169 / 274
5
INCREMENTAL COLLABORATIVE STUDY
The results of a traditional collaborative study are typically reported separately for each
concentration level measured for each matrix. Repeatability, reproducibility, recovery and
comparative results frequently are different for different matrices; and repeatability,
reproducibility and recovery are typically concentration dependent (cf. ‘HORRAT’ index).
DESIGN ELEMENTS COMMON TO ALL SCHEMES FOLLOWING
All of the proposed versions of incremental collaborative studies will have the following design
elements:
1. Fixed number of replicates. (2 are suggested)
2. Repeatability conditions for replicates (same equipment and reagents, same technician, same
point of time).
3. Specified and constant method protocol across all measurements and all collaborators
(reproducibility conditions).
4. Controls to maintain study integrity.
5. Specified reporting formal for results.
6. Randomization and masking wherever possible and desirable (replications, order of testing
concentrations).
INCREMENTAL BY MATRIX
The first major line of demarcation for splitting a collaborative study into modules is at the
matrix level. For example, if the plan is to validate a test method for three different matrices, then
three different increments of the collaborative study might be performed, one for each matrix
involved. Generally, this will involve studies that are still fairly expensive, given the multiple
concentration levels and replication involved. The order of the matrices studied may be arranged
in declining order of importance so that early termination of the study yields maximum value at
minimum cost. If the confounding of time sequence with matrix is unacceptable, the order of the
matrices may be randomized.
Different collaborators may be used for each increment
, which
will greatly improve ease of enrollment.
Current thinking proposes study of various matrices at the single laboratory level, with a
subsequent single worst-case matrix chosen for the collaborative study. Note, however, that this
does not allow measurement of reproducibility, and should only be considered when the number
of replicates used provides a statistical power to test method equivalency or performance
requirements at the necessary level (and no less than that provided from a collaborative study). If
reproducibility varies with matrix, as it frequently does, this should be taken into account in
selecting the worst-case matrix. Also note that testing only a single worst-case matrix in a
collaborative study will characterize the candidate method by its worst-case reproducibility.
116