Adaptive sample size methods

Adaptive Designs - Plan B for Trials

• Allows for pre-planned modifications to a trial based on interim data analysis, offering flexibility over fixed designs.
• Primary Goals: Increase trial efficiency, reduce costs, and improve ethical outcomes by stopping futile trials early.
• Common Adaptations:
  • Sample size re-estimation (SSR).
  • Stopping early for futility or overwhelming efficacy.
  • Dropping or adding treatment arms.
• ⚠️ Requires complex statistical adjustments (e.g., alpha-spending functions) to prevent inflating Type I error.

⭐ The most frequent adaptation is Sample Size Re-estimation (SSR). It uses interim data to recalculate the required sample size, ensuring the study remains adequately powered.

Sample Size Re-estimation - Sizing It Up

• An adaptive trial design method allowing for sample size adjustment during the study.
• Performed at a pre-planned interim analysis based on accumulating data.
• Primary Goal: To correct initial, potentially inaccurate, assumptions about effect size or data variability, ensuring adequate statistical power.

• Blinded Re-estimation: Adjusts sample size based on variance without unblinding treatment assignments. Lower risk of bias.
• Unblinded Re-estimation: Uses interim treatment effects to adjust size. More powerful but carries a higher risk of operational bias.

⭐ Unblinded sample size re-estimation can inflate the Type I error rate (false positives) if not controlled by specific statistical methods (e.g., promising zone design).

Group Sequential Designs - Peek-a-Boo Trials

• Concept: Allows for planned interim analyses of accumulating data in a clinical trial. The trial can be stopped early if results are compelling for efficacy, futility, or safety.
• Mechanism:
  • Data is analyzed at pre-specified time points or after a certain number of events.
  • Formal stopping rules (boundaries) are used to determine if the trial should continue.
• Challenge & Solution:
  • ⚠️ Repeatedly "peeking" at the data inflates the overall Type I error rate (family-wise error rate).
  • To maintain the nominal overall $\alpha$ (e.g., 0.05), the significance level for each interim look must be adjusted downwards.
• Common Boundary Methods:
  • Pocock: Uses the same significance boundary for each interim analysis.
  • O'Brien-Fleming: Uses conservative boundaries early on, making it hard to stop, and less conservative ones later.

⭐ High-Yield: The O'Brien-Fleming method is often preferred as it maintains statistical power better than the Pocock method by being very conservative at the initial interim looks.

• Adaptive designs allow for pre-planned sample size recalculation mid-trial based on interim data.
• This ensures the study remains adequately powered if initial effect size or variability estimates were inaccurate.
• The primary goal is to improve trial flexibility and resource efficiency.
• All adaptation rules must be prospectively defined in the study protocol to prevent bias.
• Helps avoid underpowered studies (↑ Type II error) or unnecessarily exposing patients in overpowered studies.
• Maintains statistical validity and blinding when executed properly.