Citation

Finkelstein EA, Chen H, Miller TR, Corso PS, Stevens JA. Med. Care. 2005; 43(11): 1087-1091.

Affiliation

Pacific Institute for Reseaerch and Evaluation, Calverton, Maryland; National Center for Injury Prevention and Control, Centers for Disease Control and Prevention, Atlanta, Georgia.

Copyright

(Copyright © 2005, American Public Health Association, Publisher Lippincott Williams and Wilkins)

DOI

unavailable

PMID

16224301

Abstract

OBJECTIVES:: Although the case-crossover design has been used widely in epidemiological and cost-offset studies as an alternative to the case-control design, it is rarely applied to cost-of-illness studies. In this study, costs for a series of hospitalized and nonhospitalized fall-related injuries were computed using the 2 approaches to allow for a direct comparison of the results. RESEARCH DESIGN:: We used claims data from the Medicare fee-for-service 5% Standard Analytical Files. For the case-control design, those who sustained nonfatal fall-related injuries were tracked for 1 year after their first fall, and costs were compared, using regression analysis, to annual costs for a comparison sample of nonfallers. The case-crossover design used a modified regression approach that compared monthly costs of fallers before and after fall. RESULTS:: We present unit costs for falls requiring (1) a hospitalization resulting in a live discharge, (2) an emergency department visit not resulting in an admission, and (3) falls requiring office-based or hospital outpatient visits only. Using the case-control design, these costs were $22,260, $3890, and $5040 respectively. Using the case-crossover design, these estimates were reduced to $20,920, $3230, and $4200. CONCLUSIONS:: On average, estimates of the costs of fall injuries from the case-control design were between 6% and 17% greater than those from the case-crossover approach. These differences likely result from our inability to control for comorbidity differences between fallers and nonfallers in the case-control design. Under several scenarios, including unobserved heterogeneity between cases and controls, the case-crossover design, although computationally more intensive, produces more accurate results.