Why is statistical analysis so important for clinical research?

Most treatments are not sufficiently effective for you to tell whether or not they work based solely on clinical experience. You need statistical analysis!

Consider the question of whether or not to anticoagulate patients with atrial fibrillation (a condition where the heart beats irregularly) and normal heart valves. Such patients are predisposed to emboli (blood clots that travel to other parts of the body). Although anticoagulation with warfarin prevents strokes due to emboli, it can cause serious side effects (bleeding). So what do you do if you have a patient with atrial fibrillation and normal heart valves?

I remember distinctly how Dr. Kanu Chatterjee, one of the greatest cardiologists to have ever practiced medicine, answered this question in 1987. I was among the medical residents congregated around him at University of California, San Francisco Medical Center waiting for pearls of wisdom. He took a deep breath and said: “What you do is you anticoagulate all your patients with atrial fibrillation until one of them bleeds into his head. Then you don't anticoagulate any of your patients until one of them has a stroke. Then you go back to anticoagulating all of them.”

Dr. Chatterjee was admitting with an honesty and humility often missing in clinical medicine that it was not clear whether the benefits of anticoagulation outweighed the risks. He was also capturing the tendency of physicians to base their decisions, in the absence of definitive evidence, on their most recent experience.