FDA Expectations for N=2 Studies
When n=2, the FDA does not accept this as adequate evidence for drug approval and requires substantially larger sample sizes to demonstrate safety and efficacy, typically requiring hundreds to thousands of patients depending on the endpoint and disease context. 1
Why N=2 is Insufficient
The FDA's regulatory framework fundamentally requires robust statistical evidence that cannot be generated from only 2 patients:
- Statistical power requirements: For noninferiority trials, approximately 700 primary events are needed to provide convincing evidence, requiring trials of many thousands of patients 1
- Safety assessment: The FDA requires adequate sample sizes to detect potential harm, as demonstrated by recent examples where early efficacy endpoints failed to predict overall survival detriments 1
- Regulatory approval standards: Even for rare diseases with high unmet need, the FDA expects sufficient data to establish benefit-risk profiles 1
Minimum Sample Size Contexts
The evidence shows what the FDA actually accepts:
- Full approvals: Typically require overall response rates >70% with median duration of response of at least 9 months, still requiring substantially more than 2 patients 1
- Accelerated approvals: Even with lower ORR (13-70%), these still involve cohorts large enough to demonstrate statistical significance 1
- Rare disease trials: While the FDA acknowledges challenges in rare diseases, they still require adequate sample sizes to assess safety and efficacy 1
Special Circumstances
The only context where extremely small sample sizes might be considered:
- N-of-1 trials: These involve repeated measurements within individual patients across multiple treatment periods, not simply 2 patients total 1
- Case series for contextualization: Real-world evidence with very small samples may provide supportive context but cannot serve as primary evidence for approval 1
Critical Pitfall
The FDA explicitly criticizes inadequate sample sizes and selection bias in submissions, noting that limited data are "subject to selection bias and other problems inherent in the use of an external historical control" 1. With n=2, these concerns would be insurmountable for any regulatory decision.