The sample size has to be large enough for the results to have statistical significance. If something happens in the patient that never happens in comparable cases, and you have piles of information about suitable controls, then you can place a high certainty in the treatment being able to work. You don't have a success rate, but nobody's claiming one.