Sample size for non-inferiority trials

Non-Inferiority Trials - Not Worse, Just Different

• Goal: To show a new treatment is not unacceptably worse than the standard. Used when new options offer other benefits (e.g., ↑safety, ↓cost).
• Non-Inferiority Margin (Δ): The pre-specified, largest clinically acceptable difference to still be considered "good enough."
• Hypotheses:
  • H₀ (Null): The new treatment is inferior (Difference > Δ).
  • H₁ (Alternative): The new treatment is non-inferior (Difference ≤ Δ).
• Sample Size: Influenced by α, β (power), variance, and Δ. A smaller, stricter margin (↓Δ) requires a ↑ sample size.

⭐ For non-inferiority to be claimed, the entire confidence interval for the treatment effect difference must be less than the non-inferiority margin (Δ).

NI Sample Size - The Secret Sauce

• Goal: Prove a new treatment is not unacceptably worse than the standard. The sample size hinges on the non-inferiority margin (δ).

• Core Formula (per group):

  • $n = \frac{(Z_{\alpha} + Z_{\beta})^2 \times (2\sigma^2)}{(\Delta - \delta)^2}$
  • $\Delta$: Assumed true difference between treatments.
  • $\delta$: The pre-defined non-inferiority margin.

• Key Relationship: The required sample size is highly sensitive to the gap between the true effect ($\Delta$) and the NI margin ($\delta$).

⭐ Exam Pearl: Counterintuitively, non-inferiority trials often require a larger sample size than superiority trials, especially if the new drug's efficacy is expected to be very similar to the standard (i.e., Δ is small).

The Formula - Cranking the Numbers

• Calculates subjects needed to prove a new treatment is not unacceptably worse than standard treatment.
• Formula for continuous outcomes (per group): $$ n = \frac{2 \sigma^2 (Z_{\alpha} + Z_{\beta})^2}{(\Delta - \delta)^2} $$
  • Key Inputs:
    • $Z_{\alpha}$: Significance level (e.g., 1.96 for α=0.025)
    • $Z_{\beta}$: Statistical power (e.g., 0.84 for 80% power)
    • $\sigma^2$: Data variability (variance)
    • $\delta$: The non-inferiority margin (critical value)
    • $\Delta$: Expected difference in effect (often assumed to be 0)
• Sample Size Drivers:
  • Sample size ↑ as power ↑, significance ↑ (α ↓), or variance ↑.
  • Crucially, sample size ↑ dramatically as the margin (δ) ↓ (becomes stricter).

⭐ The non-inferiority margin (δ) is the most critical choice. It must be smaller than the active control's established benefit over placebo, ensuring the new drug preserves a clinically meaningful effect.

Sample Size Levers - Dialing It In

• Non-Inferiority Margin (δ): The most critical lever.
  • Smaller (stricter) margin → ↑ sample size.
  • Larger (lenient) margin → ↓ sample size.
• Power (1-β):
  • Higher power (e.g., 90% vs 80%) → ↑ sample size. Reduces Type II error risk.
• Significance Level (α):
  • Lower α (e.g., 0.01) → ↑ sample size. Reduces Type I error risk.
• Outcome Variability (σ²):
  • Higher data variability → ↑ sample size for precise estimates.

⭐ The non-inferiority margin (δ) isn't arbitrary. It's set based on historical data of the active control's effect over a placebo, ensuring the new drug preserves a clinically meaningful effect.

High‑Yield Points - ⚡ Biggest Takeaways

• The goal is to show a new treatment is not unacceptably worse than the standard one.
• A pre-specified non-inferiority margin (δ) sets the boundary of acceptable difference.
• Success requires the entire confidence interval of the effect to be above -δ.
• Sample size is driven by the margin (δ), power (1-β), and significance (α).
• A smaller (stricter) margin demands a larger sample size to achieve adequate power.
• If the CI crosses -δ, the result is inconclusive, not a confirmation of inferiority.