| Medical advances have allowed patients in the United States a very high quality of care, but not without cost. In fact, growing healthcare costs have become a hot-button political issue as the nation faces what has been termed an "obesity epidemic" and many are threatened with chronic illness. Insurers have an interest in using methods from actuarial science and statistics to better understand and predict these costs. Actuarial science typically analyzes health care data using two-part, or frequency-severity, models. These models aim to describe first the probability of a healthcare event, such as a doctor visit, occurring. Then, conditional on this event occurring, the second part of the model describes the cost (severity) of the healthcare event. Together, a two-part model describes the total cost of healthcare. When considering several types of healthcare events, it may be helpful to use multivariate techniques that can capture association between events. For instance, a visit to the emergency room may require a follow-up office-based doctor visit. This thesis will consider methods to jointly model the frequency of different types of healthcare events. The models considered are regression models estimated using maximum likelihood. In each case, models will be evaluated based on clarity of inference, ease of estimation, and predictive ability using a held out sample from the Medical Expenditure Panel Survey. |
