|
Abstract:
|
Copulas provide a way to jointly model data that may come from discrete, continuous, or a mixture of discrete and continuous distributions. Annual healthcare costs such as inpatient, outpatient, nursing home, and prescription costs commonly have an inflated zero distribution and are often correlated, even after adjusting for patient-specific differences. In this talk, we will discuss the joint modeling of these correlated costs. We propose using zero-inflated or two-part models to marginally describe each cost combined with a copula to correlate them. This approach provides a convenient expression for the joint distribution of costs, such as the total overall cost. The benefit of using this approach over alternatives will be demonstrated using simulated and real data.
|
