A mouse is not a small person
Preclinical testing relies heavily on animals and cultured cells. These systems share much of the basic biology of humans, which is why they are used at all, but they differ in ways that matter. Drug metabolism, immune responses, and the timing of biological processes vary between species. A compound can be safely cleared by a mouse liver and accumulate to toxic levels in a human one, or it can work through a mechanism that simply does not exist in the same form in the test animal. Cultured cells are even further removed: they are isolated from the whole-body context of circulation, immune surveillance, and organ interaction, so a result in a dish says little about what happens in a living patient.
A trial is not a controlled experiment
Even when a mechanism does carry over, human trials introduce variability that no laboratory study reproduces. Participants differ in age, genetics, other illnesses, and other medications. Real patients often have conditions that the clean laboratory model excludes. A treatment that performs well in a uniform, carefully controlled setting can behave differently across a diverse population, and effects that only appear in a small subgroup are easy to miss until enough people have been exposed.
The gap is not a sign that early testing is worthless. It means early results are evidence about a model system, not a verdict about humans, and the translation step is where many candidates are lost.