A carcinoembryonic antigen standardization experiment for the conduct of multi-institutional clinical trials....

The Gastrointestinal Tumor Study Group conducted a carcinoembryonic antigen standardization experiment to address the issue of reproducibility of assays performed at participating institutions and a central laboratory. Samples were drawn in quadruplicate from 37 patients, representing five participating institutions. Systematic biases were observed in four of the five institutions: two series were consistently higher and another two series were consistently lower than the corresponding central laboratory determinations. Significantly higher variations were found in three of the five series, comparing the sample variances associated with the matching assay series. Overall, the standard deviation associated with the central laboratory was one third that of the participating institutions. These results point toward the limitation of pooling multi-institutional carcinoembryonic antigen data. The use of multivariate models with parameters for institution comparison is proposed as a solution to this problem.