Jarvinen and colleagues argue in the BMJ that even if a drug is proven to be effective in large, randomised, clinical trials, it still needs to be shown how it works in real world settings before one can calculate the cost-effectiveness of the intervention. See the full discussion here.