The 131st Annual Meeting (November 15-19, 2003) of APHA

Abstract #64362

Donald E. Reed, PhD, School of Health Sciences, Ohio University, Grover Center E317, Athens, OH 45701, 740-597-1695, reedd@ohio.edu

Individual health care utilization data are inherently non-normal. Data are in the form of low-valued, non-negative integers, and exhibit over-dispersion(excess zeros). Nonlinear statistical models exist,but may be difficult to interpret and not included in many statistical software packages.

During the 1990's, Georgia moved much of its Medicaid population into a managed care program(GBHC). Transition required sixteen quarters, permitting comparison of utilization by members with non-members for fifteen quarters. Analyses reported to date have been based on nonlinear models,however it is instructive to compare these results with those obtained from linear models.

Reduction of unnecessary specially physician services was a goal of GBHC. Two-stage regression models were developed to measure GBHC’s impact. The first stage measured the probability of a Medicaid recipient having at least one specialty provider contact in a quarter. The second stages measured the frequency services were provided – given at least one contact. The dependent variable was contact / no contact for the first stage, and frequency of contact for the second. Independent variables were GBHC membership and demographic information – coded as indicator variables.

GBHC membership was found to be associated with a statistically reduction in contact by the linear model in fourteen of fifteen quarters and in thirteen of fifteen quarters by the non-linear model. Linear models detected no statistically significance differences in frequency in any quarter, while the non-linear models detected significant differences in three of fifteen quarters. In general, standard errors for the GBHC variables from linear models were smaller than those from non-linear models.

Learning Objectives:

At the conclusion of the session, the participant in this session will be able to:
• Compare and contrast the performance of linear and non-linear regression models in health care utilization studies,
• Select the appropriate model based on the purpose of the study.

Keywords: Statistics, Health Care Utilization

Presenting author's disclosure statement:
I have a significant financial interest/arrangement or affiliation with any organization/institution whose products or services are being discussed in this session.
Relationship: During a portion of his doctoral studies, the author was employed as a part-time data analyst for the Georgia Department of Community Health, the agency which administers the state's Medicaid program.