A system-theoretic approach to the clinical stage classification problem

In order to determine an appropriate treatment, it is important to classify the patients with a particular disease into a few stages in the aspect of disease advancement. Clinical stage classifications (or staging systems) are used for this purpose. For instance, they indicate the stage (or disease advancement) for each combination of categories of prognostic factors. In the paper, two methods are proposed to solve the clinical stage classification problem; one using genetic algorithms (GA) and one using simulated annealing (SA). In both methods, the AIC (Akaike information criterion) is used as the objective function. The former method using GA tries to optimize statistically the staging system as well as the number of stages. In this method, a new crossover technique named “hyperplane crossover” is used, by which the characteristics of individuals (i.e. staging systems) can be maintained. The latter method using SA tries to optimize statistically the staging system only whereas the number of stages is fixed. To confirm the effectiveness and usefulness of the proposed methods, they are applied to a practical classification problem of pancreatic cancers. Data of 3801 patients who were registered in the Japanese Congress of Pancreatic Cancer (JCPS) and followed up over one year are used. Staging systems obtained by the two methods were better than the one given by JCPS