Using hierarchical models for the identification of factors associated with best practices and hospital profiling: Application to California birth data


Back to Annual Meeting Page		American Public Health Association 133rd Annual Meeting & Exposition December 10-14, 2005 Philadelphia, PA

3059.0: Monday, December 12, 2005 - 8:30 AM

Abstract #106177

Using hierarchical models for the identification of factors associated with best practices and hospital profiling: Application to California birth data

Moshe Fridman, PhD, Statistics, AMF Consulting, Inc., 1018 So Stanley Ave., Los Angeles, CA 90019, 323 857 6618, fmoshe@sbcglobal.net, Lisa M. Korst, MD, PhD, Departments of Pediatrics and Obstetrics & Gynecology, USC Keck School of Medicine, 4650 Sunset Boulevard, MS#30, Los Angeles, CA 90027, and Kimberly D. Gregory, MD, MPH, Department of Obstetrics & Gynecology, Divison Director, Maternal Fetal Medicine, Cedars Sinai Medical Center & UCLA School of Medicine, 8700 Beverly Blvd. Suite 160 West, Los Angeles, CA 90048.

Hierarchical models have become the models of choice for the analysis of multilevel nested data were individuals are grouped into larger units and inferences need to be adjusted for correlations among outcomes of individuals within a unit. This methodology allows separate modeling of two or more levels of random variation (error terms) such as within and between healthcare providers outcomes. The main advantages that this class of models provides are flexibility in specification of first and second moments for the error terms' distributions and the ability to work with small sample sizes or event rates in specific units by pooling information from other levels of the model. We implement a two-level hierarchical logistic regression model that adjusts for patient case mix variables (level 1) and hospital characteristics (level 2). Statewide patient data was obtained from a maternal and newborn linked hospital delivery discharge and birth certificate data for 2002 from the California Office of Statewide Health Planning & Development. To collect detailed hospital data we conducted structured interviews of nurse managers at hospitals in California reporting greater than 50 deliveries in calendar year 2002 regarding their hospital resources and clinical policies. This study was IRB approved. We illustrate two potential applications for the model: (1) Identification of hospital level factors affecting outcomes (route of delivery, maternal/neonatal outcomes), and (2) Profiling hospital variation in outcome rates as a potential marker for best practices and/or quality and safety issues.

Learning Objectives:

At the conclusion of the session, the participant will be able to

3. Describe the main advantages of hierarchical models and identify potential areas of their application.
6. Recognize the benefits of using hierarchical models with multilevel data.
9. Evaluate applications using hierarchical models that identify hospital factors contributing to best practice or profile healthcare providers.

Presenting author's disclosure statement:

I wish to disclose that I have NO financial interests or other relationship with the manufactures of commercial products, suppliers of commercial services or commercial supporters.

Recorded presentation

Statistical and Modeling Techniques for Health Outcomes Research

The 133rd Annual Meeting & Exposition (December 10-14, 2005) of APHA