Regency B
Mediation Analysis with an Ordinal Outcome Using Empirical Data (303932)
Suad Ghaddar, The University of Texas Rio Grande Valley
*Kristina P Vatcheva, The University of Texas Rio Grande Valley
Keywords: ordinal outcome, mediation analysis
Mediation analysis is a powerful tool in health related research to disentangle the indirect effect of an exposure on an outcome through a given mediator variable. When analyzing ordinal outcome variable ordinal logistic regression is recommended to take into account the ordinality. We investigated the mediation effect of Telehealth attitudes on the effect of health information knowledge on Telehealth services use using South Texas community health survey data. We used the common approach by fitting two regression models but in the settings of an ordinal outcome; and the Structural Equation Modeling (SEM) technique. Both methods detected that Telehealth attitudes fully mediated the effect of the predictor on the outcome. Using proportional odds regression model we generated the natural indirect effect (NIE) with a meaningful interpretation: by affecting the Telehealth attitudes, for each increase in health information knowledge score the odds of higher vs. lower health services increase by 5% (OR=1.05, 95% CI: 1.02, 1.08). SEM for the hypothesized mediation model was a good fit to the data, but the estimated effect was on a continuous scale (NIE =0.018, 95% CI: 0.006, 0.031).