EPI-820 Evidence-based Medicine Clinical Decision Analysis – Background David R. Rovner MD OVERVIEW: Questions having to do with uncertainty and probability are deeply ingrained in the process of patient care and the fabric of clinical medicine. They arise from several sources: the ambiguity of clinical data and variations in interpretations; uncertainty about relations between clinical information and the presence of disease; uncertainty about the effects of treatment; and errors in clinical data. Decision analysis is a systematic way of making decisions under conditions of uncertainty. It organizes alternatives and choices to incorporate both the probability of an outcome and the value or utility of that outcome. Decision analysis is often the organizing foundation to performing cost-effectiveness and cost-utility analysis. It often results in a “decision tree”. RATIONALE FOR USE: Several heuristics and biases limit human’s ability to interpret uncertain data. Among them are representativeness, anchoring and adjusting, regression toward the mean, and conditional probability. Expected utility theory attempts to help individuals to overcome these biases. GOAL: To understand the analytic framework of decision analysis and it’s use in clinical decision making. OBJECTIVES: e familiar with terminology used in decision analysis. Understand the four steps of Clinical Decision Analysis (CDA). Understand how to define a question answerable by CDA. Be able to identify the information needed for CDA. Know the sources of data. Understand the role of incidence, prevalence, ethnicity, prior patient history, and personal sense of value or worth (utility) in constructing a decision tree. Be able to appropriately place the probabilities and utilities obtained on a decision tree. Be able to calculate expected utility and understand its usefulness. Understand the concept and application of sensitivity analysis. e able to apply decision analysis techniques to a clinical problem. TOOLS Terminology Decision node: A branch point on a decision tree where a decision must be made. Chance node: A branch point on a decision tree where the results are governed by probabilities or likelihood rather than choices. The sum of all probabilities emanating from the chance node must sum to 1.0. 1 Probability: May be thought of as either the frequency with which an event occurs in a population or a measure of the strength of belief that an event will occur or that a state of the world is true. Odds: A ratio of the probability of an event happening divided by the probability that it will not. Utility: The value or expected worth of a particular outcome. Expected outcome: The expected result of following a specific pathway in the tree. Expected utility: The multiplied product of probability and utility. The expected utility may apply to any chance or decision node. Markov Analysis: A method of evaluating decision trees that uses time as a variable. It can calculate expected utility at intervals chosen by the analysts. Monte Carlo: A method of analysis that evaluates the decision tree by allowing a large number of trips down the tree all determined by chance alone. The article by Kattan et al. shows an up-to-date decision analysis about a subject that is not settled by a large randomized control trial. The steps of the analysis are very similar to those of the example appended below, except that the Markov method of analysis is used. For this analysis computer techniques are almost mandatory. Steps of CDA 1. Identify and bound the question: What is the decision at hand. What should be considered including: a. Which aspects of the patient's health are of particular concern. b. Alternative actions - frequently not given enough thought. c. What clinical information will be available to make the decision. d. Possibly in this age, cost or more accurately benefit/cost. Usually results in a list. 2. Structure in time General decision tree. Includes boxes for decision points. Circles for chance outcomes of decisions made. Line between the former 2. It is read from left to right Called a tree because main trunk and branches remind people of a tree. 3. Obtain the data a. Probability - Chance of a clinical event occurring. Made up of baseline population values as modified by personal factors. 2 b. Utilities - Sense of worth or value as evaluated by a particular individual. We suggest strongly that the patient be the one to express this utility. 4. Choose the best outcome by evaluating the whole tree at one time by maximizing expected utility. Synthesizes the information Requires nothing more than simple multiplication. The mechanics of “solving” a decision tree include the following. 1. The decision tree must include all relevant alternatives and their outcomes. 2. The probabilities of the outcomes that could result from a decision must encompass all possibilities. That is they must sum to 1. For example if patients who receive surgery have a probability of dying of 0.1, then the probability of surviving must be 0.99 (thus .01 + .99 = 1). 3. To calculate an expected outcome, multiply the probabilities assigned to branches extending from the chance node. 4. The value or utility of an outcome can be obtained. Utility is expressed as a number between 0 and 1. In the medical literature several formal methods are used to determine the utility of a health state. Among these are Standard Gamble, Time Trade-Off and Visual Analog Scale. Of these the Visual Analog Scale or “Feeling Thermometer” is the most direct and easiest to use, although not the most accurate. Utilities can be assigned by the physician, patient, caregiver, or typical members of society. If the Clinical Decision Analysis is for an individual patient the patient should (in our opinion) give his or her Utility. Utilities for different outcomes do not have to sum to 1. 5. To calculate an EXPECTED UTILITY of a chance node, multiply the probabilities assigned to branches extending from the chance node by the utility of each outcome and sum. At the decision node, pick the branch with the highest expected utility as the one that a “rational” person would use as the best decision. 6. Perform a sensitivity analysis on probabilities and utilities because uncertainty always exists. This is done by varying the literature or personally measured values over the plausible range. If the decision changes because of a different hierarchy of expected utilities, the decision is said to be sensitive to the particular values used. Using a hand calculator one can easily calculate the expected utility of each branch at the extremes of the probability range. One is better served by a computer program to calculate the overall relationship between probabilities and expected utility, especially if one wishes to graph the results. 3