PMA7: CONTROLLING FOR BIASES FROM MEASUREMENT ERRORS IN HEALTH OUTCOMES RESEARCH USING STRUCTURAL EQUATION MODELING

Abstract

OBJECTIVES: Measurement errors in independent variables may lead to attenuated estimates of their effects and may contaminate estimates for other covariates in conventional linear regression models (LRM). However, the direction and magnitude of these biases are difficult to determine theoretically. Measurement error is a serious problem in health services research as health status is a latent variable that can only be measured with error using proxy variables. This study empirically evaluated the validity of LRM models in health outcomes research. METHODS: SEMs with a latent health construct are proposed and compared with LRM models to examine the bias of measurement errors in general health status using data from a study of the impact of pharmacist's consultation on both health outcomes and costs (KP/USC study). Perceived health status at a given time point was modeled as a latent variable measured by the multiple scales of the SF-36. RESULTS: The latent health construct with multiple scales of the SF-36 and its SEMs for health outcomes and costs are empirically supported by the KP/USC data. SEM estimations of the latent health construct in both the measurement model and the structural model were all statistically significant with expected signs. As predicted, LRM estimates for the SF-36 scales were attenuated. However, there is no strong evidence that LRM estimations of treatment effect were contaminated by the measurement errors in the SF-36 or that the simultaneity between health outcomes and costs. CONCLUSIONS: Measurement errors in health status variables may result in attenuated estimates of health status effects on patient outcomes. Fortunately, careful study design can eliminate the contamination of treatment effect estimates due to errors in measuring health status. Moreover, SEM methods can be used to control both attenuation and resonation biases.