The overall goal is to understand how patient information should be used to diagnose disease, both for the individual patient and at the level of population studies. Classical statistical techniques, including discriminant analysis, regression, and classification have recently been joined by newer technologies, including probabilistic expert systems, neural networks and fuzzy logic. At the population level, public screening programs aim at early detection of disease in the hope that early diagnosis will lead to more cures or longer survival. Implementation costs for those programs amount of billions of dollars. Issues such as the initial age to begin participating in a screening program, the number of and timing of diagnostic exams, optimal methods for incorporating accruing patient information over a sequential series of exams and differential screening strategies for populations at varying levels of risk are open questions. There are expanding efforts to plan programs to detect some chronic diseases early in the expectation that earlier diagnosis combined with therapy will result in greater cure rates and/or survival. This is especially true in cancer, diabetes and heart disease. The range of problems ranges from the planning and analysis of early-detection clinical trials to the planning of public health programs. There are serious controversies in many of the applications to specific diseases. For example in breast cancer, randomized trials have shown that mammography is effective in reducing mortality only for women over 50 years of age. The American Cancer Society recommends that women over 50 should receive and annual mammogram, yet in Scandinavian countries the recommendation is every two years and in the United Kingdom every three years. Nearly all of these controversies may be clarified by development of new statistical theories. Probabilistic models, incorporating trial results, can lead to the solutions of many of the problems for specific chronic diseases.

The design of mass screening trials and programs can benefit from decision theoretic analysis methods aiming at minimizing costs and optimizing health outcomes. The statistical issues raised here impact most areas of statistics and health.

The statistical issues include classification/discriminate analysis, regression, classification and newer technologies, including expert systems, notably probabilistic systems and neural networks. At the population level, screening studies are important. The statistical issues include length-based sampling, number and scheduling of screening studies, and differential screening strategies in populations at varying levels at risk. The statistical issues raised here impact most areas of statistics and health.

top of page