Mice

Mice in a scientific experiment. Image from Science Daily.

Science has an interesting read on the frequent disconnect between animal studies and human trials in drug development. Many treatments that look promising in mice models, fail spectacularly in humans. They cite a study on stroke treatments by Malcolm Macloed:

Macleod and his colleagues identified 603 drugs tested in animals, 374 of which had helped heal the brain. Of those, 97 had been tried in humans—and only one had worked. And that one, Macleod is quick to point out, wasn’t tested because of animal data at all, but because it had already benefitted patients with heart attacks.

Currently animal studies have several limitations. For example, in the vast majority of studies animals are not assigned to treatment or control populations randomly. This means that any observed effect might be attributable to something intrinsic to the group and not the drug in question. Additionally, the studies are frequently not blinded, meaning the researchers know throughout the experiment which groups are receiving treatment or a placebo. This tends to introduce bias into studies where researchers observe an effect even that does not exist or overstate its significance. The last problem the article addresses is that many studies use small sample sizes, making the study more likely to observe an effect even when there isn’t one:

What they found was telling. If the two groups [control and treatment] contained just four animals each, there was a 30% chance that an illusory life expectancy gap would show up. With 10 animals per group, the risk dropped to 10%. “You can imagine 10 labs doing this experiment,” says Shai Silberberg, a program director at the National Institute of Neurological Disorders and Stroke (NINDS) in Bethesda, Maryland. “One gets an effect, and they publish it.” The other nine are much less likely to submit a paper. Suddenly, the literature is skewed.

The article ends by saying most of these limitations aren’t attempts at obfuscation or intentional misinformation. It also lists some of the steps that researchers and journals are implementing to help improve the current state of affairs. Researchers would like for studies to be more reproducible and translated to humans with greater success. For example the NIH is asking for a 15 question risk of bias assessment be performed for some studies. And journals like Nature are asking for more information on population selection, randomization and blinding with animal study submissions.