Sensitivity analysis of limit of detection estimation using probit regression.

Area of Science:

• Clinical Chemistry
• Analytical Chemistry
• Biostatistics

Background:

• Accurate determination of the lower limit of detection (LLOD) is crucial for reliable analytical measurements.
• Existing guidelines recommend specific testing designs for LLOD evaluation.
• The impact of variations in concentration number and distribution on LLOD estimates requires further investigation.

Purpose of the Study:

• To investigate how the number and distribution of tested concentrations around the presumed LLOD influence the LLOD estimate and its confidence interval.
• To assess the sensitivity of LLOD estimates to restricted testing designs.

Main Methods:

• Review of published lower limit of detection (LLOD) evaluations.
• Sensitivity analyses involving systematic reduction in the number of tested concentrations.
• Evaluation of different concentration distributions (centered vs. top-weighted) around the LLOD.

Main Results:

• Restricting data sets while maintaining centered concentrations led to lower LLOD estimates.
• Top-weighted concentration distributions resulted in lower LLOD estimates and significantly wider confidence intervals.
• Model fit, assessed by Akaike information criterion, deteriorated across all data restriction scenarios, most severely with top-weighted distributions.

Conclusions:

• Findings support existing recommendations from the Clinical and Laboratory Standards Institute regarding LLOD estimation.
• Caution is advised when employing constrained testing designs for LLOD determination due to potential biases and reduced precision.
• The study highlights the importance of comprehensive concentration ranges in LLOD evaluations for robust analytical performance assessment.