Multistate Models for the Evaluation of Screening Interventions in Family Designs

• Activity Number: 253
• Type: Contributed
• Date/Time: Monday, August 5, 2013 : 2:00 PM to 3:50 PM
• Sponsor: SSC
• Author(s): Laurent Briollais*+ and Yun-Hee Choi and Yildiz Yilmaz
• Companies: Samuel Lunenfeld Research Institute and Western University and University of Toronto
• Keywords: Multistate model ; Family design ; screening ; survival analysis ; sampling bias ; colorectal cancer

Abstract

In diseases where multiple outcomes or events can occur over time, multistate models (MSM) are a popular approach to analyze the study data. The outcomes are defined as states, which could be first and second disease events and/or death. We propose here a MSM to evaluate screening interventions in Lynch Syndrome (LS) families. As a clinical disorder, LS is defined by the clustering of related cancers (colorectal, endometrial, etc.) across generations of kindreds and sometime within the same individual. We propose to extend the illness-death model to account for screening interventions, where the first state is the study entry, the second state is the occurrence of screening and state three, the occurrence of a first cancer. From this model, we derive the transition probabilities from Study Entry to a First Cancer, with or without going through Screening. We develop a general likelihood framework for MSMs applicable to family designs that can also handle missing genetic data among family members. Finally, we perform extensive simulations and a real data application to LS families from the Colon Cancer Family Registries to illustrate the interest our novel method.