Title: Issues of Simultaneous Tests for Non-Inferiority and Superiority
1Issues of Simultaneous Tests for Non-Inferiority
and Superiority
- Tie-Hua Ng, Ph. D.
- U.S. Food and Drug Administration
- Ng_at_cber.fda.gov
- Presented at
- MCP 2002
- August 5-7, 2002
- Bethesda, Maryland
- _______
- The views expressed in this presentation are
not necessarily of the U.S. Food and Drug
Administration.
2Simultaneous Tests for Non-Inferiority and
Superiority
- Multiplicity adjustment is not necessary
- Intersection-union principle (IU)
- Dunnett and Gent (1996)
- Closed testing procedure (CTP)
- Morikawa and Yoshida (1995)
- Indisputable
3A Big Question
- Is Multiplicity Adjustment Necessary?
4- Is
- Multiplicity
- Adjustment
- Necessary?
5Outline
- Assumptions and Notations
- Switching between Superiority and Non-Inferiority
- Is Simultaneous Testing Acceptable?
- Use of Confidence Interval in Hypothesis Testing
--- Pitfall - Problems of Simultaneous Testing
- Conclusion
6Assumptions/Notations
- Normality and larger is better
- T Test/Experimental treatment (?t)
- S Standard therapy/Active control (?s)
- ? Non-Inferiority Margin (gt 0)
- For a given d (real number), define
- Null H0(d) T ? S - d
- Alternative H1(d) T gt S - d
- Non-Inferiority d ?
- Superiority d 0
7Non-Inferiority (d ?)
- H0(?) T ? S - ? against H1(?) T gt S - ?
H1(?)
H0(?)
?
S
Worse
Better
Mean Response
8Superiority (d 0)
- H0 (0) T ? S against H1 (0) T gt S
H1(0)
H0(0)
S
T
Worse
Better
Boundary
Mean Response
9Switching between Superiority and
Non-Inferiority
- CPMP (Committee for Proprietary Medicinal
Products), European Agency for the Evaluation of
Medicinal Products - Points to Consider on Switching Between
Superiority and Non-Inferiority, 2000. - http//www.emea.eu.int/htms/human/ewp/ewpptc.htm
10Switching between Superiority and
Non-Inferiority (2)
- Non-Inferiority Trial
- If H0(?) is rejected, proceed to test H0(0)
- No multiplicity issue, closed testing procedure
- Superiority Trial
- Fail to reject H0(0), proceed to test H0(?)
- No multiplicity issue
- Post hoc specification of ?
11Switching between Superiority and
Non-Inferiority (3)
- Non-inferiority Trial
- Intention-to-treat (ITT)
- Per protocol (PP)
- Superiority Trial
- Primary Intention-to-treat (ITT)
- Supportive Per protocol (PP)
- Assume ITT PP
12Simultaneous Testing
- One-sided 100(1 - ?) lower Confidence Interval
for T - S
Superiority
Non-inferiority
Neither
-?
0
Test is worse
Test is better
Mean Difference (T S)
13Simultaneous Testing (2)
- Multiplicity adjustment is not necessary
- Dunnett and Gent (1996)
- Intersection-Union (IU)
- Superiority Both H0(?) and H0(0) are rejected
- Morikawa and Yoshida (1995)
- Closed Testing Procedure (CTP)
- Test H0(0) when H0(?)?H0(0) is rejected
14Simultaneous Testing (3)
- Discussion Forum (October 1998)
- London
- PSI (Statisticians in Pharmaceutical Industry)
- Is Simultaneous Testing of Equivalence
Non-Inferiority and Superiority Acceptable? - Superiority trial
- Fail to reject H0 (0)
- No equivalence/non-inferiority claim
- Ok Morikawa and Yoshida (1995)
- Ref Phillips et al (2000), DIJ
15Is Simultaneous TestingAcceptable?
16Use of Confidence Interval in Hypothesis Testing
- H0(d) T ? S - d (at significance level ?)
- One-sided 100(1-?) lower CI for T-S
- Reject H0(d) if and only if the CI excludes -d
Reject H0(d)
Do not reject H0(d)
-d
Test is worse
Test is better
Mean Difference (T S)
17Use of Confidence Interval in Hypothesis Testing
(2)
- If CI (L, ?), then H0(d) will be rejected for
all -d lt L. - A Tricky Question
- Suppose CI (-1.999, ?), L -1.999
- H0(2) T ? S - 2 is rejected (d2) since -d lt L
- Can we conclude that T gt S - 2?
