Modeling Complex Medical Decision Problems with the Archimedes Model

How often should women 40 to 50 years of age undergo mammography? Could an HIV vaccine with only limited effectiveness be a cost-effective means of slowing the spread of HIV infection? What is the best work-up for a patient who has a suspected pulmonary embolism?

In the ideal world, one would like free, fast, and ethical clinical trials that would provide the evidence needed to answer such questions. In the real world, these clinical trials may be too time-consuming, too expensive, unethical, or even impossible to perform. How, then, can we obtain useful answers about appropriate methods for disease prevention and treatment? We need a structured framework that uses the best evidence, imperfect as it may be, and captures relevant complexities and interactions. Decision models fulfill these requirements.

A Markov model is one way to represent a decision problem. A Markov model represents a dynamic process (for example, a patient's health over time) by using a set of distinct states (for example, benign tumor, malignant tumor, or death) and transition probabilities that describe how the disease process transitions from one state to another over time (for example, the probability that a benign tumor will become malignant). The analyst enters the states, the transition probabilities, and the outcomes into a computer program that simulates the time course and outcomes for each of many patients and calculates average values. Markov models are commonly used to analyze medical problems. A recent MEDLINE search on “Markov model” generated almost 3000 hits.

The article by Eddy and colleagues in this issue (1) evaluates the potential cost-effectiveness over 30 years of a lifestyle modification program used in the Diabetes Prevention Program (DPP) …