Topic: Title: Robust joint modelling: Revealing the impact of outliers Authors:  Lisa McCrink - Queens University Belfast (United Kingdom) [presenting]
Abstract: The use of joint modelling techniques has grown in popularity as rapidly as the collection of the data that such models analyse. The appeal of simultaneously analysing an individual's changing repeated measurements over time and the effect such changes have on their time-to-event process is evident, with literature confirming the improvements in estimation compared to independent models. Despite such growth in this research field, only a limited amount of study has investigated the impact of longitudinal outliers on the commonly used normality assumptions of the random terms. The need to investigate this further is highlighted both through a simulation study and an illustrative example exploring the factors that affect the survival of Northern Irish renal patients who are undergoing haemodialysis. The detrimental impact of outliers on both the accuracy and efficiency of estimates is demonstrated, whilst introducing novel methodology for accommodating outliers in one of the most common types of joint models, that which links a linear mixed effects model with a Cox proportional hazards model. Motivated by outlying patients shown to have significantly worse survival than typical renal patients, the focus is on the identification of outliers presented using graphical techniques which are shown to accurately identify outliers. This has the potential to aid clinicians to change the treatment plan of outlying renal patients to improve their survival rates.