- Yes, if H0(2) is prespecified.
- No, otherwise.
18Use of Confidence Interval in Hypothesis Testing
(3)
- Post hoc specification of
- H0(d)
- is a
- No No
19Simultaneous Testing Problems
- H0(d1) and H0(d2), for d1 gt d2
- One-sided (1 - ?)100 lower CI for T - S
Reject H0(d2)
Reject H0(d1)
Neither
-d2
-d1
Test is worse
Test is better
Mean Difference (T S)
20Simultaneous Testing Problems (2)
- H0(d1), H0(d2) and H0(d3), for d1 gt d2 gt d3
- One-sided (1 - ?)100 lower CI for T - S
Reject H0(d3)
Reject H0(d2)
Reject H0(d1)
None
-d3
-d1
-d2
Test is worse
Test is better
Mean Difference (T S)
21Simultaneous Testing Problems (3)
- H0(d1), H0(d2),, H0(dk), for d1 gt d2 gt gt dk
- One-sided (1 - ?)100 lower CI for T - S
Reject H0(dk)
. . .
Reject H0(d2)
Reject H0(d1)
None
-dk
-d1
-d2
-d3
Test is worse
Test is better
Mean Difference (T S)
22Simultaneous Testing Problems (4)
- Choose k large enough
- Pr-d1 lt Lower limit lt -dk close to 1
- Max dk - dk-1 lt a given small number
- Simultaneous testing of H0(di), i 1,, k
- ? Post hoc specification of H0(d)
23Simultaneous Testing Problems (5)
24Simultaneous Testing Problems (6)
- What is wrong with IU and CTP?
- Nothing
- PrRejecting at least one true null ? ?
- What kind of problems?
25Simultaneous Testing Problems (7)
- Post hoc specification of H0(d)
- Let -d0 100(1 - ?) lower limit - ?
- Reject H0(d0), since -d0 lt lower limit
- Repeat the same trial independently
- PrRejecting H0(d0) 0.5
26Simultaneous Testing Problems (8)
- Simultaneous testing of many H0(d)
- Repeat the same trial independently
- Low probability of confirming the finding
- 1st trial Reject H0(dj) but not H0(dj1)
- 2nd trial PrRejecting H0(dj) is low (e.g.,
0.5)
27Simultaneous Testing Problems (9)
- Simultaneous testing of H0(?) and H0(0)?
- Confirm the finding
- ? 2
- Known variance
- Let ? ? T - S
- Significance level ? 0.025
- 80 power for H0(?) (at ? 0)
28Simultaneous Testing Problems (10)
f?(?) PrRejecting H0(?) ? f0(?)
PrRejecting H0(0) ?
29Simultaneous Testing Problems (11)
- Test one null hypothesis H0(?)
- Suppose that H0(?) is rejected
- Repeat the same trial independently
- PrRejecting H0(?) again f?(?)
30Simultaneous Testing Problems (12)
- Test H0(?) and H0(0) simultaneously
- Suppose that H0(?) or H0(0) is rejected
- Repeat the same trial independently
- PrRejecting the same null hypothesis again 1
- w(?) f?(?) w(?) f0(?) - f?(?) - f0(?) 1 f0(?)/f?(?),
- where w(?) f0(?)/f?(?)
31Simultaneous Testing Problems (13)
1 - w(?) f?(?) w(?) f0(?) where w(?)
f0(?)/f?(?)
Simultaneous tests in the 2nd trial
32Simultaneous Testing Problems (14)
- Ratio 1 f0(?)/f?(?) 1 f0(?)/f?(?)
- Ratio may be as low as 0.75
33Conclusion
- Many H0(d) Problematic
- Not type I error rate
- H0(?) and H0(0) Acceptable?
- If zero tolerance policy No
- If 25 reduction cannot be tolerated No
- If 25 reduction can be tolerated Yes
34Is Simultaneous Testing of H0(?) and H0(0)
Acceptable?
35You be the judge
36References
- Dunnett and Gent (1976), Statistics in Medicine,
15, 1729-1738. - Committee for Proprietary Medicinal Products
(CPMP 2002). Points to Consider on Switching
Between Superiority and Non-Inferiority.
http//www.emea.eu.int/htms/human/ewp/ewpptc.htm - Morikawa T, Yoshida M. (1995), Journal of
Biopharmaceutical Statistics, 5297-306. - Phillips et al., (2000), Drug Information
Journal, 34337-348.
37God Bless America