Slashdot Mirror
Those Sleeping Pills May Be Killing You
dstates writes "A recent article in in BMJ Open reports a strong association between the use of prescription sleeping pills and mortality. The study used electronic health records for 2.5 million people covered by the Geisinger Health System to find 12 thousand who had been prescribed sleeping pills and a matched set of controls. Death rates were much higher in the patients taking sleeping pills and the risk increases with age. Kudos to the authors for publishing this in an open access journal."
This is not really true. The purpose of a double-blind experiment is to set up a study with a controlled variable and observe the outcome. This is a meta-analysis, which looks at previously gathered data and tries to see if there are interesting patterns. The problem with such analysis is that although "blind" in the sense that it does not influence results, it is not "blind" in the choice of data. Whether intentionally or not, by cherry-picking data it is easy to create associations where none exist. This is further biased by the fact that only positive results are reported - no one writes of all the "no correlation" results they may have found through different choices of matched sets.
For example, I am sure that I could take a piece of data such as daily temperature and pick a subset of the stock market that happened to correlate with it - something that is likely entirely a figment of the data sets. This is the danger in such studies and it explains why they are NOT in any way the same as a double-blind trial.