Computerized selection of donors for inclusion in red cell antibody panels

This study evaluated various algorithms for the generation of a red cell antibody identification panel by computer. The algorithms depended on schemes that optimized Fisher's exact p value for each specificity in the panel rather than using predetermined formulations of desired phenotypic combinations. Panels produced by these computer-driven algorithms offered the distinction of a greater number of antibody specificities than manually created panels. They were also evaluated by a group of technologists to be more useful for general serologic work than panels created manually. Although the panels were created from a database of only 252 donors, an objective numeric evaluation showed that the panels were more useful than commercial panels. The computer approach used in this study may offer means of efficiently creating panels for general use as well as creating panels for resolution of particular serologic problems. They may also facilitate and encourage the construction of panels from small donor databases.