Statistical power in the detection of matrix effects

Area of Science:

• Clinical Laboratory Science
• Biostatistics
• Analytical Chemistry

Background:

• Matrix-induced bias can negatively impact clinical laboratory performance evaluations on proficiency testing.
• Accurate detection of matrix effects is vital to prevent laboratories from being unfairly penalized.

Purpose of the Study:

• To evaluate an experimental protocol developed by the College of American Pathologists for detecting matrix effects in proficiency testing.
• To define and analyze the statistical power of this protocol in identifying matrix-biased systems.

Main Methods:

• Generated power curves to illustrate the impact of five key parameters on detecting matrix effects.
• Investigated the influence of bias, lack of fit (natural variation), pure error (random variation), number of patient samples, and replicates per sample.
• Assessed the effect of the statistical test's significance level on detection probability.

Main Results:

• Statistical power is most influenced by lack of fit and pure error components of variance, with lack of fit being more influential.
• Increasing fresh patient samples from 10 to 20 significantly increases power; moderate increases occur up to 40 samples.
• Increasing replicates per sample from 1 to 5 markedly increases power; 10 replicates are needed for substantial further increases.

Conclusions:

• The developed protocol aids in distinguishing matrix effects from true performance issues.
• Lack of fit and pure error are critical factors affecting the detection of matrix bias.
• Optimizing the number of patient samples and replicates is essential for enhancing the power to detect matrix effects in proficiency testing.