A balanced hazard ratio for risk group evaluation from survival data

Area of Science:

• Biostatistics
• Clinical Trials
• Genomics

Background:

• Clinical studies often assess prognostic factors to stratify patients into risk groups for tailored therapies.
• Survival data analysis commonly uses hazard ratios to evaluate these risk groups and prognostic factors.
• Existing hazard ratio methods have limitations, including arbitrary group definitions and inconsistencies with statistical significance.

Purpose of the Study:

• To identify limitations in using the hazard ratio for assessing prognostic factors and risk groups in survival data analysis.
• To propose a novel metric, the balanced hazard ratio, to overcome these limitations.
• To demonstrate the utility of the balanced hazard ratio in defining risk groups and evaluating prognostic markers.

Main Methods:

• Critically analyzed the limitations of the traditional hazard ratio in assessing risk groups.
• Developed and introduced the balanced hazard ratio as a new performance metric.
• Validated the balanced hazard ratio using controlled experiments and real-world cancer prognosis data.

Main Results:

• The traditional hazard ratio can lead to arbitrarily unbalanced risk groups and inconsistencies between effect size and statistical significance.
• The balanced hazard ratio provides an intuitive and easily computable metric for evaluating prognostic factors.
• The balanced hazard ratio naturally facilitates the selection of cut-off points for defining risk groups from continuous risk scores.

Conclusions:

• The balanced hazard ratio offers a more robust and reliable method for assessing the quality of prognostic factors compared to the traditional hazard ratio.
• This new metric can improve the definition and comparison of patient risk groups in various clinical settings.
• The balanced hazard ratio has broad applicability for evaluating prognostic markers across different diseases and study designs.