Citation

Kalsbeek WD, Bowling JM, Morris PD. Am. J. Epidemiol. 1992; 135(5): 552-563.

Affiliation

Department of Biostatistics, University of North Carolina, Chapel Hill 27599-2400.

Copyright

(Copyright © 1992, Oxford University Press)

DOI

unavailable

PMID

1570821

Abstract

The authors examine an approach for identifying which criteria to use when screening for persons with a relatively rare medical condition. They considered various subsets of an inclusive set of criteria by examining the statistical effectiveness per unit cost of deleting various criteria using three statistical measures of effectiveness. An application to injury surveillance, where categories of the patient's chief complaint were used for screening via emergency department logs, illustrates two implications of the approach. First, deletion priorities may differ somewhat by effectiveness measure. When using sensitivity to measure statistical efficiency, the tendency is first to drop the larger categories with the smaller proportions of injury. On the other hand, using specificity to gauge effectiveness calls for large categories with the highest injury rates to be deleted first; while for a measure of effectiveness derived from relative bias, the tendency is first to drop those categories with the fewest injuries. Second, an unequivocal set of criteria may not emerge, thus forcing one to decide among options. The authors suggest several rationales for making this decision.

Language: en