Protopathic bias in observational studies on statin effectiveness