How to read a peptide study

Start with the species. Perel and colleagues compared, in the BMJ (2007), treatment effects in animal experiments against results from clinical trials for six interventions. For clot-busting drugs, antenatal corticosteroids and bisphosphonates, the animal and human data lined up reasonably well. For two others they did not: corticosteroids for head injury worked in animal models but gave no benefit in clinical trials, and tirilazad improved outcomes in animals but was associated with worse outcomes in stroke patients. The authors put the mismatch down to either bias or animal models that do not resemble the human disease closely enough. The conclusion is not that animal data are worthless — it is that they do not reliably predict what happens in people. For several popular peptides, rodent data are effectively the entire evidence base. Józwiak and colleagues note in Pharmaceuticals (2025) that for BPC-157, "most, if not all, studies are limited to small animal models (i.e., rats and mice)".

Then look at size and length. A randomised, double-blind study sounds robust until you see how many people were in it. Aruan and colleagues published, in the Journal of Clinical and Aesthetic Dermatology (2023), a double-blind randomised study of creams containing acetyl hexapeptide-3 and palmitoyl pentapeptide-4 for crow's feet. It ran for eight weeks with 21 women — seven per group. The authors describe their own work as preliminary and call for a more adequate number of participants and longer follow-up. Check what was measured, too. In that study the outcomes were instrument readings and self-assessment: stand-ins for the thing you care about, not the thing itself.

Factor in who paid. Lundh and colleagues updated a Cochrane review of industry sponsorship and research outcomes in 2017, based on 75 included papers. Studies funded by the manufacturer more often reported favourable results — relative risk 1.27 (95 per cent confidence interval 1.17–1.37) — and more often drew favourable conclusions, relative risk 1.34 (1.19–1.51). The authors argue this reflects an industry bias that standard risk-of-bias assessment does not explain away. Which means a study can look methodologically spotless and still lean the sponsor's way.

Four more checks before you conclude anything. First: a lab model of skin, a cell culture or a skin equivalent shows that a molecule can affect cells under laboratory conditions — not that it gets through the outer layer of real skin in a finished product. Second: tell a preprint from a peer-reviewed paper. A preprint has not been reviewed by anyone. Third: tell a registered trial from a published result. Józwiak and colleagues describe how a phase 1 study of BPC-157 in 42 healthy volunteers (NCT02637284) began in 2015, but the researchers withdrew the submission of results in 2016. An entry in a trial registry is not a result. Fourth: check whether the outcome is a stand-in measure or something that actually matters to a person. A changed biomarker is not the same as changed health.