Which Covariates Should Inform Treatment Decisions? Variable Importance in an Optimal Treatment Context

• Activity Number: 138
• Type: Contributed
• Date/Time: Monday, August 10, 2015 : 8:30 AM to 10:20 AM
• Sponsor: WNAR
• Author(s): Jeremy Coyle* and Alan Hubbard and Mitchell J. Cohen
• Companies: UC Berkeley and UC Berkeley and UC San Francisco
• Keywords: optimal treatment ; variable importance ; traumatic injury

Abstract

Optimal treatment methodologies answer a prescriptive question: "what is the optimal treatment strategy for a given patient based on covariates?". Extending this, variable importance can answer a descriptive question: "which covariates should inform treatment decisions?". By estimating a fully optimal treatment rule, and restricted optimal treatment rules (eliminating each covariate in turn), we can estimate the improvement in outcomes due to considering each covariate when making treatment decisions. This is directly relevant to precision medicine -- starting with the wealth of patient information available to a doctor making a treatment decision, we can narrow the covariates under consideration to the set of truly relevant factors, simplifying treatment decisions and improving outcomes. This methodology will be illustrated using data from longitudinal clinical studies of outcomes from acute trauma.