Prioritized concordance index for hierarchical survival outcomes

Area of Science:

• Biostatistics
• Epidemiology
• Clinical Research Methodology

Background:

• Evaluating prognostic biomarkers for complex diseases with multiple outcomes is challenging.
• Existing methods may not optimally leverage information from prioritized or multiple disease endpoints.

Purpose of the Study:

• To propose and validate a novel prioritized concordance index for assessing biomarker prognostic utility in diseases with multiple, prioritized outcomes.
• To enhance the efficiency and power of prognostic variable identification compared to single-outcome indices.

Main Methods:

• Extension of Harrell's concordance (C) index to a "prioritized concordance index" for prioritized outcomes.
• Utilizing generalized pairwise comparisons based on the most severe outcome, similar to the win ratio.
• Employing inverse probability weighting for censoring correction and U-statistic properties for asymptotic analysis.

Main Results:

• Simulation studies demonstrate increased efficiency and power when a predictor is associated with both primary and secondary outcomes.
• The prioritized concordance index effectively identifies prognostic variables compared to using only the primary outcome.
• Application to type II diabetes risk and lung cancer incidence/mortality models showcases its utility.

Conclusions:

• The prioritized concordance index offers a robust method for evaluating prognostic biomarkers in diseases with multiple, prioritized outcomes.
• This approach enhances the ability to detect true prognostic signals, particularly when biomarkers influence various disease aspects.
• The index provides valuable insights into risk prediction for complex diseases like diabetes and cancer.