Common analysis pitfalls in longitudinal ctDNA studies: lead time, sensitivity and immortal time bias

Area of Science:

• Oncology
• Biomarker Research
• Statistical Methodology

Background:

• Circulating tumour DNA (ctDNA) is a promising biomarker in oncology for various clinical applications.
• Longitudinal ctDNA measurements are increasingly used for disease monitoring and treatment guidance.
• Methodological and statistical challenges exist in analyzing longitudinal ctDNA data.

Purpose of the Study:

• To critically examine recurring methodological and statistical pitfalls in longitudinal ctDNA studies.
• To raise awareness of time-dependency in performance metrics (sensitivity, specificity) and bias in ctDNA analysis.
• To provide guidance for the design, analysis, and reporting of ctDNA trials.

Main Methods:

• Review and critical examination of methodological and statistical issues in longitudinal ctDNA studies.
• Demonstration of time-dependency in performance metrics using examples.
• Explanation and illustration of immortal time bias in minimal residual disease (MRD) studies.
• Use of toy data, simulations, and re-analysis of published data.

Main Results:

• Performance metrics like sensitivity and specificity are inherently time-dependent in ctDNA studies.
• Immortal time bias can arise from incorrect classification of patients in MRD surveillance.
• Lead time definition is a critical factor in longitudinal ctDNA analysis.

Conclusions:

• Awareness of methodological and statistical pitfalls is crucial for accurate longitudinal ctDNA analysis.
• Improved study design, analysis, and reporting are needed to overcome identified challenges.
• Further evidence is required to establish definitive solutions for longitudinal ctDNA data interpretation.