Abstract:
|
Rubin (1978 Annals of Statistics) was truly innovative. The proof that strongly ignorable random assignment of treatments to patients allows unbiased estimation of efficacy from observable outcomes instead of potential outcomes has been the justification for randomized controlled trials (RCTs). This presentation gives two newer insights. The first is that, originally written for efficacy defined as a difference, for ratio efficacy this proof can and needs to be amended to take the prognostic effect into account (in medical terms). The second insight is that, in defining Estimand measuring efficacy of a new treatment against a control as the expected difference of outcome measured on the same patient, it may also be possible to take measurement error into account, if a randomized controlled trial (RCT) has repeated measures. This second insight opens up the possibility of Confirmability assessment of clinical trials, and patient level personalization of medicine.
|