Decision analysis is useful, because it clearly defines the responsibilities that various parties have in making a decision. In medicine, this leads to an optimal decision and protects the patient and the healthcare provider, as will be described below.

Recall that expected utility is where is the probability of outcome given action a and initial state s and is the patient’s reward, or utility.

First, let’s define the roles in decision analysis.

The provider’s role is to give the probabilities of outcomes – e.g., the probability of not having a side effect and going into complete remission. The provider is responsible for this information based on clinical experience or the literature.

The patient’s role is to give the utility of the outcomes. Only the patient knows how they will feel if they have, e.g., partial remission and a side effect vs. complete remission and a side effect.

The final role is that of “optimization.” Given the probabilities from the provider and the utility from the patient, one uses optimization to find the decision that maximizes the utility in expectation (or for some quantile).

A lot of times, the decision making gets derailed when one or more parties, often with good intentions, overstep into the role of another party.

If I am a patient who oversteps into the provider’s role, I may override the provider’s opinion of the probability of some outcome, based on what I have learned from another resource. If this probability is incorrect, the decision will not maximize my expected utility.

If I am a provider who oversteps into the patient’s role, it may be because I project my own utility onto the patient. I may be younger and would seek a riskier treatment to live longer, whereas the patient might be older and prefer a treatment that gives better quality of life.

Finally the provider or the patient might overstep into the role of optimization. This occurs, oddly, if one or the other makes the decision. The optimization required to actually choose a decision is not a feasible problem for a human to mentally solve.

BTW: note that nothing stops a patient from overriding the decision made by optimization (there are valid reasons for why this would be a good idea, but if the decision analysis is done correctly, they should not have to).

If the patient will not choose a treatment, which will often occur, because the optimization is intractable, they might ask the provider for an opinion.

The provider is then tasked to, in addition to giving correct probabilities of outcomes, try to do this intractable optimization, for another person, mentally. Without using a formal optimization procedure, this will almost surely lead to a suboptimal decision for the patient.

Mentally doing the optimization also puts the provider at risk. They might be held accountable for randomness or accidentally incorporate, despite their best efforts, their own utilities.

The decision analysis, although it is cumbersome, protects both the patient and the provider.

The decision analysis protects the patient’s right to incorporate their own preferences, via a clearly defined utility function. This utility is subject to outside contamination not just by the provider but also by the payer, the patient’s family, and even society at large.

The provider is protected from having to make the decision for the patient. With a decision analysis, if the provider gives probabilities of outcomes that are valid based on the current medical literature or their clinical experience, they have done their job.

So decision analysis leads to optimal decisions and it protects everyone. Why don’t we use decision analysis, then?