Moving Away From Binary Decision Making: A Lesson From History

Doc, does this treatment work? It is a reasonable question, but statistics is a tricky business and one where clinicians should tread with care. So, how do you try to decide whether the results of a clinical trial show that a treatment works or not? Palesch has described some common P value pitfalls, so that decisions based on an arbitrary P value of <0.05, to test whether a trial result is significant or not, is a bit like fast food—tempting but potentially unwise. So, significance testing and looking for P<0.05 are not necessarily the best way to decide whether therapy works/ does not work, yet I often catch myself thinking in this dichotomous way (eg, good prognosis versus bad prognosis and treat versus do not treat). The 50-year history of the evaluation of corticosteroids for acute stroke and traumatic brain injury (TBI) provides valuable examples (with some fictional additions) to illustrate the value of confidence intervals (CIs).