30% Sample Size Reduction with CUPED
HER2+ metastatic breast cancer single-arm Phase II evaluating objective response rate (ORR)
$3.6M
Cost Savings
30%
Sample Reduction
3.6 mo
Faster Phase III Start
Scenario
A biotech company was designing a single-arm Phase II trial for a novel HER2-targeted therapy in metastatic breast cancer. The primary endpoint was objective response rate (ORR) assessed by RECIST 1.1.
Standard design required 240 patients to detect a clinically meaningful response rate with 80% power. Per-patient cost was estimated at $50,000, for a total trial cost of $12.0 million.
The CUPED Opportunity
Baseline Tumor Burden as Covariate
Published literature shows that baseline tumor burden (sum of longest diameters of target lesions) is correlated with treatment response in HER2+ breast cancer, with correlation coefficient ρ = 0.55.
Using the CUPED variance reduction formula:
This means the adjusted variance is 70% of the unadjusted variance, enabling a 30% sample size reduction.
Design Comparison
| Metric | Standard Design | CUPED-Adjusted |
|---|---|---|
| Sample Size | 240 patients | 168 patients |
| Duration | 14 months | 10.4 months |
| Trial Cost | $12.0M | $8.4M |
| Statistical Power | 80% | 80% |
Additional Benefits
Phase III start 3.6 months earlier
Faster enrollment completion accelerates subsequent development
BLA submission 6-9 months earlier
Earlier market entry compounds throughout development timeline
$150-300M additional revenue
Market entry acceleration for a successful therapy
Methodology
CUPED (Controlled-experiment Using Pre-Experiment Data) was originally developed at Microsoft for A/B testing but has direct application to clinical trials. The FDA's May 2023 guidance on covariate adjustment explicitly encourages this approach, calling it “low-hanging fruit” for improving trial efficiency.
The variance reduction comes from the statistical principle that adjusting for a correlated baseline covariate reduces residual variance. The adjusted treatment effect estimate has lower variance by a factor of (1 - ρ²), where ρ is the correlation between baseline and outcome.
This reduction in variance translates directly to reduced sample size requirements while maintaining the same statistical power.
Regulatory Support
“Sponsors are encouraged to adjust for prespecified covariates that are expected to be prognostic for the primary outcome.”
— FDA Guidance: Adjusting for Covariates in Randomized Clinical Trials (May 2023)
Calculate your CUPED savings
Enter your baseline-outcome correlation to see potential sample size reduction.