1. Statistical Considerations for an Adaptive Design for a Serious Rare Disease
Gary R Cutter, PhD
On behalf of
Kaushik Patra, PhD, Bruce A.C. Cree,
Eliezer Katz, Erik Pulkstenis, Alex Dmitrienko,
and myself (authors of this paper)

2. Neuromyelitis optica (NMO) and Neuromylelitis Optica Spectrum Disorder (NMOSD)
Acute/subacute demyelination, necrosis of optic nerves, spinal cord
Primarily in females (9 to 1)
Often preceded by viral illness, associated with systemic autoimmune disease
Significant residua common
Partial responses to steroids, other immunosuppressants
The term NMOSD reflects a variety of disorders associated with anti-AQP4 antibodies thought to be related to NMO but do not quite fit the diagnostic criteria for definite NMO.
This includes brainstem disorder with clinical features including intractable hiccups and nausea, or vomiting

3. NMO and NMOSD
NMO/NMOSD prevalence is estimated to be between 0.5 and 4.4/100,000
To date, no randomized controlled clinical studies of NMO have been completed
No treatment has received regulatory approval

4. Diagnosis of NMOSD All 3 findings are required: Optic neuritis
Transverse myelitis
2 of 3 of the following supportive criteria:
Contiguous cord lesion involving more than 3 vertebral segments
Anti-aquaporin-4 IgG seropositivity
Brain MRI not meeting criteria for Multiple Sclerosis

5. Treating NMOSD
One area of controversy within the NMOSD community and among regulatory agencies is the appropriateness of a placebo-controlled design in this disease.
Although placebo control can provide unequivocal evidence of efficacy, there is a legitimate concern about not treating patients
Untreated subjects may be at higher risk of relapses (and their consequences)
Can be fatal

6. Treatments For NMOSD Immunosuppressive medications such as:
azathioprine, mycophenolate mofetil, mitoxantrone, prednisone, and rituximab are all used empirically to prevent attacks
the evidence to support their use is limited, but are widely viewed by clinicians as worthwhile
Either because of positive experience
Or because it allows them to do something

7. Design and Statistical Aspects Considered in an NMOSD Trial
1. Determination of the primary endpoint 2. Choice of comparator and design consideration 3. Stratification, randomization ratio, and sample size calculation 4. Interim analyses 5. Secondary endpoints and multiplicity adjustment 6. Ethical considerations 7. Assessment of primary endpoint, adjudication process and related analyses strategies
MediImmune Trial: A Double-masked, Placebo-controlled Study With Open Label
Period to Evaluate MEDI-551 in Neuromyelitis Optica and Neuromyelitis Optica
Spectrum Disorders NCT

8. Determination of the primary endpoint
Given that sometimes NMOSD relapses cause irreversible disability, the occurrence of a single relapse on treatment has been accepted by investigators and regulatory agencies to be a primary outcome.
Clinically relevant criteria for each of the typical NMO types of relapse were developed by expert consensus:
for optic neuritis (11 criteria)
myelitis (4 criteria)
brain stem (2 criteria), and
hemispheric events (1 criterion).
A central committee is responsible for adjudicating each relapse to be used for the primary endpoint.

9. Choice of comparator and design consideration
Could a placebo be used?
Great debate on this topic
A single relapse event with rescue was considered adequate to balance the FDA preference for a placebo, while respecting the clinicians unease
Minimize the exposure to placebo
Use 3 to 1 randomization

10. How Long on a Placebo is Tolerable?
The traditional time-to-event designs follow patients on their treatment assignment until a specified number of medically relevant events have accrued.
This strategy was not tenable for the current study because indefinite placebo exposure was considered to potentially be an excessive burden or risk for the study subjects.
Therefore the duration of follow up was truncated.

11. A time-to-event design (Figure 1) with each patient on the randomized controlled treatment for a maximum of 197 days (approximately 7 months) or until an NMOSD relapse occurred (whichever occurred first).

12. In order to protect the subjects from further worsening
as a consequence of NMOSD relapses, immediate rescue
treatment with high-dose steroids, plasmapheresis or intravenous immunoglobulin (interventions commonly used to treat NMOSD relapses) were incorporated into the study.

13. Stratification, Randomization Ratio, and Sample Size Calculation
Since not all NMOSD are AQP4+
Expected rate of AQP4+ of 4 to 1
seronegative patients may have lower relapse rates – but not assumed for overall analyses
stratification by antibody positivity was deemed appropriate
Sample size estimates were based on pooled non-randomized trials to
estimate the number of events needed
and the placebo hazard rate was calculated as the negative logarithm of the proportion of relapse-free subjects over the first 12 months between the first two relapses prior to treatment from historical data.

