Author + information
- George A. Diamond, MD, FACC∗
- ↵∗Address for reprints: George A. Diamond, MD, Cedars-Sinai Medical Center, Division of Cardiology, 8700 Beverly Boulevard, Los Angeles, California 90048.
Stepwise regression procedures are often used to identify a small set of variables that serve as important predictors of clinical outcome and to construct prediction models based on those variables. Several theoretical and practical limitations of this process are discussed and highlighted with a variety of examples from published reports. Wider appreciation of these limitations should encourage the development of more relevant models, and thereby improve the quality of clinical prediction.
☆ This work was supported in part by a Specialized Center of Research (SCOR) grant from the National Institutes of Health (HL-17651), Bethesda, Maryland and is based on a presentation given at the 1987 Regenstrief Conference (Quality and Cost-Conscious Cardiovascular Care: Role of Decision Modeling) sponsored by the Regenstrief Institute for Health Care, Indiana University Medical Center, Indianapolis, Indiana.