Power estimation for multireader ROC methods an updated and unified approach

Area of Science:

• Medical Imaging Analysis
• Statistical Methodology in Medical Research

Background:

• Multireader receiver operating characteristic (ROC) studies are essential for evaluating diagnostic test performance.
• Accurate power and sample size estimation are critical for the design and validity of these studies.
• Existing methods for power and sample size calculation in ROC studies require updates to incorporate recent statistical developments.

Purpose of the Study:

• To describe a step-by-step procedure for estimating power and sample size for planned multireader ROC studies.
• To unify power estimation approaches for the Dorfman-Berbaum-Metz (DBM) and Obuchowski-Rockette (OR) analysis methods.
• To incorporate recent methodological advancements and allow flexibility in input parameters (conjectured values or pilot study outputs).

Main Methods:

• A detailed step-by-step procedure for power computations is presented with its theoretical underpinnings.
• Updates include using recommended denominator degrees of freedom and accounting for varying normal-to-abnormal case ratios between pilot and planned studies.
• A novel method for calculating the OR test-by-reader variance component is introduced.

Main Results:

• The procedure was illustrated using a real dataset, demonstrating power computation for planned studies with identical and different case ratios compared to the pilot study.
• A simulation study confirmed that the proposed procedure yields mean power estimates closely approximating the true power.
• The updated procedure is straightforward to apply.

Conclusions:

• The updated procedure provides a reliable method for power and sample size estimation in multireader ROC studies.
• Ensuring comparability of pilot data (modalities, reader expertise, case selection) to the planned study is crucial for accurate estimates.
• Further investigation into the variability of power estimates is warranted.