Combining Retrospective and Prospective Data To Improve Markov Transition Parameter Estimation for Characterizing the Accumulation of HIV-1 Drug Resistance Mutations

• Activity Number: 517
• Type: Contributed
• Date/Time: Thursday, August 10, 2006 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Brian Healy*+ and Victor DeGruttola and Marcello Pagano
• Companies: Harvard School of Public Health and Harvard School of Public Health and Harvard School of Public Health
• Keywords: hidden Markov models ; branching trees ; HIV resistance mutations ; constrained optimization ; genetic pathways

Abstract

Prospective studies of HIV-infected patients permit investigation of acquisition of new HIV resistance mutations, but patients often have many such mutations at study entry. We propose methods for modeling genetic pathways that combine Markov models for prospective data and branching trees for cross-sectional baseline data. Most links between the two sets of model parameters are functions of the time on treatment before study entry---information that may not be available. Nonetheless, some link functions eliminate the dependence on this time under certain assumptions. Approaches to fitting models to the combined information that use these functions include constrained maximization and weighted least squares. The latter easily accommodates error in the link function due to uncertainty in the necessary assumptions. Using both sources of information improves precision of the estimation.