Cooking the Books:The statistical games behind “off-label” prescription drug use
Study patients took the anti-seizure drug Neurontin, and researchers measured tons of possible outcomes (like pain with touch, pain with cold, excessive pain with pinpricks, more than a dozen different scales for psychiatric symptoms, and so on). By random chance, if you measure enough outcomes, at least some of them will appear better after drug treatment. When the time came to report the findings, however, the researchers systematically omitted the outcomes on which the drug had no effect—and presented only the data showing benefit. That’s like dealing dozens of hands of poker to yourself but showing only the hand with good cards.