We don't know that "it works in nearly all the cases".
We know that in nearly all the cases, any issues don't rise to the level where the FDA or the public identifies them. I doubt that the system is constructed in such a way that nearly all the problems are identified.
It seems to me that because drugs are developed to meet the absolute minimum standards for proving efficacy and safety, there's inherently going to be little if any margin for error in the manufacturing process and hardly any way to analyze problems. When reporting potential side effects, most drugs are going to be just above the noise level in their principle effect, so how are you going to determine what is a random unrelated issue and what is a negative side effect, and what is the consequence of a QA or design failure?