Willi Horner-Johnson, PhD¹, Gloria L. Krahn, PhD, MPH², and Charles E. Drum, JD, PhD¹. (1) Oregon Office on Disability and Health, Oregon Health & Science University, PO Box 574, Portland, OR 97207-0574, 503-494-9273, hornerjo@ohsu.edu, (2) Oregon Institute on Disability & Development, Oregon Health & Science University, PO Box 574, Portland, OR 97201-0574

The BRFSS Health Behaviors and Outcomes study engaged in an iterative process of model building and testing to develop a multiple-variable predictive model of positive health status. Beginning with analyses of data from the 1998 Behavioral Risk Factor Surveillance System aggregated across 14 states and D.C., the project developed an initial model that included demographic and socioeconomic variables, type of health insurance, individual health behaviors, and disability status. There were 1561 respondents with non-missing data for all variables. Results of regression analysis indicated that having a spouse or domestic partner, type of insurance, binge drinking, and vegetable consumption did not significantly contribute to differences in health status. The model was re-run without these variables. A total of 17,762 respondents had non-missing data for all remaining variables. The variables in the streamlined model accounted for 28.6% of the variance in health status. This model was then tested using the 2001 data set with disability identifiers for all 50 states. In this larger dataset, 175,905 respondents had non-missing data for all of the variables. The model accounted for 29.1% of the variance in health status. Disability, even when entered last, accounted for 7.3%. This finding led to the development of separate models for people with and without disabilities. Variables such as pain and need for assistance were significant, but relatively small, predictors of health status for people with disabilities. Details of the predictive model for people with disabilities will be discussed and contrasted with the model for people without disabilities.

Learning Objectives:

• Participants will gain greater understanding of multiple variable predictors of health status for persons with disabilities.
• Participants will learn about modeling methods for analyzing the BRFSS data.
• Participants will become aware of future research implications and opportunities for large-scale data sets to predict health status of persons with disabilities.

Presenting author's disclosure statement:
I do not have any significant financial interest/arrangement or affiliation with any organization/institution whose products or services are being discussed in this session.