Proposed Guidelines for Interim Monitoring of Prospective Observational Studies Planning to Invoke Propensity-Scoring as the Primary Analysis

• Activity Number: 583 - Statistical Applications in Observational Studies
• Type: Contributed
• Date/Time: Wednesday, August 2, 2017 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Michael Swartz* and Thomas Jay Greene and Erin E Fox and Savitri N. Appana and Misung Yi and Soeun Kim and Elysia A. Garcia and Jada M. Johnson and Jeanette Podbielski and Jeffrey S. Tomasek and Charles E. Wade and John B. Holcomb and Stacia M DeSantis
• Companies: The University of Texas Health Science Center At Houston, School of Public Health and The University of Texas Health Science Center at Houston, School of Public Health and The University of Texas Health Science Center, Center for Translational Injury Research and The University of Texas Health Science Center at Houston and The University of Texas Health Science Center at Houston, School of Public Health and The University of Texas Health Science Center at Houston, School of Public Health and The University of Texas Health Science Center at Houston, School of Public Health and The University of Texas Health Science Center at Houston, School of Public Health and The University of Texas Health Science Center, Center for Translational Injury Research and The University of Texas Health Science Center, Center for Translational Injury Research and The University of Texas Health Science Center, Center for Translational Injury Research and The University of Texas Health Science Center at Houston and UTHealth SPH
• Keywords: Propensity Score analysis ; prospective observational study ; study design ; guidelines

Abstract

Propensity scores (PS) provide powerful tools to estimate causal effects from observational data. Traditionally, propensity scores are applied retrospectively in a secondary analysis of pre-existing observational data to provide an estimate of a treatment effect comparable to a randomized controlled trial (RCT) for an intervention or exposure of interest. Although results from such observational studies are not considered equivalent to RCT findings, some clinical researchers in the acute setting are beginning to consider prospective observational studies using a causal analysis, such as using PS, as the primary analysis when randomization may be infeasible or considered unethical. The Prehospital Resuscitation On Helicopter Study (PROHS) is one such study. Designed as a multi-center prospective, pragmatic observational study, PS were described in the protocol analysis plan to adjust for confounding. We present our experience and lessons learned from PROHS (PS did not balance confounders across treatment groups due to dramatic disparities), as well as provide practical guidelines to consider when designing a prospective observational study with a propensity score analysis.