Evaluating multiple surrogate markers with censored data.

Area of Science:

• Biostatistics
• Clinical Trials Methodology
• Epidemiology

Background:

• Surrogate markers can shorten randomized trials but evaluating multiple markers is challenging.
• High-dimensional surrogate markers pose difficulties for traditional nonparametric methods due to the curse of dimensionality.
• Existing methods for evaluating single surrogate markers may not scale to complex, multi-marker scenarios.

Purpose of the Study:

• To define a quantity for assessing the utility of multiple surrogate markers in time-to-event outcomes.
• To propose a robust estimation approach for evaluating multiple surrogate markers with censored data.
• To develop and validate novel statistical methods for surrogate marker assessment in clinical research.

Main Methods:

• Developed a dimension reduction procedure for multiple surrogate markers measured at a landmark time.
• Proposed a robust estimation approach incorporating weights to mitigate model misspecification.
• Introduced three novel estimators, two of which demonstrate double robustness properties.
• Utilized simulation studies to assess finite sample performance under various conditions.

Main Results:

• The proposed dimension reduction and robust estimation methods provide a feasible approach for evaluating multiple surrogate markers.
• The developed estimators show good performance in simulation studies across different scenarios.
• The methods are illustrated effectively using data from the Diabetes Prevention Program (DPP).

Conclusions:

• The novel approach offers a valuable tool for assessing the utility of multiple surrogate markers in time-to-event settings.
• This methodology can enhance the efficiency and reduce the cost of clinical trials.
• The application to DPP data demonstrates the practical utility of the proposed estimators for diabetes research.