Comparison of quantitative diagnostic tests: type I error, power, and sample size

Area of Science:

• Medical Diagnostics
• Biostatistics
• Laboratory Medicine

Background:

• Quantitative laboratory tests commonly use the 0.975 fractile as a discrimination limit.
• Sensitivity is defined as the proportion of diseased subjects exceeding this limit.
• Ignoring sampling variation in discrimination limits can significantly inflate type I error.

Purpose of the Study:

• To investigate the impact of ignoring discrimination limit variability on sensitivity estimation.
• To identify correct statistical procedures for comparing diagnostic test sensitivity.
• To analyze statistical power and sample size requirements for quantitative diagnostic tests.

Main Methods:

• Statistical analysis of sensitivity estimates.
• Evaluation of type I error rates under different conditions.
• Power and sample size calculations for Gaussian and log-Gaussian distributions.

Main Results:

• Comparing sensitivities without accounting for discrimination limit variation increases type I error up to sevenfold.
• Correct statistical procedures mitigate this inflation.
• Power and sample size are influenced by distribution type (Gaussian vs. log-Gaussian).

Conclusions:

• Accurate statistical methods are crucial for reliable sensitivity evaluation in quantitative diagnostic tests.
• Awareness of discrimination limit variability is necessary to avoid erroneous conclusions.
• The findings aid in planning studies for quantitative diagnostic test evaluation.