Animal testing is not particularly accurate at predicting how drugs will work in humans. Only about 5% of therapies that show promise in animal studies ultimately receive regulatory approval for human use, and roughly 89% of new drugs fail in human clinical trials despite passing preclinical animal testing. These numbers paint a clear picture: animal models capture some biological reality, but they miss far more than they catch.
What the Approval Numbers Show
A large-scale analysis published in PLOS Biology tracked what happens to therapies after they show positive results in animals. About 50% progressed to any kind of human study, 40% made it into randomized controlled trials, and just 5% earned FDA approval. That approval rate ranged from 0% to 20% depending on the medical field, but the overall pattern held across the board.
The failure rate in clinical trials tells a similar story. Around 12% of drugs that pass preclinical animal testing enter human trials. Of those, only 60% survive Phase I (the earliest safety testing in small groups of people). Roughly half of all drug failures in human trials are caused by toxicity that animal studies failed to predict.
Why Animal Bodies Process Drugs Differently
The core problem is biological. Laboratory animals generally eliminate drugs from their bodies much faster than humans do. When you swallow a medication, your liver processes it before it reaches your bloodstream. This “first-pass” effect is often more pronounced in animals, meaning a drug that seems safe in a rat or dog may build up to harmful levels in a human body that clears it more slowly.
Protein binding adds another layer of unpredictability. Drugs attach to proteins in the blood, and only the unattached portion actually reaches tissues and has an effect. This ratio varies enormously between species. For one well-studied drug (valproic acid), mice and hamsters had five times more free drug circulating in their blood than monkeys and humans did. That kind of difference can make a drug appear far more or less potent depending on which animal you test it in.
The enzymes that break down drugs also differ across species. Humans rely on a specific set of liver enzymes to metabolize most medications. Animals use overlapping but distinct versions of these enzymes, sometimes activating drugs into harmful compounds that wouldn’t form in a human liver, and sometimes neutralizing toxins that would persist in human tissue. These aren’t small technical details. They’re fundamental differences in how a living body handles a foreign chemical.
How Well Animal Tests Predict Toxicity
A study published in the British Journal of Cancer looked specifically at whether toxic side effects seen in animals matched what happened in people during early cancer drug trials. The results were mediocre. When animal studies flagged a specific type of toxicity, they were correct about 65% of the time. When animal studies predicted a drug would be safe for a particular organ or system, that prediction was right only 50% of the time, essentially a coin flip.
Combining data from both rodent and non-rodent species improved things somewhat, with 71% agreement between animal and human toxicity profiles. Non-rodent animals alone (typically dogs or monkeys) matched human results 63% of the time, while rodents alone matched just 43% of the time. Testing in multiple species helps, but significant gaps remain.
Famous Failures and Successes
Thalidomide is the most notorious example of animal testing failing to protect humans. The drug, prescribed to pregnant women in the late 1950s for morning sickness, caused severe birth defects in thousands of children. Yet thalidomide does not cause birth defects in many of the animal species used for testing. The biological pathways that make it dangerous to developing human embryos simply don’t exist in the same form in those animals.
Aspirin presents the opposite problem. In animal models, aspirin produces a range of toxic effects that aren’t typically seen in humans, or that appear only under very different conditions. Had aspirin been subjected to modern animal-based regulatory testing when it was introduced in 1899, it might never have been approved. Its widespread acceptance was, in a sense, fortunate timing.
On the other hand, animal testing played an essential role in one of medicine’s greatest breakthroughs. The discovery of insulin depended heavily on experiments in dogs. In the late 1800s, researchers removed the pancreas from dogs and observed that they developed diabetes, establishing the organ’s role in blood sugar regulation. Decades of further work in dogs and cats eventually led Frederick Banting and Charles Best to produce a pancreatic extract that controlled blood sugar in diabetic animals. That success gave researchers the confidence to try it in humans in 1922, saving countless lives. Later refinements used extracts from cattle and pig pancreases, which proved even more effective.
The insulin story worked because the basic biology of blood sugar regulation is remarkably similar across mammals. When the underlying mechanism is conserved between species, animal models can be genuinely predictive. The trouble is that researchers often can’t know in advance whether a given biological pathway will translate well or not.
Efforts to Improve Predictive Value
One approach to closing the accuracy gap involves genetically engineering mice to carry human versions of key drug-processing genes. Researchers have created mouse lines where 33 mouse liver enzymes were deleted and replaced with the major human enzymes responsible for breaking down medications. These “humanized” mice can predict human drug interactions and blood drug levels with much greater accuracy than standard lab mice.
However, even these models have limitations. The remaining mouse enzymes that weren’t replaced can still interfere with results, catalyzing the same reactions researchers are trying to study. Each new generation of humanized mice gets closer to mimicking human drug processing, but the biology of a mouse is never fully human, no matter how many genes you swap in.
What the 5% Approval Rate Really Means
It’s worth putting that 5% figure in context. Not every failure represents a flaw in animal testing specifically. Some drugs fail in human trials because the disease turns out to be more complex than expected, because the dose that works in animals doesn’t scale properly, or because human clinical trials are (rightly) held to strict statistical standards. Poor design in the animal studies themselves, including small sample sizes, lack of blinding, and publication bias toward positive results, also inflates the apparent promise of therapies before they ever reach human volunteers.
Still, the low translation rate points to a real and persistent problem. Animal models are better at confirming basic biological mechanisms (like the role of the pancreas in diabetes) than at predicting the specific safety and efficacy profile of a new drug in human patients. They provide a rough filter, catching the most obviously dangerous compounds before human exposure, but they miss a great deal. For every drug that animal testing correctly identifies as promising, many more either appear safe in animals but harm humans, or appear dangerous in animals but would have worked perfectly well in people and are quietly abandoned.