Design: systematic review and meta-analysis of randomised controlled trials, searched to March 2025
Population: 48,765 patients with type 2 diabetes across 48 randomised trials of SGLT2 inhibitors versus placebo or active comparator
Primary outcome: gastrointestinal neoplasms as reported in publications, supplementary material or trial registries
Effect: overall OR 1.10 (95% CI 0.84-1.44; p = 0.46; I-squared 0%); all site-specific and subgroup analyses non-significant

Forty-eight randomised trials totalling 48,765 patients with type 2 diabetes were pooled to test whether SGLT2 inhibitors raise the risk of gastrointestinal neoplasms. Overall there was no association, with an odds ratio of 1.10 and no heterogeneity between trials. Site-specific analyses — oesophageal, gastric, hepatic, pancreatic, colonic, colorectal and rectal — were all non-significant, as were subgroups by individual agent, age, BMI, HbA1c, treatment duration and dose.

The question arose from pharmacovigilance signals and from the general anxiety that attaches to any drug class taken indefinitely by tens of millions of people. This is the most complete randomised evidence available on it, and it is reassuring.

The authors are careful to say how far the reassurance goes, and their caution is well placed. About half the trials had follow-up of a year or less, which is short for an oncological outcome with a long latency. Events were few. And cancers were captured as reported adverse events rather than as centrally adjudicated cancer endpoints, which means both detection and classification were incidental to the trials' purposes.

For practice the reading is straightforward. A gastroenterologist asked whether a patient's SGLT2 inhibitor should be stopped over cancer concern can say the randomised evidence shows no signal, while being clear that long-term data do not yet exist.

Do not stop an SGLT2 inhibitor over gastrointestinal cancer concern; the randomised evidence shows no signal
Keep standard screening as it is; nothing here argues for extra surveillance in patients on these drugs
Explain the limitation honestly where a patient asks — short follow-up, not a long-term safety guarantee
Continue to report suspected drug-associated neoplasms through pharmacovigilance; that is where a long-latency signal would first appear

Don't overread it

Cancers were captured as reported adverse events rather than adjudicated endpoints, and about half the trials ran for a year or less — too short for an oncological outcome.

The statistics, in plain English

The pooled odds ratio of 1.10 with a 95% confidence interval of 0.84 to 1.44 includes 1.0, and an I-squared of 0% means the trials agreed closely with each other. The site-specific intervals are much wider — oesophageal ran from 0.37 to 3.45 — because each site had few events; those analyses cannot exclude a meaningful risk at any individual site. The deeper limitation is ascertainment: when cancers are picked up as adverse-event reports rather than sought systematically, both arms may under-record them, which biases the comparison towards showing no difference.