Describing and analyzing complex disease history in retrospective studies

Abstract

Blood-related diseases are complex diseases with diverse origins, treatments and prognosis. In haematology studies, investigators are interested in multiple outcomes and multiple prognostic variables that may change value over the course of follow-up. These time-dependent variables can be of different nature. Time-dependent events such as treatment with haematopoeitic stem cell transplant (HCT) and acute or chronic graft-versus-host disease (GVHD) typically interact with outcomes respectively after diagnosis or HCT. Longitudinal measurement such as immune response do influence survival after HCT. Effect of these time-dependent variables on outcomes can be investigated using different approaches, such as time-dependent Cox regression, landmark analysis, multi-state models or joint modelisation. In this paper we review basic principles of these different approaches using examples from haematological studies.

Keywords: Joint models; Landmark; Longitudinal measurements; Multi-state models; Observational studies; Outcomes; Time-dependent variables; Time-to-event analysis.