Variance functions, detectability and bias: a re-evaluation.

Area of Science:

• Clinical Chemistry
• Biostatistics
• Analytical Chemistry

Background:

• Previous simulations using internal quality control data suggested small negative biases in limit of blank and limit of detection estimates.
• These initial estimates exhibited high uncertainty due to insufficient simulation size.

Purpose of the Study:

• To re-evaluate the accuracy of variance functions for estimating the limit of blank and limit of detection.
• To investigate the impact of simulation size and data quantity on bias and uncertainty.

Main Methods:

• Simulations were repeated 25 times with varied random number generator seeds.
• Data quantities were increased 100-fold, and replicates per specimen were reduced (40 down to 2).
• Variance functions were used to estimate the limit of blank and limit of detection.

Main Results:

• Previously observed negative biases were identified as artifacts of inadequate simulation size.
• With 20 or more replicates, biases were minimal (<0.1%).
• Lower replication (e.g., duplicates) introduced positive biases, particularly with large variances (up to +1.23%).

Conclusions:

• Variance functions yield essentially unbiased estimates for the limit of blank and limit of detection when using 20 or more replicates.
• Minimal biases are observed at lower replication levels if measurement errors are small.
• The study confirms the reliability of variance functions for these critical analytical parameters with sufficient data.