GLP-1 receptor agonists and their possible link to cancer risk: A combined analysis of clinical trials

Type 2 Diabetes mellitus (T2DM) and obesity have both been associated with an increased incidence of cancer.1, 2, 3 The risk is further increased in individuals with both T2DM and obesity4 and in long‐term obesity.5 In particular, 13 malignancies have been identified as obesity‐associated cancers, for which the evidence is sufficient to identify obesity as a risk factor6; overall, weight excess may cause up to 20% of total cancer mortality.7

The possible pathogenetic mechanisms implied seem to be different for each neoplasm8: both altered adipocyte function and insulin resistance determine an increase in circulating levels of hormones, such as oestrogens, insulin and insulin‐like growth factors, which may in turn enhance tumour proliferation and inhibit apoptosis8; furthermore, adipokines may contribute to oncogenesis by stimulating chronic inflammatory mediators, increasing oxidative stress and impairing immune responses.9 Excess weight is also associated with conditions, such as gastroesophageal reflux and steatotic liver disease, which in turn increase the risk of specific cancers (oesophageal adenocarcinoma and hepatocellular carcinoma, respectively). Conversely, intentional weight loss has been associated with a reduction in oncogenic risk.10 Long‐term observations that patients undergoing bariatric surgery show a reduced incidence of cancer in comparison with age‐ and sex‐matched subjects11 suggest that treatment of obesity, when producing a relevant weight loss, could reduce the incidence of malignancies.

The relationship between diabetes and cancer is similarly complex. The association between diabetes and incident malignancies has been extensively reported for type 2 diabetes, whereas data on type 1 diabetes are less impressive.12 Some associations (e.g., type 2 diabetes and pancreatic cancer) could be partly due to reverse causation, with the still undiagnosed incident malignancy inducing hyperglycaemia.13 Furthermore, the effect of confounders (particularly, visceral adiposity) could play a major role in the epidemiologic association of type 2 diabetes with cancer. It is also possible that hyperinsulinaemia contributes to the risk of malignancies in people with diabetes.14, 15 There is no clear evidence of an association between higher A1c levels in people with diabetes and a greater risk of cancer.16 In clinical trials, intensified treatment of type 2 diabetes, leading to improved glycaemic control, did not reduce cancer incidence,16 but the sample size and duration of available trials could have been insufficient for this endpoint.

Glucagon‐like peptide‐1 receptor agonists (GLP‐1 RA) are very effective in improving glucose control in people with diabetes and achieving weight loss in subjects with overweight or obesity,17, 18 while also reducing cardiovascular adverse events in these high‐risk populations.19, 20 It is plausible that GLP‐1 RA also reduce the incidence of obesity‐associated cancers, as recently suggested by an observational study21; on the other hand, another investigation failed to detect any effect of these drugs on the incidence of cancer.22

Conversely, the possible detrimental effects of GLP‐1RA on the risk of some specific malignancies are still controversial. The Food and Drug Administration and the European Medicines Agency state that GLP‐1RA should be used with caution in patients with a personal or familial history of medullary thyroid cancer.23, 24 This statement is based only on preclinical (rodent) data, without further clinical evidence.25, 26 Furthermore, a possible increase in overall thyroid cancer in GLP‐1RA users has been reported in epidemiological studies27, 28 and clinical trials,29 whereas other observational studies failed to detect either beneficial or detrimental effects of GLP‐1RA on this endpoint.21, 30, 31

The results of observational studies, although interesting, can never be conclusive, because of the inevitable effect of confounders. Although some of the studies21, 27 provided adjusted analyses incorporating the effects of many concurrent parameters, the possibility of residual confounding can never be excluded. Therefore, results of observational studies should be verified through the analysis of data from randomized trials.

Data from individual trials do not show significant between‐group differences for any malignancy, but the number of events recorded in each study is too small to draw any conclusion. A previous meta‐analysis32 did not detect any significant association between GLP1‐RA and cancer incidence, but recent long‐term trials,33, 34 which were not available at the time, may have added new information on this topic.

The assessment of the effects of GLP‐1RA on cancer incidence is relevant for clinical practice. The demonstration of favourable effects, if any, would strengthen the motivation to choose these drugs, instead of possible therapeutic alternatives, in the treatment of obesity and type 2 diabetes. Conversely, an increased incidence of some types of cancer, if present, should be considered in the assessment of the risk–benefit ratio, particularly when treating relatively healthy overweight/obese patients.

