Use of Area-level Exposures as Instruments to Study the Effects of Treatment on AMI Mortality

• Activity Number: 177
• Type: Topic Contributed
• Date/Time: Tuesday, August 10, 2004 : 8:30 AM to 10:20 AM
• Sponsor: Section on Health Policy Statistics
• Author(s): Therese Stukel*+ and David Wennberg and Lee Lucas
• Companies: Institute for Clinical Evaluative Sciences and Maine Medical Center and Maine Medical Center
• Keywords: instrumental variable ; observational study ; selection bias ; health services ; propensity score

Abstract

We compared survival of heart attack patients exposed to different levels of medical and invasive management in their region of residence. Studies comparing these therapies need to account for selection bias as to who receives treatment. We undertook a national retrospective cohort study of 158,831 Medicare patients from the Cooperative Cardiovascular Project (CCP) database. Region of residence met the criteria for an instrumental variable (IV). We compared treatment effects using propensity score models (PSM) with patient-level exposures; instrumental variable (IV) techniques adapted to generalized linear models; and standard risk-adjustment using area-level exposures. PSM and adapted IV did not remove patient selection bias. Analysis using area-level exposures appeared to be equivalent to IV in the non-OLS setting, and to remove unmeasured confounding. Increased regional intensity of either treatment improved survival; no increased survival was seen with more invasive management beyond that seen with medical management. We interpret our findings in the context of observational studies with binary, censored outcomes in the presence of selection bias.