The problem here is pervasive throughout the world of medicine and science: publication bias. This is the reason that a review of 53 studies relating to cancer drug targets only led to 6 being actually replicable. The positive results, which say that something works or is connected, are almost always published, at the same time almost all negative results (saying it is not related) weren’t published.
A review of data in 2008 dealing with antidepressants, by Erick H. Turner et al, found that from a total of 37 studies with positive results, all but 1 was published, whereas of the 36 negative results: 22 were not published and 11 were published to convey a positive outcome. This makes the job of both doctors and truth-seekers/debunkers much more difficult, as the average person does not have access to the unpublished clinical data.
So even when regulators are deciding about approving food or medicine is making their decision: there is a huge skew towards whatever it is being approved. This has caused death, birth defects, and countless suffering. The word countrless is deliberate: as confusion when dealing with subjects as complex as human health (and sickness) makes clear results difficult, and little financial interest exists in finding negative qualities or verifying that drug really is working.
Insider Ben Goldacre puts these and many other facts into perspective (really well, I encourage watching it), and what he conveys isn’t a pretty picture:
A further, and tendentially unmentioned problem, is that we are not aware of all the potential inter-drug reactions and potential drug binding with an unintended target: we are taking largely unnecessary risks in the name of biased results. More transparency is needed, negative results should be published, and existing scientific research should continue to be increasingly made publicly accessible.