Meta-analysis muddle: reviews of evidence are too often flawed

Pooling many studies to get at the truth of which food is healthy or which pills work is being let down by bias and bad data, says John Ioannidis