Abstract: Discriminant function analysis can be used to make predictions of the group to which a patient most likely belongs (e.g., groups defined by high/low risk of developing diabetic retinopathy within a year). The use of longitudinally observed data can be particularly powerful in discriminant analysis. In this talk I will give an overview of a flexible multivariate longitudinal discriminant analysis (LoDA) approach, where longitudinal markers of different types (continuous, counts, binary) are jointly modelled. We will explore the benefits of taking into account the level of uncertainty of group membership probabilities to improve classification. We will apply this multivariate approach to identify patients who develop sight threatening diabetic retinopathy (STDR) within a year. Diabetic retinopathy is a common complication of diabetes and early detection of STDR is important to avoid significant vision loss. The allocation rules and classification accuracy will be presented and discussed.

Speaker: Dr M Garcia-Finana, reader in Biostatistics based at University of Liverpool, with expertise in multivariate and longitudinal statistical modelling, stereological techniques, design and analysis of laboratory-based studies.

---