Sensitivity analysis of limit of detection estimation using probit regression

Our objective was to explore how variations in the number and distribution of tested concentrations around the presumed lower limit of detection (LLOD) affect the estimate and its confidence interval.

We reviewed published LLOD evaluations and conducted sensitivity analyses to assess how limiting the number and distribution of input concentrations affects LLOD estimates. We systematically reduced the number of tested concentrations and varied their distribution, either centered or top-weighted around the LLOD, to evaluate impacts on the estimate and confidence interval.

When data sets are restricted but remain centered, the estimated LLOD lowers. When data sets are restricted to top-weighted concentrations, the estimated LLOD lowers, and the confidence intervals widen considerably. In all cases of data restriction, model fit, as measured by the Akaike information criterion, suffers, the effects of which are most severe in the top-weighted scenarios.

These findings reinforce recommendations from the Clinical and Laboratory Standards Institute and highlight the need for caution when constrained testing designs are used in LLOD estimation.