The consequences of proportional hazards based model selection

Area of Science:

• Biostatistics
• Clinical Trials
• Survival Analysis

Background:

• The Cox proportional hazards (PH) model is standard for survival data analysis in clinical trials.
• Validating the PH assumption is crucial before model application.
• Failure of the PH assumption necessitates alternative methods, raising concerns about sequential testing bias.

Purpose of the Study:

• To investigate the impact of sequential model fitting on clinical trial efficacy testing.
• To evaluate the performance of common correction methods when the PH assumption is violated.
• To compare the power of permutation adjustment and a two-stage testing strategy.

Main Methods:

• A simulation study was employed to assess statistical power and bias.
• The study examined a simple resampling technique: permutation adjustment.
• A recently proposed two-stage testing strategy was also analyzed.

Main Results:

• The permutation adjustment method demonstrated significant power limitations.
• Sequential model fitting procedures can introduce notable bias.
• The two-stage testing strategy was explored for its potential to mitigate these issues.

Conclusions:

• Sequential testing in clinical trials with survival data requires careful consideration of potential bias.
• The permutation adjustment technique is suboptimal in terms of statistical power.
• Further investigation into alternative strategies, such as the two-stage approach, is warranted for robust survival data analysis.