Decision Analysis

Decision analysis is the art and science of informing difficult decisions. It has a long and varied history, with roots in economics, business, psychology, engineering, and other fields. It is inherently multidisciplinary, especially with regard to analyses that involve the health of individuals or populations. Decision analysis can be used to inform clinical, health funding, or policy decisions.

The basic steps in decision analysis are universal to most rational and systematic decision-making processes. Briefly, a problem is defined, including the decision situation and context. Objectives, based on what the different stakeholders (participants in the decision) value or deem important, are defined and quantitative measures or scales (i.e., ‘attributes’) are determined. Alternative choices are defined. The problem is then modeled, using ‘expected value’ methods (described later), and the alternatives are ranked in terms of how well they satisfy the objectives. Sensitivity analyses are performed to examine the impact of uncertainties, and the need for further analysis or refinement is determined. A framework that is often cited is Hammond, Keeney, and Raiffa's (1998) PrOACT, which stands for problems, objectives, alternatives, consequences, and trade-offs. An important aspect of decision analysis frameworks is that they are iterative in nature and function as decision support tools rather than the ‘final answer.’

In health care and public health, most decisions revolve around improving survival, health state, and/or quality of life. Thus, in decision analyses involving health outcomes, an important consequence measure is typically some measure of health status, such as mortality, morbidity, or a combined measure, such as quality-adjusted life years (QALYs). These measures are assumed to represent utilities or measures of preference; that is, an alternative that improves, say, survival over another alternative (all other things being equal) will be preferred. There is an extensive literature on this subject, including pros and cons of different measures. Good summaries are provided by Brent (2003) and Drummond and McGuire (2001). As health interventions or policies nearly always involve resource limitations, cost per unit utility gained or ‘cost-effectiveness’ is often used as a decision criterion. Less common are cost-benefit analyses, in which all consequences, including health status or survival, are measured in monetary terms. Health decision analyses are sometimes incorrectly viewed as synonymous with economic evaluations. Indeed, it is informative to address multiple objectives important to multiple stakeholders in what is termed amultiattribute decision analysis, but to date these applications have been limited in the health care field, although multi-attribute analyses have been applied in the public health field.

The analytical aspects of decision analysis center on estimation of the ‘expected value’ of different alternatives. The expected value of an alternative is a function of the probability-weighted consequence(s). This is typically estimated using a decision tree or influence diagram. As an example, a decision may be whether one should choose intervention or Treatment a, b, or c to reduce mortality from disease X.

Say that Treatment a represents ‘watchful waiting’ or even doing nothing. Each ‘branch’ of the tree represents the probability that the patient will live or die, given a particular treatment. In the case of the present example, say that a utility of 1 is assigned to life, and a utility of 0 is assigned to death. If Treatment b increases the probability of survival by a greater degree than a or c, then its probability-weighted consequence will be larger, and thus it will be the preferred alternative under the axioms of utility theory.

...