I’ve expounded on the benefits of decision analysis for medical decision making, discussing how it helps define the roles in a medical decision (which can be easier on the involved parties) and has the potential to interact favorably with entities in the legal, pharmaceutical, and insurance areas. 

From a research perspective, though, what do we need to actually make decision analysis possible?

The expected utility equation tells us.  To perform decision analysis, we need to know the probability of different outcomes given different treatments and patient characteristics.  I.e., we need to design our trials to estimate the joint probability of, e.g., mortality and adverse effects as a function of the chosen treatment and other information about the patient, such as the history, labs, etc.  In mathematical notation, for patient covariates treatment and outcomes we need to estimate which gives us, e.g.,

P(mortality, adverse effects| treatment, history, labs, etc …) 

and

P(mortality, adverse effects| no treatment, history, labs, etc …).

Then, in collaboration with the patient, who helps define utilities, it’s a matter of computing the expected utility and choosing the treatment that maximizes it. 

In the end, when making medical decisions, healthcare providers and patients must take risks and benefits into account.  If trials estimate the probabilities mentioned above, it will be a significant step toward formalizing this.

Note that these probabilities are different from a trial’s usual estimand, which is the average treatment effect, (where is mortality), which gives us, e.g.,

E[mortality|treatment]-E[mortality|no treatment].  

These probabilities are less different from the conditional treatment effect, which gives us, e.g.,

E[mortality|treatment, history, labs, etc…]-E[mortality|no treatment, history, labs, etc…].

Possibly, a trial could estimate all of these things: , and