The one study that looked found a signal and then withdrew it. A target trial emulation put erectile dysfunction at 35.2 per 1,000 person-years among GLP-1 initiators against 28.0 among DPP-4 initiators [1]. The hazard ratio was 1.26, 95% CI 1.08 to 1.46. The authors then ran their own bias check, and after it the association was no longer significant. Sexual function on this drug class remains barely studied, as the contraception evidence also shows.
What the study measured
All participants were men with type 2 diabetes in a US health system. The raw result held across subgroups, sensitivity analyses and an external validation cohort. On its face that is a robust finding.
That technique matters because most observational research skips it. A study without a negative control reports its 1.26 and stops, and a reader has no way to tell how much of it is real. Here the authors quantified their own residual confounding and published both numbers, which is the opposite of a mortality difference arriving with no such check.
Why a coded diagnosis is a poor instrument here
Erectile dysfunction in this study is a diagnostic code. Three things must happen for one to exist: a man notices the problem, mentions it, and a clinician writes it down. Men who have just started an injectable weight drug are in front of clinicians more often, with more appointments in which something might come up.
That is detection rather than disease, and it is exactly what a calibration step is designed to remove. The same instrument problem runs through side effects counted from records.
The population carries the risk factors already
Erectile dysfunction in men with type 2 diabetes tracks vascular disease, depression and medication. A Norwegian registry matched 190,818 people starting a weight-loss drug to population controls [2]. Between 71% and 78% carried at least one recorded comorbidity, against 49% to 51% of controls. Opioids had been dispensed to 24% to 30% of them against 13% to 15%.
Those ranges span five drugs rather than expressing uncertainty, and the study measured nothing about sexual function. What it establishes is that this population arrives carrying more of everything that causes erectile dysfunction than the people it gets compared with.
The psychiatric half is measured too. A Swedish cohort followed 95,490 people with depression or anxiety [3]. Semaglutide was associated with less worsening mental illness, adjusted hazard ratio 0.58, 95% CI 0.51 to 0.65. Exenatide and dulaglutide showed nothing. Depression and its treatment are among the commonest causes of erectile dysfunction, and no study has connected these two literatures.
What a buyer should take from it
That one study looked, found a signal, tested whether its own method could be trusted, and concluded it could not support the finding. That is a good paper and a weak answer.
No randomized trial has measured erectile function on these drugs. Silence is not safety, and it is not evidence of harm either. The wider pattern of outcomes nobody has measured is collected in 175 outcomes across two million records.