On the stability of log-rank test under labeling errors

Area of Science:

• Biostatistics
• Survival Analysis
• Computational Biology

Background:

• The log-rank test is a standard statistical method for comparing survival distributions between groups.
• Accurate sample labeling is a critical assumption for the validity of log-rank test results.
• Potential errors in sample labeling can compromise the reliability of survival analysis outcomes.

Purpose of the Study:

• To investigate the impact of sample labeling errors on log-rank test outcomes.
• To develop a method for quantifying the uncertainty introduced by labeling errors in survival analysis.
• To provide a practical tool for assessing the robustness of log-rank test results to labeling inaccuracies.

Main Methods:

• Introduced and defined a novel concept of uncertainty arising from labeling errors in the log-rank test.
• Developed a new algorithm for efficiently calculating a stability interval around the log-rank P-value.
• Proved the mathematical correctness of the developed algorithm.

Main Results:

• Demonstrated the algorithm's effectiveness on multiple real-world datasets.
• Quantified the potential impact of labeling errors on the statistical significance of survival differences.
• Established a method to assess the stability of log-rank test results.

Conclusions:

• Labeling errors introduce uncertainty into log-rank test results.
• The developed stability interval provides a robust measure of P-value reliability.
• The LoRSI tool facilitates the assessment of survival analysis robustness in the presence of labeling noise.