Multistate Modeling of Intermittent Observations with Application to Viral Load Measurements in HIV-Positive Patients

• Activity Number: 253
• Type: Contributed
• Date/Time: Monday, August 5, 2013 : 2:00 PM to 3:50 PM
• Sponsor: SSC
• Author(s): Narges Nazeri Rad*+ and Jerald F Lawless
• Companies: University of Waterloo and University of Waterloo
• Keywords: estimation ; intermittent observation ; Markov multistate models ; relative efficiency

Abstract

In many chronic disease studies where multistate models are used to describe aspects of disease progression, data on individuals can only be collected intermittently. For instance, in cohorts of individuals infected with HIV, the states in multistate models might be defined in terms of viral load levels, whose values are measured at intermittent visit times. In this situation, only the states occupied at the discrete observation times are known. The states visited between successive observation times and the exact transition times are all unknown. In this talk, we discuss the estimation of transition rates and covariate effects in Markov models, and examine loss of efficiency due to intermittent observation. In particular, we consider the effect of the time interval between visits, and the number of visits, on the precision of estimation in both progressive and reversible multistate models. We present applications involving an analysis of time to viral failure (viral rebound) in HIV-positive persons whose virus has been suppressed by combination antiretroviral therapy.