TL32: Statistical Considerations for Handling Treatment Switches in Observational Studies
Deborah Bookwalter, Quintiles   *William Hawkes, Quintiles RWLPR  Christina Mack, Quintiles  Zhaohui Su, Quintiles  Priscilla Valentgas, Quintiles 

Keywords: observational research, treatment switch, intention to treat, time-dependent variable, marginal structural model

Statistical Considerations for Handling Treatment Switches in Observational Studies William Hawkes, Zhaohui Su, Christina Mack, Deborah Bookwalter and Priscilla Velentgas

Treatment switches are one of the most complex issues faced in the analysis of observational data. Medication changes, particularly switches from one study medication to a comparator, can dilute effective sample size, reduce statistical power and create challenges in analysis and results interpretation. Various solutions, including Intention-to-Treat (ITT) and Per-Protocol analyses, treating exposure as a time-dependent variable, Inverse Probability Weighting, and Marginal Structural Models have been proposed. This roundtable discussion will promote discussion and sharing of ideas among practicing statisticians from government, industry, and academia on this difficult and common issue. The moderator will be William Hawkes, PhD, Statistical Scientist, Quintiles Real World and Late Phase Research division.

Questions for Roundtable Participants: 1. How have you handled treatment switches in your own statistical work? 2. What are the strengths and weaknesses of these approaches? 3. What are the best practices to account for treatment switches in specific settings?