Not true! Economists have devoted person-millenia developing the theory of instrumental variables, which lets you extract causal relationships from historical non-experimental data.
That said, it's damn hard to find a good instrument. Many economists use it badly, and most doctors don't even know the theory.
The problem is we conduct science on systems we don't fully understand. Because we don't fully understand it, it's easy to miss important information we didn't know is important.
The double blind with random selection allows us to control for all important information, determine causation, and is the best known method for controlling for bias. We don't have to worry about missing important variables because the nature of random selection means those variables are randomly distributed.