Osteoporosis is a major health problem in our aging society. As a silent disease, diagnosis and risk assessment of osteoporosis rely on diagnostic tests. In the past decade, a great number of diagnostic techniques have been developed and used to assess bone density and quality. The accuracy and cost of these tests vary greatly. With such a variety of diagnostic tests, the clinicians must determine the best diagnostic strategy for specific patient populations, both for screening and for selecting and monitoring treatment. Because of the lack of appropriate statistical tools to assess diagnostic utility of combining multiple tests in their accuracy in predicting osteoporotic fractures and cost-effectiveness, it is difficult to identify the optimum combination of tests.
The specific aims of our study are the following: 1. Development of receiver operating characteristics (ROC) region analysis for parallel tests. 2. Development of recursive partitioning trees for test series using cost criteria. 3. Development of cost-effectiveness models of optimal decision thresholds for sequential tests. We will study theoretical properties, develop computer software for above statistical tools and apply them to examine the combinations of DXA, SXA, QCT, QUS, biomarkers, and epidemiological risk factors such as age, weight, fall rate, family and fracture history, etc., based on data from the Study of Osteoporotic Fractures (SOF) and simulation models using cost-effectiveness and population distribution parameters published in literature.
