Ideal estimand for a trial

To maximize expected reward,

E_{\pi} R(S,A,S') = \int_s \int_{s'} \sum_a R(s,a,s') p(s'|a,s)\pi(a|s)p(s)ds'ds,

an important target estimand is p(s'|a,s).

E.g., if we are performing a trial for an anticoagulant, we might consider our outcome s' to contain the variables stroke and bleed, and then p(s'|a,s) is the joint probability of having a stroke and bleed post treatment, given patient covariates.

Instead, often, the treatment effect of an anticoagulant is estimated. This then requires that one designate a primary outcome such as stroke. Often, it seems, this can detract from estimation of the secondary outcomes, if the trial is not powered for the secondary outcome, for example.

One response to “Ideal estimand for a trial”

  1. Decision analysis and medical research – Statistics, medicine, humanity

    […] expected utility equation tells us.  To perform decision analysis, we need to know the probability of different outcomes […]

    Reply

Leave a comment