Clinicians and patients with complex chronic illnesses like type 2 diabetes face numerous potential treatment choices at every 10-minute primary care visit. Computer modeling reveals that the value of these choices is extremely sensitive to individual patient characteristics, including the number and kinds of treatment already in place. Committee-developed, printed guidelines cannot capture this complexity and, partly as a result, are not always used. Validated computer simulation models have the potential to do much better. We propose to use the Global Diabetes Model (GDM), a validated, comprehensive micro-simulator of diabetes treatments and complications, to test the potential of simulation-based treatment prioritization in complex chronic illnesses. We will compare GDM's recommendations against the actual clinical choices of primary care clinicians at Kaiser Permanente Northwest Region (KPNW), a large non-profit medical care system with excellent informatics resources and a record of very high quality diabetes care. We will perform three specific in silico experiments. Experiment 1 will estimate the extent to which one additional model-recommended treatment per patient would save total discounted medical care costs over 20 years, and increase quality-adjusted life-expectancy, compared with observed behavior. We also will measure differences in specific complication events and cost-effectiveness. Experiment 2 will measure the net improvement that could result if the best model-recommended treatment were substituted for the worst existing treatment. Experiment 3 will measure the sensitivity of the benefits identified in Experiment I to assumptions about the downstream stability and quality of medical care. Several modifications to the GDM will be necessary to conduct these studies and to prepare the model for future real-world implementation trials. We will add the capacity for additional treatments and treatment algorithms; add database structures to recognize additional treatment attributes; and build new, prioritization-optimized front- and back-ends onto the model.
