Seite - 46 - in Applied Interdisciplinary Theory in Health Informatics - Knowledge Base for Practitioners

electronic record, and calculate the expected utility to this point only. Alternatively, one could add decision support to the electronic record, which would change the utility and disutility associated with system use. Since some electronic records have decision support, and some do not, these separate calculations of utilities allow us to make comparisons using the value chain. For a decision tree, we calculate expected utility by multiplying the utility of a terminal node by the probabilities of each step in the path to that node. We calculate a similar path expected utility in a value chain, but can do so for each node in the chain (see Table 2). This path expected utility for a node in a value chain represents the expected utility of ending the chain at a given node. A related question is whether the utility of one node directly determines the value of the subsequent nodes. The answer is that earlier nodes in a chain do influence the utility of later ones, but not in an easily definable way. A value chain is typically an open world. Each node has a separate utility because different populations of patients and users, technologies and external factors all might contribute to each node’s utility. So whilst each earlier stage does shape downstream utility, we do not know the specific mathematical function that describes how it contributes, and there is no easy way to infer one directly from the other. For this reason we re-measure utilities at every node. Although value chain theory is essentially quantitative – it asks us to calculate the value of information at different steps – it is important to remember that in many cases we will be making qualitative comparisons between different stages in the chain. This means that in some cases where great precision in value calculation is difficult, approximating the value of information still allows meaningful qualitative comparisons to be made – usually where there is substantial difference in the VOI at different stages in the chain. As with any theory that relies on quantitative measurements, it is important to ensure that data used in any analysis actually measures what it is meant to. Standard epidemiological challenges such as dealing with confounding factors and noise, as well as temporal variations such as seasonality in disease and service patterns, all need to be addressed. It is important to recognize that value chain theory does not attempt to provide detailed mechanistic explanations for the impact of information technology beyond the causality implied in the structure of the chain itself. From this perspective it provides a lens to focus on areas of concern or benefit, and other approaches to analysis that assist in untangling the reasons for a particular outcome are then needed. Value chain theory can also help answer questions about the need for automation, and thus help decide which tasks should or should not be automated [19]. Recognizing that there will likely be different expected utility profiles for completing a task by machine or by human, we can calculate both profiles and plot the resulting curve to generate a summary profile (Figure 6). Undertaking this type of analytic exercise allows us to identify whether tasks are better automated, left to humans, or performed jointly [2]. Understanding the answer has fundamental implications for the strategy taken and its likelihood of success. Whilst the generic value chain in Figure 1 is applicable to a broad class of information and communication systems, there appears to be no theoretical restriction to imagining different chains of events, or adapting this chain to meet the needs of a specific setting, technology or purpose. One alternate formulation by Parasuraman et al. uses a simplified four step information processing model to create a similar pipeline [20], in contrast to the model used here, which is instead based on human decision making. E.Coiera /AssessingTechnologySuccessandFailureUsing InformationValueChainTheory46