Trajectories Through the Disease Process: Cross Sectional and Longitudinal Studies

This paper explores the use of two different techniques for building models of disease progression from clinical data. Firstly, it explores the use of non-stationary dynamic Bayesian networks to model disease progression where the underlying model changes over time (as is common with many diseases where some tissue or organ becomes damaged throughout the duration of disease progression. Secondly, the fitting of trajectories through cross-sectional data in order to build models of progression from larger cohorts but without any stamps. The methods are applied to simulated data and real clinical data based on visual field tests from sufferers of glaucoma, the second largest cause of blindness in the world. Results demonstrate the importance of integrating cross-sectional and longitudinal data, both of which offer different advantages to understanding disease progression, and the use of models that account for changing underlying structures.