Sample size calculation for mixture cure model with restricted mean survival time as a primary endpoint

Area of Science:

• Biostatistics
• Clinical Trial Design
• Survival Analysis

Background:

• Many clinical trials, particularly in oncology, observe long-term survivors or 'cured' patients, necessitating cure fraction considerations.
• Traditional sample size calculations for mixture cure models often rely on the proportional hazards assumption, which is frequently violated in practice.
• Existing methods using the log-rank test may lead to inaccurate sample size estimations when proportional hazards do not hold.

Purpose of the Study:

• To develop a sample size calculation formula for mixture cure models using restricted mean survival time (RMST) as the primary endpoint.
• To evaluate the accuracy and efficiency of the proposed RMST-based sample size formula through simulations and real-world examples.
• To provide a more robust method for sample size determination in clinical trials with a cure fraction, especially when proportional hazards assumptions are not met.

Main Methods:

• Derived a novel sample size formula for mixture cure models based on restricted mean survival time (RMST).
• Conducted simulation studies to compare the performance of the RMST-based sample size calculation with the traditional log-rank test-based method.
• Applied the developed formula to an endometrial cancer trial example.

Main Results:

• The RMST-based sample size calculations for mixture cure models were found to be accurate, irrespective of whether proportional hazards assumptions were met.
• In scenarios where proportional hazards assumptions were violated, the RMST method generally required smaller sample sizes compared to the log-rank test.
• RMST provides a clinically interpretable measure of treatment effect (e.g., difference in survival years), enhancing patient-physician communication.

Conclusions:

• The proposed RMST-based sample size calculation is a more reliable approach for mixture cure models, particularly when proportional hazards are violated.
• This method offers accurate and often more efficient sample size estimations, leading to better-designed clinical trials.
• RMST provides a practical and easily communicable measure of treatment benefit, improving the clinical utility of trial findings.