Studies using animal subjects to study human brain disorders are often biased, first claiming positive results and then failing in human trials, a new study suggests.
The findings by John Ioannidis and colleagues at Stanford University could help explain why many treatments that appear to work in animals do not succeed in humans.
Bias also wastes money and could harm patients in clinical trials, said the study in PLoS Biology.
Researchers examined 160 previously published meta-analyses of 1,411 animal studies on potential treatments for multiple sclerosis, stroke, Parkinson's disease, Alzheimer's disease and spinal cord injury, all done on more than 4,000 animals.
Just eight showed evidence of strong, statistically significant associations using evidence from more than 500 animals.
Only two studies seemed to lead to "convincing" data in randomized controlled trials in humans, it said.
The rest showed a range of problems, from poor study design, to small size, to an overarching tendency toward publishing only studies in which positive effects could be reported.
Statistically, just 919 of the studies could be expected to show positive results, but the meta-analysis found almost twice as many -- 1,719 -- that claimed to be positive.
"The literature of animal studies on neurological disorders is probably subject to considerable bias," concluded the paper.
"Biases in animal experiments may result in biologically inert or even harmful substances being taken forward to clinical trials, thus exposing patients to unnecessary risk and wasting scarce research funds."
Animal studies make up a "considerable portion" of biomedical literature, with some five million papers archived in the medical PubMed database, it said.
While animal research exists to test safety and efficacy before new treatments are attempted in humans, most interventions fail when they reach human clinical trials, said the researchers.
"Possible explanations for this failure include differences in the underlying biology and pathophysiology between humans and animals, but also the presence of biases in study design or reporting of the animal literature."
Researchers said the bias likely originates when scientists conducting the animal studies choose a way of analyzing the data that appears to give a better result.
Also, scientists tend to seek out high-profile journals to publish their work, and those journals tend to prefer studies with positive results.
Solutions may include stricter guidelines for study design and analysis, pre-registration of animal studies so that the results must be published whether positive or negative, and making raw data available for other scientists to verify, the study said.