Analyzing Pregnancy Costs with Finite Mixture Models: An Opportunity to More Adequately Accommodate the Presence of Patient Data Heterogeneity

Abstract

The choice of a model in the analysis of patient health care costs and utilization is critical for a clear understanding of the behavior and estimation of quantities like incremental costs or cost-effectiveness. In studying heath care claims related to pregnancy, it would not be surprising that a small portion of the women have costs associated with their care and treatment that might be extreme or outlying. Many strategies exist for accommodating outliers; however, is one approach superior to the others because it may be implemented over a broader set of conditions without making unreasonable assumptions about the prevailing data characteristics? In this study, the author will show an example of a data set based on the medical claims for over 300K pregnant women, aged 15-49, where the traditional, or widely used Generalized Linear Model (GLM) approach to modeling costs may be less than optimal due to the presence of patients with very large, or very small expenditure values. These values, in some sense “contaminate” the typically employed GLM and cause it to violate its underlying requisite statistical assumptions. Finite Mixture Models (FMMs) have been employed in other areas of clinical research to model health care utilization. The author will introduce FMMs as an alternative to the commonly used GLM model and show that in his example data set, the fit of the FMMs is superior for the modeling of maternity expenditures in the presence of extreme or outlying cost values.