Modelling Correlated Healthcare Costs with Many Zeros: A Two-Part Random Effects Approach

Abstract

Healthcare costs typically exhibit a substantial proportion of zero values together with very large positive values. For example, healthy people have no costs recorded in a given year whereas certain individuals may incur large medical expenses that increase tremendously by disease severity. Moreover, such semi-continuous data are often collected in hierarchical form and therefore correlated. A flexible two-part modelling approach is proposed to analyze the heterogeneous and correlated cost data. In the binary part, the odds of cost being positive are modelled using a logistic mixed regression model. In the continuous part, the mean cost given that costs have actually been incurred is assessed by a gamma mixed regression model. Random effects are incorporated within the two parts to account for correlation of the observations. Model fitting and inference are performed through the Gaussian quadrature technique, which can be implemented conveniently in statistical packages. The method is applied to evaluate the effectiveness of an occupational safety intervention program using longitudinal compensation claims cost data. The findings have important implications on healthcare administration and financial planning.