Ideally. In reality it rarely happens. And it's certainly not needed, the entire basis upon which evidence-based medicine is founded is that an effective intervention should be statistically measurable in it's effect regardless of our knowledge of the underlying mechanics.