1. Sensitivity Analysis for Clinical Trials with Missing
Outcome Data Using Repeated Measures
A Simulation Study Design
H. Terry Liao, PhD
Boston Scientific Corporation
Ying Yang, PhD
FDA/CDRH
Presented to Joint Statistical Meetings
August 10th, 2015
Contact: terry.liao@bsci.com
1JSM 2015

2. Motivation
 Renal denervation
–Treating uncontrolled hypertension
 Minimizing bias through:
–Study design proposal
–Repeated measures
–Missing data
–Sensitivity analysis
2JSM 2015

3. Renal Denervation
JSM 2015 3
TCT 2013: SFO Oct 27-Nov 1, 2013 by Dr. Horst Sievert

4. Study Endpoints
 Efficacy
– Office-based systolic blood pressure (OSBP);
average of 3 readings
– 24-hour ambulatory systolic blood pressure
(ASBP); average 24-hour average SBP
– Endpoint = (∆ SBPTest - ∆ SBPControl)
 Safety
– Composite endpoint (MAE)
– All-cause death, renal failure, hospitalization
for hypertensive crisis, hospitalization due to
severe hypotension/syncope, etc.
JSM 2015 4

5. High Blood Pressure
Hypertension (HTN)
Blood Pressure
Category
Systolic
mm Hg (upper #)
Diastolic
mm Hg (lower #)
Normal less than 120 and less than 80
Prehypertension 120 – 139 or 80 – 89
High Blood Pressure
(Hypertension) Stage 1
140 – 159 or 90 – 99
High Blood Pressure
(Hypertension) Stage 2
160 or higher or 100 or higher
Hypertensive Crisis
(Emergency care needed)
Higher than 180 or Higher than 110
JSM 2015 5This chart reflects blood pressure categories defined by the American Heart Association.

6. Study Design Proposal
 Feasibility design; superiority 2:1 RCT with N=~100
 Repeated measures: baseline, w4, and w8
 Demonstrate effect size = 0.6 at w8
 Uncontrolled hypertension (US only)
– OSBP in 150 – 180 mmHg
– ASBP in 135 – 170 mmHg
– Required 4 weeks washout phase prior to randomization
 Two treatment strategies
– Renal denervation
– Masked procedure (renal angiogram)
 Medication rescue (if OSBP ≥ 180 mmHg)
 Primary assessments
– ASBP change in 8W (mean reduction difference)
– MAE for safety (not powered)
JSM 2015 6

7. Medication Rescue
 Subjects will remain off antihypertensive
medication through the primary efficacy
assessment unless rescue changes are
required
 Rescue medications will be introduced if a
subject’s OSBP is ≥180mmHg at two
consecutive visits (recheck within 3 days)
 To use vs. not to use data after medication
rescue
JSM 2015 7

8.  Two time points: baseline and w8
 Expected effect size = 0.6
 Test significance level = 0.025 (1-sided)
 Test vs. control ratio = 2:1
 Power ≈ 80%
 Attrition rate = 7%
 Number of subjects randomized ≈ 100
JSM 2015 8
Sample Size

9. Repeated Measures Scenarios
(Data Mechanism)
 Subjects complete treatment strategy and
all assessments are performed
 Subjects complete treatment strategy but
some assessments are not performed
 Subjects discontinue treatment strategy
due to dropout so assessments are not
performed after dropout
 Subjects discontinue treatment strategy
due to rescue medication but assessments
are still performed after rescue
JSM 2015 9

10. Analysis Sets
 Intention To Treat (ITT)
– Subj assigned A vs. subj assigned B regardless of
whether or not receiving or completing that
treatment strategy by protocol
– Will include assessments after rescue medication
 Modified ITT (MITT)
– Will not include assessments after rescue
medication
 Worse Case Scenario (WCS)
– Impute any missing assessment with worst in
test arm, best in sham arm
JSM 2015 10

11. Power Simulation
 T-test vs. repeated measures T-test
 Correlation structure: utilize observed
correlation matrix from historical data
 Endpoint effect size (w8) = 0.6
 Interim effect size (w4) = 0.6, 0.3, 0.0
 Example:
– T=(152,149,146); C=(152,152,153); std=10
– Reduction: T=(-3,-6); C=(0,+1)
– Treatment effect (w4,w8): (-3, -7)
JSM 2015 11

12.  Data mechanism
– Subjects complete treatment strategy and all
assessments are performed (~97%)
– Subjects discontinue treatment strategy due to
rescue medication but assessments are still
performed after rescue (~3%)
– Minimal dropouts (rule out MAR/MCAR) (~0%)
 Sensitivity
– To use data after medication rescue (ITT)
– Not to use data after rescue (MITT)
– Impute any missing with worst case scenario (WCS)JSM 2015 12
Power Simulation (Cont’d)

13. Power Analysis
(T-Test vs. Repeated)
w4 ∆Tt – ∆Tc = -6 ∆Tt – ∆Tc = -3 ∆Tt – ∆Tc = 0
w8 ∆Tt – ∆Tc = -6 ∆Tt – ∆Tc = -6 ∆Tt – ∆Tc = -6
T-Test Repeat T-Test Repeat T-Test Repeat
ITT 79% 82% 79% 68% 79% 45%
MITT 74% 80% 75% 64% 76% 45%
WCS 48% 51% 46% 36% 43% 18%
JSM 2015 13
Assuming all standard deviation = 10 mmHg

14. JSM 2015 14

15. Take-Aways
 In superiority ITT is the analysis of choice
with appropriate support provided by MITT
 To preserve the property of ITT analysis,
subjects treated with rescue medication
should also be analyzed
 Direct comparison without baseline
adjustment may be favorable; repeated
measures can be as supplementary analysis
 Meaningful interim effect is not ignorable
when planning sample size in repeated
measures modelJSM 2015 15

16. Thank You! Enjoy Your Stay!JSM 2015 16