The present meta‐analysis aims to summarize all the evidence from randomized trials on the effects of GLP‐1RA on the incidence of different malignancies.

This is a post hoc analysis of a review aimed at assessing the incidence of thyroid malignancies, previously registered on the PROSPERO website (registration number CRD42023456382).35 The present meta‐analysis was conducted following the Preferred Reporting Items for Systematic Reviews and Meta‐Analyses reporting guidance.36

Randomized trials have already shown that GLP‐1RA reduce cardiovascular morbidity and mortality in people with obesity and/or diabetes.33, 56, 57 However, they still fail to demonstrate a major beneficial effect on incident cancer, another long‐term outcome related to obesity and diabetes.

Indeed, among the obesity‐related cancers, only a reduction in the incidence of uterine cancer in trials performed in obese non‐diabetic subjects was detected, whereas the effect was not significant for studies in diabetes. It is possible that beneficial effects on cancer risk are more evident with higher doses of those molecules that produce a greater weight loss, which are indicated for the treatment of obesity, although other mechanisms cannot be ruled out. However, the difference between trials on obesity and diabetes was not statistically significant; the limited number of observed cases prevents a more accurate assessment of the effects of treatment in different patient populations. In fact, trials on obesity were performed on subjects with a higher baseline body mass index, using drugs (i.e., liraglutide or semaglutide) that produce a greater weight loss than other GLP‐1RA,58 and with higher doses than those used for diabetes. On the other hand, the result observed in trials on obesity should be interpreted cautiously because of the limited number of observed events, which were largely retrieved from one single trial.33 In addition, this result was observed in a post hoc analysis, which should be considered hypothesis‐generating, needing to be further confirmed.

Among different neoplasms, those with a greater association with obesity are uterine and oesophageal cancer.6 In the present analysis, a trend toward a reduction of oesophageal cancer was observed with GLP‐1RA, although differences between treatment arms did not reach statistical significance. This is not surprising, considering the relatively low incidence and therefore the small number of observed cases of oesophageal cancer. The limited sample size and the insufficient duration of exposure could well explain the lack of significant results for all the other obesity‐related cancers, despite the relevant weight loss induced by GLP‐1RA. This could also explain the discrepancy of the present results from those of observational studies.21, 28

The increase in the number of reported cases of colorectal cancer in patients allocated to GLP‐1RA is surprising since the weight loss induced by GLP‐1RA should reduce the risk of this malignancy.59 Interestingly, this apparent detrimental effect of treatment is evident in shorter‐term but not in longer‐term trials. The results of such post hoc subgroup analysis should be interpreted very cautiously, considering that the limited number of observed cases prevents the detection of significant differences between subgroups of trials. However, it can be speculated that side effects of GLP‐1RA treatment, such as anorexia, nausea, vomiting, constipation and diarrhoea, induce physicians to perform diagnostic procedures that reveal underlying (and pre‐existing) tumours; this phenomenon would produce an apparent increase in incident cases at the beginning of treatment, which fades in the longer term. This highlights a major limitation of the present study: in the included RCTs, malignancies were not enlisted among pre‐specified endpoints; therefore, they were investigated and subsequently reported as SAEs only when symptoms referred by patients suggested the possibility of a malignancy, with the possibility of underdiagnosis of incident cases in the control arm or of misclassifying as incident cases pre‐existing malignancies in the active treatment arm. On the other hand, a direct effect of GLP‐1RA on colon oncogenesis may have some biological plausibility: GLP‐1 receptor expression has been detected in rodents and human colorectal epithelial tumours, even though reduced in comparison with adjacent healthy colon mucosa samples60; in rodents, it has also been proposed that GLP‐1 receptor signalling may regulate the expansion of colon mucosa through fibroblast growth factor 7,60 although an effect of GLP‐1RA on tumorigenesis has been demonstrated only in mice distal small intestine and not in colon mucosa.60, 61 On the other hand, GLP‐1 seems to exert an anti‐inflammatory effect on the mucosa of mice through increased antioxidant enzymes and reduced expression of transforming growth factor‐1, phosphatidylinositol‐3‐kinase and inflammatory cytokines such as nuclear factor kappa B, interleukin‐6 and interferon‐γ.62 Therefore, data from preclinical studies may either support a beneficial or detrimental effect of GLP‐1RA on colon tumorigenesis.

