A comparison of generalized linear model, two-part model, and finite mixture model in the estimation of healthcare utilization and expenditures for obese individuals.

摘要

Obesity has become an epidemic. Approximately one-third of the American population is affected by obesity. The number of overweight and obese Americans has continued to increase since 1960, a trend that is not slowing down.;National costs attributed to both overweight and obesity medical expenses accounted for 9.1% of total U.S. medical expenditures in 1998 and reached to ;This study is designed to compare different econometric models that describe healthcare expenditures for obese individuals while controlling for the effects of demographics, comorbid conditions and health behaviors such as physical activity and cigarette smoking.;This is the first study to compare a generalized linear model (GLM), a two-part model, and a finite mixture model (FMM) in the estimation of healthcare utilization and expenditures for obese individuals. Healthcare utilization and expenditures have traditionally been estimated by means of Ordinary least Squares (OLS).;Healthcare utilization data are highly positively skewed data and have many zeros. Thus, healthcare utilization does not meet the normal distribution assumptions that are required by OLS.;In the literature, GLM has been proven superior to OLS in the analysis of healthcare utilization and expenditures; however, excess zeros are a problem that GLM does not account for. The two-part model accounts for the "zero-problem" by combining a logistic regression on any use of the service and linear regression on the log of positive charges given use of the service. FMM is an alternative to the traditional two-part models that classifies the sample into sub-samples of infrequent users and frequent users.;According to the results, the FMM model is superior in the prediction of healthcare utilization and expenditures. It is already established that obese individuals utilize more healthcare than normal or overweight individuals; nevertheless, the results from this study suggest that there is more than one subpopulation within obese individuals: those who utilize healthcare resources occasionally and those who utilize healthcare resources more often.;Accurate prediction of healthcare expenditures is of enormous importance for decision-making authorities in the appropriate allocation of resources, in the determination of risk population, and in the decrease of per-capita healthcare utilization.