|
Abstract:
|
Collecting information on multiple longitudinal outcomes is increasingly common in many clinical settings. In large datasets (with large numbers of patients and large numbers of measured variables) computational challenges can inhibit the ability to perform joint analysis of longitudinal variables. This talk will describe a mean field variational Bayes algorithm (MFVB) we have developed for multivariate generalised linear mixed models with longitudinal data with different outcome types (Gaussian, Poisson, binary). Variational Bayes can often estimate posterior means very well, but give poor estimates of variance. We will use simulation studies to assess factors affecting the accuracy of MFVB estimates compared to standard Markov chain Monte Carlo. We will assess the computational time, parameter accuracy and covariance estimation, which can often be poor in MFVB approaches. Clinical applications (in the field of diabetic retinopathy and liver cancer) will be used to show that MFVB can give fast and accurate estimation of complex longitudinal models.
|
