|
Abstract:
|
Stephen Senn raised the question in 2004 and as of 2013 still saw no better way to prevent selection bias and provide a theoretical basis for calculation of probability statements. Flexible minimization with an improved prospective detector of selection bias may change that. If a patient is being submitted for assignment to treatment whose characteristics or timing are suspicious of selection bias, the computer will have three options, ask for a delay until another patient is available, flip a coin, or deliberately place the patient in a different group. The only action the submitting investigator can take to help remove that suspicion is to have his patient wait. Proper analysis of the outcome with opposite limiting signs on the primary and secondary analyses will tend to steepen the power curve improving the repeatability of the study. It should reduce the fear of sharing individual data sets and thus advance the goal of personalized medicine. Changes in the formula for the null hypothesis and labeling for the alternative should help correct our misuse of p-values and high rates of non-reproducible studies.
|
