In a previous post, I described how decision analysis might be viewed as a way to assign roles in a medical decision, where the patient’s role is to specify a utility function, the provider’s role is to give the probabilities of different outcomes, and mathematical optimization’s role is to find the best decision, taking into account the utilities and probabilities.

Interestingly, in the real world, we rarely see adherence to these roles.

If I am a patient who helps with the provider’s role, I may redefine probabilities. For example, I may due this based on what I have learned from another resource. This can sometimes be unfavorable. For example, as a patient, I might assume that surgery will not work, since one of my friends recently had a failed surgery. Although it can sometimes be unfavorable when the patient influences probabilities, other times, they might have access to information, such as their own likelihood of compliance, which may impact a treatment’s probability of working. In this case, the patient’s participation in assessing probabilities can be important.

If I am a provider who helps with the patient’s role, I might help with defining utilities. This can sometimes be unfavorable for the patient: for example, despite good intentions, I might project my own utility onto the patient. For example, I may be young, and, in order to maximize my own life expectancy, I might prefer a riskier treatment. However, the patient might be elderly, and they might prefer a more conservative treatment that prioritizes quality of life. Sometimes, however, it is important that the provider guides the patient when defining a utility function. Consider an ileostomy, which is a potentially life-saving opening that is created in abdomen. A provider might have seen many patients with ileostomies, and, therefore, might know better than the patient how an ileostomy would feel. In such a case, the patient would benefit from the provider’s input in specifying the utility function. As shown here, with the help of providers, researchers can sometimes pre-specify a set of utility functions, one of which can then be selected by the patient.

The last player in a decision analysis is optimization. If I am a provider or patient who steps into the role of optimization, it may be because I attempt to determine which decision is best. This might sound odd. However, in the theoretical world of decision analysis, it is optimization, not the provider or the patient, that combines probabilities and utilities to determine which decision is best. In other words, neither the provider nor the patient makes the actual decision. In the real world, however, the patient or provider almost always makes the decision, because the tools and ingredients for decision analysis (and perhaps the rationale—maybe I am missing something after all) are not yet readily available. Often, the patient asks for a recommendation, and, in doing so, the patient implicitly requests that the provider make a decision.

Hence, in the real world, the patient and provider often step into different roles from those assigned by decision analysis. As I will argue in a future post, though, it might be better for everyone, in certain cases, if we adhere more to the roles assigned by decision analysis. Performing a decision analysis might be laborious, especially because the framework is not yet in place, but it might be easier in other ways.