What Is or What If
Much of contemporary analytics focuses on tabulating and portraying characteristics of existing systems, whether they are for energy supply, health delivery or a wide range of other complex systems. This type of analytics addresses “what is” or in many cases “what was.” This approach is backward looking, which makes a lot of sense if there are important lessons to learn from the past and carry forward.
There are some situations, however, where the current system is not one to be emulated. Health delivery is one of these cases. While medical science has steadily advanced, the delivery of health has not. The delivery system is a federation of millions of entrepreneurs with no one in charge. Information systems are highly fragmented and rife with incompatibilities. The incentive system rewards delivery of procedures rather than health outcomes.
We need a very different system in terms of how health delivery is organized, operated, and financed. This requires that we move from “what is” to “what if” in the sense that we need to explore delivery models that do not yet exist. We cannot rely on empirical data from systems that have not been designed and deployed. Further, as these will be very expensive systems, we need some way to drive the future before we write the check.
Computational approaches can provide the means to this end. What we need is interactive organizational simulations that enable key stakeholders to explore alternatives, eliminate bad ideas, and refine good ideas. As stakeholders come from a wide range of disciplines, these simulations have to include compelling interactive visualizations that allow extensive “what if” explorations. When stakeholders move the sliders for key model parameters and choose their own assumptions, they become increasingly committed to the shared models they are developing.
Creation of these types of capabilities requires several ingredients. First, several types of computational models must be linked, e.g., agent-based for patients, discrete-event for delivery processes, microeconomic for providers and payers, rule-based for policy, and system dynamics for exogenous phenomena. Linking such a disparate range of models can be a substantial challenge.
Second, the parameters for component models must be gleaned from large data sets including clinical data, financial data, and claims data. Using such data to parameterize process models, for example, can be quite difficult, as most providers and payers have not structured their data sets in terms of processes. Instead, data are organized by codes for diagnoses, procedures, and locations. This requires that processes be inferred from data sets never intended to support such inferences, which often involves filtering out special cases as well as mistakes.
Third, interactive visualizations are needed for decision makers to understand and be comfortable with computational approaches. They need to view the computational models as a means for exploring a range of possibilities rather than as a “magic box” that produces optimal but, unfortunately, often opaque answers. This requires core competencies in interactive computing and decision support systems.
The three competencies outlined above — computational modeling, statistical estimation, and interactive visualization — are rarely found in one individual or even one discipline. Multiple disciplines are needed, working as a team to tackle large-scale “what if” problems. Any initiatives to address the transformation of healthcare requires the multi-disciplinary team needed to assure the availability of these competencies.
Smart health is not just about doing what we now do better. Indeed, Peter Drucker has cautioned us to never invest in improving something that you should not be doing at all. For health, being smart means being able explore whole new ways of doing things to eliminate bad ideas and refine good ideas, so that we can then invest in improving and deploying these good ideas to create quality, affordable health for everyone.