1. Statistical considerations for the Nipah virus treatment study
Lori E Dodd, PhD
Clinical Trials Research Section
Division of Clinical Research/NIAID

2. Power for various sample sizes
Statistical analysis
Primary Endpoint: 28-day mortality Statistical test: test of lower mortality rate (using Boschloo’s test), two-sided type I error rate of 0.05 Sample size: 100 per arm
Highly unlikely that one outbreak will provide sufficient statistical power for conclusive evidence
Mortality rate
Power for various sample sizes
per arm
SOC
m102.4 50 75
100
80%
64%
40%
58%
71%
56%
72%
89%
96%
70%
49%
57%
75%
86%
50%
30%
53%
83%
20%
76%
87%

3. Draft statistical analysis plan:
Action item
Draft statistical analysis plan:
who and what?

4. Randomization Stratified for statistical efficiency
Limited to one stratification variable:
Presence of neurologic symptoms (yes/no).
Double-blinded design requires
Process to ensure masking of treatment assignment
Randomization via database with blinded treatment codes
Pharmacist standardized operating procedure to preserve the blind
Reporting of AEs to Data and Safety Monitoring Board who should be unblinded to treatment assignment

5. Action items Randomization SOP: list generation and database process
SOP to ensure double-blinding
SOP for reporting results to DSMB

6. Interim monitoring considerations
Early data are unreliable but ethical considerations demand
that a trial stops for
Definitive early efficacy or
Definitive early harm
Monitoring too frequently increases chances of a false positive finding
Study likely to continue across multiple outbreaks
Solution: group sequential monitoring

7. Definitive early efficacy
Truncated O’Brien-Fleming boundary with 5 interim looks:
Number total 24 50
100
150
200
One-sided p-value boundary
0.0005
0.009
0.021
Mortality boundaries for first (n=24) interim analysis:
Deaths in m102.4
Deaths in SOC 8+ 1 9+ 2
11+ 3 12
Flexibility to evaluate at the end of an outbreak

8. Definitive early harm
Same frequency as for efficacy. Less evidence required to establish harm. p<0.05 for each look
Mortality boundaries for first safety analysis (at n=24):
Deaths in m102.4
Deaths in SOC 4+ 6+ 1 7+ 2 9+ 3

9. Conditional Power
Method of calculating likelihood of concluding statistical significance given accumulated (but not complete) data Given the accumulated observed data (up to time of analysis), compute the probability of achieving statistical significance assuming the remaining data (up to total of 100/arm) follow assumed alternative: *e.g., 50% reduction in mortality Evaluated during interim analyses and/or accrual pauses (due to end of outbreak) —If conditional power < 20%, continuing study may be futile

10. Sample size reassessment
After 100 participants have been accrued, sample size may be increased if: Conditional power is greater than 50%, or Pooled mortality rate is low

11. Continuing study across outbreaks
Study continues (across outbreaks) without release of results until accrual completed or DSMB recommends early stopping. Procedures/agreements needed to ensure all sites/countries understand the importance of this. Early release of data without a definitive conclusion may make continued study of m102.4 impractical.

12. Prevail II example
Randomized controlled trial comparing ZMappTM to standard-of-care during the Ebola outbreak
Powered for a 50% reduction in mortality from 40% to 20%
-200 subjects needed
By January 2016, epidemic was clearly ending.
Should study continue?
West African Ebola outbreak was unprecedented.
No pattern of regional outbreaks
Unclear if and when another outbreak would occur

13. DSMB presented with the following data
The PREVAIL II Writing Group. N Engl J Med 2016;375:

14. Recommendation
DSMB recommended study closure and release of data Results published NEJM in 2016 ZMappTM was promising but results not definitive Ongoing debate about its efficacy Questions for thought: Should DSMB have kept study open? Suppose another outbreak was expected within a year, would DSMB have kept study going?

15. Action items DSMB charter DSMB approves protocol and monitoring plans
Country-/site-wide agreement with monitoring plans
Statistical analysis plan to describe interim analyses in detail