14. From Costanzi C, Matiello M,
Lucchinetti CF, et al. Azathioprine:
tolerability, efficacy, and predictors
of benefit in neuromyelitis optica.
Neurology. 2011;77:

15. Effect of Therapy
The weighted average of the hazard rates (previous Table 1) for rituximab and azathioprine treated subjects was calculated as 0.25 and 0.58 per year, respectively.
Rituximab at 0.25 is a B-cell treatment and MEDI-551 is also a B-cell therapy
From this evidence, a conservative estimate of HR 0.40 was hypothesized and an acceptable target efficacy of MEDI-551 against placebo.

16. Sample Size Determined by Simulation*
The study was powered for the ITT population
while protecting the power in seropositive cohort
*Simulation results from 5000 iterations.

17. Expected Events Calculated as:
and

18. No Planned Interim Efficacy Analyses
(1) the target treatment effect is optimistic; (2) the required number of events is relatively small; (3) enrichment of the safety database for the experimental drug will be impaired; (4) integrity of the adjudication process around the primary endpoint may not be fully evaluable, potentially impacting regulatory review; (5) assessment of secondary objectives will be impaired; and (6) by design, all subjects eventually will be treated with MEDI-551 without an early efficacy claim.

19. Masked Sample Size Re-estimation
The fundamental theme of this method lies around re-estimating the survival probability based on the observed Kaplan-Meier failure rate (both treatments combined) at an interim time point.
Using the re-estimation approach of Todd et.al * estimation based on the average difference was used to find the adjustment factor.
* Todd S, Valdes-Marquez E, West J. A practical comparison of blinded methods
for sample size reviews in survival data clinical trials. Pharm Stat. 2012;11: )

21. At the Interim SS Re-assessment
We compute Suppose φ = and 104 subjects have completed the trial (relapsed or censored), then working through the previous equations we obtain: Snew will be (note SD = 0.67) and subsequently Nnew becomes 252, an increase of 40 subjects. If the observed event to subject ratio is approximately 27%= (1-Snew)% when 104 subjects complete the trial portion, an increase in target sample size up to 252 subjects will be considered to preserve the statistical power at approximately 94%, assuming a 4:1 ratio is maintained in seropositive versus sero-negative cohorts.

22. Potential Changes in Sample Size

23. Futility Assessment
Futility is to be assessed based on conditional power expressed by:

24. While fewer relapses are required to establish efficacy with placebo as the control compared to an active comparator with known efficacy – above the fewer relapses are achieved by assuming a larger effect of the treatment, not unreasonable for comparison with a placebo.

25. Unequal Versus Equal Allocation
A two-sided test of whether the hazard ratio is one with an overall sample size of 350 subjects (of which 175 are in the control group and 175 are in the treatment group) achieves 90% power at a significance level when the hazard ratio is actually The number of events required to achieve this power is 88 (87.5)
Unequal Allocation
A two-sided test of whether the hazard ratio is one with an overall sample size of 468 subjects (of which 117 are in the control group and 351 are in the treatment group) achieves 90% power at a significance level when the hazard ratio is actually The number of events required to achieve this power is 117
Thus, from a societal perspective almost 33% more patients experience events than in the placebo controlled trial

26. Active Versus Placebo Comparator
Placebo Controlled with 50% reduction
A two-sided test of whether the hazard ratio is one with an overall sample size of 350 subjects (of which 175 are in the control group and 175 are in the treatment group) achieves 90% power at a significance level when the hazard ratio is actually The number of events required to achieve this power is 87.5.
A two-sided test of whether the hazard ratio is one with an overall sample size of 1322 subjects (of which 661 are in the control group and 661 are in the treatment group) achieves 90% power at a significance level when the hazard ratio is actually The number of events required to achieve this power is
Thus, from a societal perspective almost 4 times as many patients experience events than in the placebo controlled trial

27. Additional Topics in the Paper
We touch on multiplicity of testing given the stratification of seropositivity
argue for strong control on the assessments with the power placed on the larger subgroup (seropositive =80% of the population)
We discuss and provide examples of the impact of the adjudication committee, based mostly on the concept that noise biases toward the null.
This is more important in noninferiority trials where non-adjudicated results may bias away from the null, but also relevant in rare diseases where finding patients is difficult

28. Conclusions Rare serious diseases present two problems
Small available sample
Often reluctance to use a placebo
A series of underpowered studies is not the answer
While not optimal from a statistical perspective N to 1 matching may be preferred to 1 to 1 matching from the clinical perspective.

29. Conclusions Continued
Limiting exposure time can be an acceptable approach when the events occur reasonably rapidly to minimize exposure to placebos
With almost all rare diseases precision of the estimates for planning may be less than adequate
thus blinded sample size adjustments may be worthwhile in these instances
The feasibility with smaller effect sizes needs to be examined in the context of the disease under consideration

30. Thanks For Listening!