|
Previous Top Next |
|
|
| Abstract
|
|
|
| Precision and Accuracy are the two primary measures of the performance of a quantitative analytical method. Developing a successful analytical method generally involves varying assay conditions (reagent concentrations, reaction times, measurement conditions, etc.); searching for the sets of conditions where the precision is adequate (optimal) over a suitable calibration range; and verifying that accuracy is maintained, even for "problematic" samples.
|
|
|
| Applying DOE techniques to immunoassay development has been hampered by the difficulty in obtaining sufficient data for reliable estimates of precision, particularly since precision is a complex function of the concentration of the analyte in the sample. Alternatively, experimenters have tried optimizing on surrogate measures of performance, such as the separation in response between calibrators, with limited success.
|
|
|
| A new approach is presented here, which efficiently utilizes data from calibrators, controls, (and unknowns, if present) to characterize a test method suitably for optimization. The precision profile is calculated, based on the pooled imprecision of the response replicates, and the slope of the calibration-response curve throughout the calibration range. This calculation provides the %CV (coefficient of variation, also called relative standard deviation) as a function of analyte concentration. Also provided are the upper and lower limits of quantification, based on the user-selected threshold for quantification. Additionally, accuracy can be directly estimated by including control samples with known target concentrations in the run. These direct measures of precision, accuracy, and range of quantification can be used for rapid assay characterization and optimization.
|
|
|
| The power of the technique comes from pooling the variance in response units from all replicate groups in the run. This is based on the observation that, in many assay systems, the assumption of uniform variance in response units is valid. The software includes tests for the validity of the uniform variance assumption, and makes response transformation available should the assumption fail. Finally, the software provides confidence interval calculations for all reported values, thereby reducing the tendency for over-interpretation of the data.
|