Page - 39 - in Applied Interdisciplinary Theory in Health Informatics - Knowledge Base for Practitioners

e(x) = p(x) ™ u(x) (1) Expected utility is thus a measure of the actual benefit that can be expected from an event over multiple trials, given uncertainty about the event occurring. The expected VOI that helps us choose between two different courses of action can now considered to be the difference in the expected utility of the different decision options [9 10] i.e.: VOI = expected utility (Option 1) - expected utility (Option 2) (2) For example, assume that the probability of a clinician finding a new pharmacogenomic test result when interacting with a patient’s electronic record is 0.4 because the clinician usually must check the EHR several times before seeing a result. The utility of this result is high at 0.9, because it allows the clinician to choose between two different drug treatments. The Expected utility of this outcome is 0.36 i.e.: 0.36 = 0.4 x 0.9 In comparison, the probability of not finding the test result is 0.6. We might assign a utility to proceed without the test result of 0.1 (because there is a good chance that the drug is ineffective for most patients who do not have the gene). The expected utility of proceeding without a gene test is thus 0.06 i.e.: 0.06 = 0.6 x 0.1 We can now calculate the VOI for a clinician accessing a gene test result: VOI = 0.36 – 0.6 = 0.3 A key idea here is that for new information to have value, the information must be actionable in some way. It is not enough that data provide us with a new diagnosis, that diagnosis must then trigger some new action in the world [11]. The action needs to result for example, in a change in morbidity, mortality, or in some other way increase a patient’s quality of life. VOI could be negative if the proposed method to gather new information does not lead to an actionable decision with potential benefits, and gathering the data has costs for the patient such as risks of complications that lead to harm, from pain through to injury and even death. 1.4. The value of events along the information value chain can be quantified Now that we have a way of calculating the value of information for any step in the information value chain, we can turn to look at the way information value changes down the chain. We first look at the frequency with which events occur at each stage in a chain. For example, over a 24-hour period, the EHR in a hospital may be accessed thousands of times, but decision support systems may be accessed only hundreds of times. One interesting property of the information value chain is that there is typically an asymmetry both in the number of events at each step, as well as in the value of the events (Figure 2). Firstly, we note that there is a probability for moving from one step in the chain to another. Thus there is a probability (but not a certainty) that interacting with an E.Coiera /AssessingTechnologySuccessandFailureUsing InformationValueChainTheory 39