420 – Biomarker Evaluation
New Method of Evaluation of Limit of Detection in Molecular Diagnostics
Jeffrey E. Vaks
Roche Molecular Diagnostics, Pleasanton, California
Current methods of estimation of LoD at 95% detection rate, described in CLSI EP17-A2 guideline for qualitative measurement procedures, are based on empirical modeling of the relationship between the analyte concentration and the probability of detection using probit analysis. The confidence interval for the LoD estimated using empirical models is often wide, and in not very rare case of lack of fit of the model to data it is very wide or does not exist making the LoD estimates unreliable and leading to new or additional data collection. The new method of LoD evaluation in molecular diagnostics is theoretical math model maximum likelihood estimation based. The math model was extensively validated with data. This method is robust to outliers and always provides reasonably tight confidence intervals for the LoD with modest amount of data and number of concentration levels tested. Also, unlike the empirical model based probit analysis, the new method allows for estimation of the minimum number of copies of target nucleic acid required for detection along with LoD. The method also provides useful information on concentrations corresponding to various detection rates, e.g., 5% and 50%. Also, consideration was given to (i) planning and conducting LoD studies more efficiently with sequential determination of the concentrations to be tested, (ii) increased LoD uncertainty in case of using more than one DNA/RNA region for detection, and (iii) in case the PCR assay is for a microorganism with large number of copies of DNA/RNA, the LoD is 3 microorganisms per sample volume, on average.