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

Abstract #108743

Chin-Lin Tseng, DrPH¹, Mangala Rajan, MBA², Drew Helmer, MD, MS³, and Leonard Pogach, MD, MBA². (1) Preventive Medicine and Community Health, University of Medicine and Dentistry of New Jersey, 385 Tremont Avenue, VAMC #129, East Orange, NJ 07018, (973)6761000 ext 2028, tseng@njneuromed.org, (2) Diabetes Research Group, New Jersey East Orange VA Hospital, 385 Tremont, VAMC #129, East Orange, NJ 07018, (3) DVA-New Jersey Healthcare System, 385 Tremont Avenue, VAMC #129, East Orange, NJ 07018

Objective: It was essential in health outcomes research to develop a risk-adjustment model to generate risk-adjusted outcomes for fair comparisons. Such models have often been validated using less-than-optimal methods. The objective of this study was to apply a bootstrapping method to internally validate risk-adjustment models of amputation outcomes. Methods: The data were from veteran clinical users with diabetes. Baseline risks (age, sex, race, lower extremity risk factors, and medical comorbidities) in fiscal year (FY) 1997-8 and events of major and minor amputations in FY 1999 were ascertained. The .632 bootstrapping method was applied to evaluate the internal validation of the various nested multinomial logistic regression models of amputation outcomes for their c statistics. A total of 100 bootstrap sample sets (drawn with replacement) of the same size of the original dataset were utilized. Results: All independent variables, except for foot deformity, were consistently significant correlates of the amputation outcomes. Our final models (foot deformity removed) had a C statistic of 0.84 for major amputations, and a C statistic of 0.79 for minor amputations. The internally validated C statistics averaged 0.785 (standard deviation (SD)=0.005; range: 0.773~0.796) for major vs. no amputation, and averaged 0.745 (SD=0.006; range: 0.725~0.768) for minor vs. no amputation. Conclusion: As expected, the internally validated C statistic values were lower than the values obtained using the original dataset. These internally validated C statistics suggest that the final risk-adjustment model had a reasonable predictive power and they were not simply artifacts of the data used to develop the model.

Learning Objectives:

• 1.Recognize the importance of internal validation of a predictive model.
• 2.Understand different validation methods.
• 3.Describe the bootstrapping method and its application in health outcomes research.

Keywords: Health Service, Statistics

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.