Another clear safety signal is related to the incidence of thyroid cancer. The present analysis is in line, on a larger data set, with our previous report of an increased incidence of thyroid cancer, which is more evident in longer‐term trials.29 This is not surprising, since the previous result was obtained on a fraction of the present dataset. In contrast with previous assertions,29 the association of GLP‐1RA with thyroid cancer retains statistical significance even after the inclusion of SELECT33 and other recent trials. This finding is in contrast with expectations for anti‐obesity agents, since the incidence of thyroid malignancies is expected to be reduced by weight loss.6 A specific effect of GLP‐1RAs on the thyroid gland may also have a biological plausibility, because GLP‐1 receptor overexpression has been observed in differentiated thyroid tumours cells, in respect to normal thyrocytes,63 although no net effect of GLP‐1 receptors on these cells has been demonstrated.63, 64, 65 Further mechanistic studies are needed to reach definitive conclusions on the biological plausibility of the association of GLP‐1 receptor stimulation with thyroid cancer. Epidemiologic studies also provided conflicting results on this point.21, 27, 30, 31 The results of observational studies on the association of GLP‐1RA with thyroid cancer should be interpreted cautiously: on one hand, the warnings of regulatory authorities on medullary thyroid cancer could lead some clinicians to over‐investigate thyroid nodules, with an increased chance of diagnosing thyroid cancer; on the other hand, the same warnings could lead some clinicians to avoid these drugs in all patients with thyroid nodules. The present meta‐analysis, performed on randomized trials only, overcomes the latter limitation. In addition, the sensitivity analysis confirming the association in double‐blind trials suggests that this result is independent of any detection bias.

Regarding preclinical studies, a clearer insight into the possible direct effect of GLP1‐RA on thyroid and colon tumorigenesis may be provided by recent machine learning‐enabled computational methods,66, 67 which may enhance the prediction of possible expression patterns and signalling pathways of GLP‐1 receptor in thyroid and colon normal and cancer tissues.

Considering the relatively low incidence of thyroid cancer, even an increase of relative risk would determine a limited increase in absolute risk, with a number needed to harm over 1300.29 This possible detrimental effect of GLP‐1RA on the incidence of thyroid cancer, if confirmed, should therefore be considered in the wider context of proven benefits of therapy, such as prevention of cardiovascular disease,33 prevention33 or improvement68, 69 of type 2 diabetes and possibly other obesity‐related complications.70

We already highlighted some limitations of the present study, that is, the limited sample size and study duration. In fact, the small number of recorded events for many malignancies could have prevented the observation of either beneficial or detrimental effects. Furthermore, the small number of recorded events prevented subgroup analyses for individual molecules of the class for each specific malignancy, and separate analyses for different doses for each available drug; this could be relevant, as different drugs have different effects on body weight, and weight reduction is dose‐dependent.71 In addition, the length of exposure to GLP‐1RA could have been insufficient to detect some longer‐term effects, increasing the chance of falsely negative results. Notably, many observational studies could have the advantage of a greater sample size, but they also usually suffer from a relatively short length of drug exposure. The fact that incident malignancies were not pre‐specified endpoints, and that cases were not formally adjudicated, is another major limitation, leading to the possibility of undetected cases and/or misdiagnosis, as already stated before. Unfortunately, a randomized trial on diabetes or obesity with cancer as a primary endpoint would be scarcely feasible, and observational studies suffer from the very same issues of underdiagnosis and misdiagnosis. Some further limitations of the present meta‐analysis should be considered for a correct interpretation of its results. The analysis was performed considering the overall number of incident cases, without any information on the date of onset; therefore, it was not possible to exclude early‐onset (and probably pre‐existing) malignancies, reported soon after the beginning of treatment. The subgroup analysis of shorter‐and longer‐term trials only partly surrogates the actual analysis based on time to event in individual patients. A further limitation is the specificity of the case mix: the population enrolled in clinical trials does not fully represent those receiving treatment in routine clinical practice. In particular, study protocols of most trials exclude patients at higher risk of medullary thyroid cancer. In addition, although most large‐scale trials included were performed versus placebo, the meta‐analysis also collected trials in which GLP‐1RA were compared with other active drugs; such comparators could theoretically be associated with either an increase or a reduction in the risk of certain cancers, possibly affecting results. Furthermore, it was not possible to retrieve data on the incidence of cancer from one large trial with liraglutide,68 although the analysis of funnel plots did not suggest selective reporting. Interestingly, analysis from this trial showed that the 5‐year cancer mortality in those with liraglutide is not different from comparators.72

On the other hand, some strengths of the present analysis should be considered. In clinical trials, the allocation to different treatments is determined by randomization. This warrants the comparability between patients on GLP‐1RA and control arms, thus overcoming the major limitation in observational studies, which may be affected by uncontrolled (or inadequately controlled) confounders. Moreover, in this meta‐analysis, the heterogeneity in results between studies is very low, thus further reducing the possibility of confounding factors. Notably, statistical heterogeneity was low despite the fact that different molecules of the class have different efficacy as weight‐reducing agents.71, 73 Another strength is represented by the retrieval of serious adverse events from multiple sources, that is, public trial repositories, and not only publications; this limits the possibility of publication bias.

