Archimedes, a simulation model for healthcare developed by the Biomathematics Unit of Kaiser Permanente’s Care Management Institute, creates a “virtual reality” in which patients, providers and institutions interact as they would in the real world.
Differential equations and algorithms—written by Leonard Schlessinger, head of the Biomathematics Unit, and his team—describe the anatomy, physiology and progress of diseases. For example, coronary artery disease is modeled as the gradual blockage of the arteries. Equations calculate the location and extent of the blockage the occurrence of symptoms; and the outcomes of the disease, based on data from epidemiological studies and clinical trials. Algorithms model the actions of patients and caregivers: patient behavior in seeking care; the performance of tests and treatments by physicians.
Archimedes lets researchers try different treatments or change the processes of care for a disease and then explore the effects. For the four diseases it models so far, Archimedes creates thousands of simulated “people” at risk of getting or who already have a condition. These “people,” each with different characteristics, grow older, get diseases, interact with the healthcare system, and live with (or die of) their diseases in similar fashion to populations in the real world. “If you’re looking at evaluating any kind of care management program with different protocols, you can program them into Archimedes and see what the results are in both outcomes and costs,” explains Matt Stiefel, associate executive director of the Care Management Institute.
As David Eddy, M.D., the visionary behind Archimedes, explains it, the starting point is to “sit down with experts who know all about the disease” and have them describe the symptoms, impact on patients and suggested treatments. At the same time, Eddy and his small team then map out the symptoms associated with progression of the disease, such as the rise of glucose levels in the case of diabetes. The software model incorporates equations that correspond to these features, as well as patient demographics. “There can easily be 50 to 100 biological variables pertinent to any particular disease,” Eddy says. To build a model of a single disease takes more than a year and costs from $500,000 to $1 million, he estimates.
Once it has modeled a disease, Archimedes can be used to evaluate a new guideline, say, for patients with chest pain: the choice of tests, the decision to hospitalize, the timing of treatments and so on. It could give a cost breakdown and describe the probable outcomes from treating a specific population with cholesterol-lowering agents. It could tell a healthcare delivery system where best to use its case managers.
By providing quantitative information, “Archimedes can help you make decisions about what to make a priority, or where to spend money,” Schlessinger says. Archimedes itself doesn’t make the decision. It just tells you how much something will cost and what the benefits could be. For instance, if you decide on a certain treatment for 1,000 people with angina, “Archimedes can tell you that you’ll have 72 fewer heart attacks and it will cost you $6 million to prevent those heart attacks,” says Schlessinger. “You can then ask whether that is the best way to allocate the resources.”
Archimedes is different from other software models because it simulates both the biological aspects of disease and the logistics of healthcare delivery. “Archimedes goes down to the continuous process of disease,” says Stiefel. “The simulation lets you change all of the relevant factors at the level of clinical decision making.”
Archimedes uses 200 Pentium 3 and Pentium 4 computers running on Windows 2000. The grid computing software is supplied by United Devices Inc. According to Schlessinger, Archimedes running on a single PC takes about 40 hours to simulate a diabetes trial; it takes less than an hour using the grid. Eddy is working on making a version that is accessible over the Web available in the next two years.