3.3.2. Reliability of the result

Some regression methods calculate a 95% confidence interval of the result by means of a variance analysis. This is complicated with logit regression.

With the measurements of the replicates from the calibration line, a confidence interval can be calculated. (See figure L) The calculated reliability is influenced by the shape of the calibration line. The closer the highest measuring point is to the calculated Bmax, the better the sigmoidal can be "predicted". Because, certainly with straighter calibration lines, Bmax far exceeds the highest response, and reliable confidence limits in the measuring range become unrealistically wide. (See figure R).

The calculated reliability does not provide a statement about the result's reliability, which is often determined from the average of multiple dilutions measured in replicate.

The confidence intervals are thus highly variable and differ with each test. Calculating reliability through variance analysis, based on the calibration line and multiple sample measurements, is quite intricate and not universally applicable without an expert evaluation.

With the calculation method used in this program, the calculated average is tested against retrospectively calculated reliability characteristics of the test performed.

The calculated result is tested by comparing the coefficient of variation of the individually calculated values of a sample with the expected coefficient of variation. This is set at 15% for ELISAs.

It must be realized that the reliability of this test increases with the number of measurements per sample.

The test should therefore be developed either with more replicates per dilution or work with smaller dilution steps to preferably produce a least five valid results.

This method assumes that the test is homoscedastic. That is, the confidence interval or reliability of the results is the same (in the same order of magnitude) over the entire measurement range.

A well-designed EIA or RIA meets this requirement, and homoscedasticity is tested in a validation study.