Wei Wang1, Wanmei Wang2, Thomas H Mosley3, Michael E Griswold4. 1. Center of Biostatistics and Bioinformatics, University of Mississippi Medical Center, New Guyton Research Building G562, 2500 North State Street, Jackson, MS 39216, USA. Electronic address: wwang@umc.edu. 2. Center of Biostatistics and Bioinformatics, University of Mississippi Medical Center, New Guyton Research Building G562, 2500 North State Street, Jackson, MS 39216, USA. Electronic address: wwang2@umc.edu. 3. Department of Medicine, Division of Geriatrics, Memory Impairment and Neurodegenerative Disease Center, University of Mississippi Medical Center, Office Annex Building, 2500 North State Street, Jackson, MS 39216, USA. Electronic address: tmosley@umc.edu. 4. Center of Biostatistics and Bioinformatics, University of Mississippi Medical Center, New Guyton Research Building G562, 2500 North State Street, Jackson, MS 39216, USA. Electronic address: mgriswold@umc.edu.
Abstract
BACKGROUND AND OBJECTIVES: The joint modeling of longitudinal and survival data to assess effects of multiple informative dropout mechanisms on longitudinal outcomes inference has received considerable attention during recent years; related statistical programs to apply these methods have been lacking. This paper provides a SAS macro implementation of a shared parameter model to accommodate the analysis of longitudinal outcomes in the presence of multiple competing survival/dropout events. METHODS: In this macro, we assumed that the associations between the survival and the longitudinal submodels are linked through a set of shared random effects. The submodel for the longitudinal outcome takes the form of a linear mixed effects model, with specifications for the random intercept and/or random slope. The survival submodel allows up to three different competing causes for dropout, each allowing either an exponential or Weibull parametric baseline hazard function. In addition, information criterion fit statistics AIC and BIC are provided to assist with parametric baseline hazard function selection. RESULTS: We illustrate the SAS Macro in a cognitive decline study sensitivity analysis using data from the Atherosclerosis Risk in Communities Neurocognitive Study (ARIC-NCS). In addition, we also conduct a simulation study to show that the joint model provides unbiased parameter estimates when informative dropout exists compared against separate model approach which assumes missing at random dropout mechanisms. CONCLUSIONS: We have presented a SAS macro to implement a shared parameter model for a longitudinal outcome and multiple cause-specific dropouts and made the macro code freely available for download. Copyright Â
BACKGROUND AND OBJECTIVES: The joint modeling of longitudinal and survival data to assess effects of multiple informative dropout mechanisms on longitudinal outcomes inference has received considerable attention during recent years; related statistical programs to apply these methods have been lacking. This paper provides a SAS macro implementation of a shared parameter model to accommodate the analysis of longitudinal outcomes in the presence of multiple competing survival/dropout events. METHODS: In this macro, we assumed that the associations between the survival and the longitudinal submodels are linked through a set of shared random effects. The submodel for the longitudinal outcome takes the form of a linear mixed effects model, with specifications for the random intercept and/or random slope. The survival submodel allows up to three different competing causes for dropout, each allowing either an exponential or Weibull parametric baseline hazard function. In addition, information criterion fit statistics AIC and BIC are provided to assist with parametric baseline hazard function selection. RESULTS: We illustrate the SAS Macro in a cognitive decline study sensitivity analysis using data from the Atherosclerosis Risk in Communities Neurocognitive Study (ARIC-NCS). In addition, we also conduct a simulation study to show that the joint model provides unbiased parameter estimates when informative dropout exists compared against separate model approach which assumes missing at random dropout mechanisms. CONCLUSIONS: We have presented a SAS macro to implement a shared parameter model for a longitudinal outcome and multiple cause-specific dropouts and made the macro code freely available for download. Copyright Â