cancer Research (award No. 15-20-27-XU), NIH Director’s New Innovator Award Program (award No. DP2HD084068), National Institute on Aging (grant Nos. AG057557, AG061388, AG062272, and AG07664), and the National Institute on Alcohol Abuse and Alcoholism (grant No. AA029831).

Role of the Funder/Sponsor: The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Data Sharing Statement: See Supplement 2.