The results of the present meta‐analysis are partly discordant from those of observational studies. Longer‐term observations on cohorts of patients treated with GLP‐1RA reported significant reductions in the risk of some obesity‐related malignancies, such as prostate28, 74 and colorectal21, 28 cancer, which were not detected in clinical trials. The discrepancy could depend on differences in case mix and/or shorter duration of observation in clinical trials, but it could also be the effects of unaccounted confounding bias in observational studies. Notably, results of clinical trials are compatible with the reported observations of a reduction in the risk of uterine28 and oesophageal75 cancer. The increase in the risk of kidney cancer in comparison with metformin, reported by one observational study21 was not confirmed in the present meta‐analysis, but the number of observed events is too small to draw any conclusion. The results of observational studies on the association of GLP1‐RA with thyroid cancer are heterogeneous, reporting either an increased27, 28, 31 or unmodified21, 30 incidence. The difference in results of clinical trials could be due either to a different case mix or to possible prescription bias: since the summary of product characteristics in the US and Europe,23, 24 on the basis of preclinical studies only, warns of possible risks in patients with a personal or familial history of medullary carcinoma, physicians could be less prone to prescribe a GLP‐1RA in patients with thyroid nodules, which are often not reported in large databases used for retrospective observational studies.

In conclusion, in randomized clinical trials available to date, GLP‐1RA do not appear to produce major effects on most malignancies, either beneficial or detrimental; however, the results of this meta‐analysis should be considered hypothesis‐generating rather than conclusive. A reduction of the risk of those malignancies that are more closely related to obesity, such as uterine and oesophageal cancers, seems possible; however, further data are needed to draw clear conclusions on this point. In this respect, observational studies, which are inevitably hampered by confounding bias, can be confirmatory of evidence from randomized trials. Conversely, very long‐term trials with malignancies as the principal outcome are hardly feasible. Results of ongoing large‐scale trials on GLP‐1RA and other molecules that also act as GLP‐1R agonists (e.g., dual GLP‐1/GIP and GLP‐1/glucagon agonists) will add further data, which will need to be incorporated in future meta‐analyses on this issue. On the other hand, a moderate increase in the incidence of colorectal cancer, which is more evident in shorter‐term trials, could be the effect of over‐diagnosis induced by the side effects of GLP‐1RA; however, this issue deserves further investigation through specific mechanistic studies and long‐term observational studies. Finally, the possibility of a moderate increase in the risk of thyroid cancer, particularly in longer‐term trials, cannot be ruled out, and it needs to be clarified because of its relevance in the assessment of the risk/benefit ratio of GLP‐1RA treatment in individual patients.

G.A.S and E.M. made the analysis plan, researched data, performed analyses, contributed to the discussion and wrote the first draft of the manuscript. C.M., G.G.D.V., C.B. and M.M. contributed to data research and reviewed and edited the manuscript. All the authors had full access to study data, approved the final version of the manuscript and took responsibility for data integrity and analysis accuracy.

GAS has received speaking or consultancy fees from Astra Zeneca, Eli‐Lilly and Novo Nordisk, outside the submitted work. MM has received speaking fees from Astra Zeneca, Boehringer‐Ingelheim, Eli‐Lilly, Merck, Novo Nordisk and Sanofi, outside the submitted work. The unit directed by EM has received research grants from Abbott, Eli‐Lilly and Novo Nordisk, outside the submitted work. EM has received consultancy fees or speaking fees from Astra Zeneca, Bayer, Boehringer‐Ingelheim, Coresearch, Dexcom, Eli‐Lilly, Molteni, Novo Nordisk, Pidkare and Sanofi, outside the submitted work. C.M., G.G.D.V. and C.B. do not have any competing interests to disclose.

The peer review history for this article is available at https://www.webofscience.com/api/gateway/wos/peer‐review/10.1111/dom.16489↗.