A New Method for Combining Experimental and Observational Data to Estimate Population Treatment Effects

• Activity Number: 350
• Type: Contributed
• Date/Time: Tuesday, August 3, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Health Policy Statistics Section
• Author(s): Richard Grieve and Erin Hartman and Jasjeet Sekhon*+
• Companies: London School of Hygiene and Tropical Medicine and University of California, Berkeley and University of California, Berkeley
• Keywords: comparative effectiveness analysis ; causal inference ; matching ; observational studies

Abstract

Policy makers are often interested in obtaining population estimates. However, Randomized Controlled Trials (RCTs) are usually available only for non-random samples of the population. This is especially problematic for comparative effectiveness studies. We present a new method for combining information from RCTs and observational studies to estimate population estimates that requires weaker assumptions than extant methods. The method combines a machine learning matching method, Genetic Matching, with maximum entropy weighting to match population moments. Our approach allows one to use two different placebo tests to help assess the validity of the identifying assumptions of the method. Key properties are proved. And we apply the method to a comparative effectiveness analysis of a clinical intervention, Pulmonary Artery Catheterization.