209760 Hospital Choice: Modeling selection to improve an instrumental variable

Monday, November 9, 2009: 9:10 AM

Michael Baiocchi , Wharton Department of Statistics, University of Pennsylvania, Philadelphia, PA
We collected birth certificates from all deliveries occurring in Pennsylvania and California between 1/1/95 and 6/30/05. These birth certificates were linked to death certificates by each state's department of health using name and date of birth. These linked records were then matched to maternal and newborn hospital discharge. Using this data set we have developed several models of patient selection of hospitals –these models are then used to create instrumental variables (IVs) for observational studies of treatment procedures. These models are both interesting for their description of patient-level hospital selection and for their potential use as IVs.

Learning Objectives:
Discuss the methods linking birth and death certificate and to demonstrate their use in patient selection models of hospitals and as instrumental variables.

Presenting author's disclosure statement:

Qualified on the content I am responsible for because: I am a fourth-year PhD candidate in the statistics department at the University of Pennsylvania and I am presenting in “3065.0 – Statistics Section Student Research Competition.” I am the second author (lead statistical author) on a peer-reviewed manuscript accepted for publication by Health Services Research. The paper is “The role of outpatient facilities in explaining variations in risk-adjusted readmission rates between hospitals,” which deals with the problem selection bias/collinearity causes in estimating the quality of outcomes attributed to a particular hospital. I am the lead author of the paper “Building a Stronger Instrument in an Observational Study of Perinatal Care for Premature Infants” – with coathurs Dr. Dylan Small, Dr. Scott Lorch, and Dr. Paul Rosenbaum. The paper is under review at the Journal of the American Statistical Association (JASA) – the top journal for new statistical methods. In this paper we develop a nonparametric approach to instrumental variables. The method is matching-based and can be considered an improvement of existing propensity score matching techniques (developed by Paul Rosenbaum and Don Rubin), as the method addresses unobserved selection bias as well as observed selection bias. In this paper we use spatial characteristics of the dataset to introduce exogenous variation into the selection of treating NICU in order to estimate the treatment effect. The research I will be presenting at APHA is an extension of the work done in the manuscript we have under review at JASA.
Any relevant financial relationships? No

I agree to comply with the American Public Health Association Conflict of Interest and Commercial Support Guidelines, and to disclose to the participants any off-label or experimental uses of a commercial product or service discussed in my presentation.