PET/CT in cancer: Methodological considerations for comparative diagnostic phase II studies with paired binary data

Area of Science:

• Medical Imaging
• Diagnostic Accuracy
• Clinical Trials

Background:

• Positron Emission Tomography/Computed Tomography (PET/CT) shows promise for cancer diagnosis and staging.
• Evaluating new diagnostic tools requires systematic clinical studies comparing them to existing standards.
• Determining whether to design a superiority or non-inferiority study can be challenging.

Purpose of the Study:

• To propose a flexible statistical analysis strategy for evaluating diagnostic imaging techniques like PET/CT.
• To provide a method adaptable for both superiority and non-inferiority study designs.
• To focus on confidence intervals for differences in accuracy measures.

Main Methods:

• Utilizes paired observations from diagnostic studies where patients undergo both PET/CT and the standard procedure.
• Focuses on point estimates and confidence intervals for differences in accuracy measures (sensitivity, specificity, predictive values).
• The strategy accommodates scenarios with incomplete gold standard data.

Main Results:

• Provides formulas for approximate 95% confidence intervals for differences in key diagnostic accuracy metrics.
• Demonstrates applicability even when gold standard results are unavailable for all patients.
• Recommends adaptive sample size determination.

Conclusions:

• Diagnostic studies for PET/CT merit should commence with phase II studies.
• Focusing on confidence intervals for diagnostic measure differences is recommended.
• The proposed statistical strategy remains applicable even with incomplete gold standard procedures.