Estimation of Dynamic Treatment Regimes and Extrapolation to Populations with Distinct Monitoring Structures: The Use of Laboratory Monitoring to Detect HIV Treatment Failure

• Activity Number: 576
• Type: Invited
• Date/Time: Thursday, August 6, 2009 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Maya L. Petersen*+
• Companies: University of California, Berkeley
• Keywords: Dynamic treatment regime ; individualized treatment rule ; inverse probability weight ; marginal structural model ; causal inference ; HIV

Abstract

Dynamic regimes are rules that assign treatment decisions over time in response to a subject's observed past. Marginal structural models can be used to estimate the effects of dynamic regimes, as well to estimate as the optimal regime within a given class. However, such estimates depend on the frequency with which the covariates used to assign treatment are measured. Covariate monitoring can differ between populations due to resource limitations and alternative approaches to diagnostic testing. This talk discusses how dynamic regime estimates from a study population can be extrapolated to a comparable target population in which a distinct monitoring strategy is employed. Methods are illustrated with an example from HIV, in which use of laboratory monitoring to detect treatment failure is investigated and the optimal covariate thresholds for modifying antiretroviral therapy are estimated.