The Importance of “Shrinkage” in Subgroup Analyses
Presented as an abstract at the Ambulatory Pediatric Association Region IX and X meeting, February 1998, Carmel, CA; and an abstract at the annual meeting of the Society for Academic Emergency Medicine, May 1998, Chicago, IL.
Received 18 February 2009; received in revised form 10 November 2009; accepted 4 January 2010. published online 08 February 2010. Corrected Proof
Study objective
Subgroup analyses examine associations (eg, between treatment and outcome) within subsets of a larger study sample. The traditional approach evaluates the data in each of the subgroups independently. More accurate answers, however, may be expected when the rest of the data are considered in the analysis of each subgroup, provided there are 3 or more subgroups.
Methods
We present a conceptual introduction to subgroup analysis that makes use of all the available data and then illustrate the technique by applying it to a previously published study of pediatric airway management. Using WinBUGS, freely available computer software, we perform an empirical Bayesian analysis of the treatment effect in each of the subgroups. This approach corrects the original subgroup treatment estimates toward a weighted average treatment effect across all subjects.
Results
The revised estimates of the subgroup treatment effects demonstrate markedly less variability than the original estimates. Further, using these estimates will reduce our total expected error in parameter estimation compared with using the original, independent subgroup estimates. Although any particular estimate may be adjusted inappropriately, adopting this strategy will, on average, lead to results that are more accurate.
Conclusion
When multiple subgroups are considered, it is often inadvisable to ignore the rest of the study data. Authors or readers who wish to examine associations within subgroups are encouraged to use techniques that reduce the total expected error.
aDepartment of Emergency Medicine, Harbor–UCLA Medical Center, Torrance, CA
bDepartment of Medicine, David Geffen School of Medicine at UCLA, Los Angeles, CA
cLos Angeles Biomedical Research Institute, Torrance, CA
dGertner Institute for Epidemiology and Health Policy Research, Tel Hashomer, Israel
Address for correspondence: Ari M. Lipsky, MD, PhD, Department of Emergency Medicine, Box 21, Harbor–UCLA Medical Center, 1000 West Carson St, Torrance, CA 90509; 310-222-3501, fax 310-782-1763
Supervising editor: Robert L. Wears, MD, MS
Author contributions: MG-H and RJL conceived the study. AML, MV, and RJL analyzed the data. AML drafted the article, and all authors contributed substantially to its revision. AML takes responsibility for the paper as a whole.
Funding and support: By Annals policy, all authors are required to disclose any and all commercial, financial, and other relationships in any way related to the subject of this article that might create any potential conflict of interest. See the Manuscript Submission Agreement in this issue for examples of specific conflicts covered by this statement. This publication was made possible by grant 1F32RR022167 from the National Center for Research Resources (NCRR), a component of the National Institutes of Health (NIH). Its contents are solely the responsibility of the authors and do not necessarily represent the official view of the NCRR or NIH. This work was previously supported by a Research Fellowship Grant from the Emergency Medicine Foundation to AML. The conduct and data analysis of the Pediatric Airway Management Project was supported by grant HS-09065-01 from the Agency for Healthcare Research and Quality, grant EMS-3036 from the State of California Emergency Medical Services Authority, and grant MCH064004-01-0 from the Bureau for Maternal and Child Health of the Public Health Service.
ABSTRACT