Online Program
WITHDRAWN: Analysis of multivariate longitudinal health outcomes via generalized linear mixed models
Libby Dismuke, Center for Disease Prevention and Health Interventions, Ralph H. Johnson Vete Keywords: cost,GLMM, longitudinal data, multivariate outcome Objective: Many health outcomes are observed longitudinally and their evaluation is interdependent. Modeling these outcomes separately ignores the interdependencies and could lead to wrong inference. The objective of this study was to develop and assess methodology for modeling multivariate longitudinal cost outcomes that accounts for the interdependence and missing data. Study Design: The response variables were a vector of longitudinal cost outcomes (inpatient, outpatient and pharmacy cost) and the key covariates were medication adherence and mental health visits. We compared several approaches that are based on a generalized linear mixed model (GLMM) framework with varying assumptions on the joint distribution of the random coefficients (intercepts and slopes) to allow for inter and intra outcome correlations. We used simulated data to assess the performance of these methods in the presence of missing data and we used data from a national cohort of 740,195 veterans with diabetes (followed 2002-2006) to demonstrate their applications. Results: Estimates of healthcare cost differences by covariates from the joint modeling approach had more accurate estimate of standard errors and were in the expected direction than those from separate models. Simulation results indicated that the proposed approach was robust to missing not at random. Conclusions: Using the proposed multivariate generalized linear mixed model (mGLMM) approach to model multiple source cost data is beneficial as it allows for the interdependence among the multiple outcomes and for estimating the global significance and differential impact of covariates on each outcome.
|
Important Dates & Deadlines
- October 9 - 11, 2013
ICHPS 2013