Human Data vs Preclinical Data
The two evidence categories are not interchangeable. This page lays out, in a comparison table, what each can and cannot establish.
Educational research-literacy content only. Not medical advice, not dosing guidance, not sourcing advice, and not a protocol for human or animal use. See our responsible information policy.
| Question | Preclinical | Human |
|---|---|---|
| Does the compound bind a receptor? | Yes | Usually unnecessary |
| Is there a measurable downstream effect? | Yes (in model) | Yes (in body) |
| Is the effect dose-related? | Yes (in model) | Yes (in body) |
| Does this dose work in humans? | No | Yes (if studied) |
| What is the human safety profile? | No | Partial |
| What is the long-term safety? | No | Usually no |
| Does it improve a clinical outcome? | No | Yes (if measured) |
| Is human translation likely? | Suggestive only | — |
The hierarchy is not optional
Human evidence wins. Preclinical evidence is essential for mechanism and for justifying a first-in-human trial — but it does not, by itself, establish that a compound works in humans or is safe in them.
Why “works in mice” is not a substitute
A 2007 systematic review found interventions that helped in animal models but not in people: corticosteroids in head injury (benefit in animal models, none in clinical trials) and tirilazad in stroke (reduced infarct volume by 29% in animal models, worse outcome in patients) (Perel 2007, PMID 17175568). Other interventions in the same review showed effects in both: thrombolysis improved outcome in stroke patients and tissue plasminogen activator reduced infarct volume by 24% in animal models, and bisphosphonates raised bone mineral density in patients and alendronate did so in animal models. Some peptide medicines have crossed the gap: semaglutide and tirzepatide now have labels. That does not show the gap is small for other peptides.
If you spot an error or have new evidence to suggest, please use our corrections policy.