Time-dependent ROC curve analysis in medical research: current methods and applications

Area of Science:

• Biostatistics
• Medical Informatics
• Clinical Epidemiology

Background:

• Classical receiver operating characteristic (ROC) curve analysis assumes fixed disease status and marker values over time.
• In reality, disease onset and marker values fluctuate, necessitating time-dependent analyses.
• Current research often overlooks time dependency, using standard ROC methods inappropriately.

Purpose of the Study:

• To comprehensively review methodologies for time-dependent ROC curves.
• To provide clarity on methods using single or longitudinal marker measurements.
• To identify software tools and illustrate practical applications of time-dependent ROC analysis.

Main Methods:

• Systematic review of existing time-dependent ROC curve methodologies.
• Extension of methods to incorporate longitudinal marker data.
• Illustration of methodologies using a real-world clinical dataset (primary biliary cirrhosis).

Main Results:

• Identified 18 estimation methods for time-dependent ROC analysis with censored event times.
• Found three additional methods applicable to non-censored event times.
• Observed a lack of widespread application of these advanced methods in clinical studies despite their availability.

Conclusions:

• Re-established the value and importance of time-dependent ROC curve methods.
• Demonstrated practical application using available software.
• Provided recommendations for future research in time-dependent diagnostic marker evaluation.