SelfDesigning Trials: Further Thoughts

1
Self-Designing Trials Further Thoughts
Advances

• Lloyd D. Fisher, Ph.D.
• Professor Emeritus, Biostatistics, University of
  Washington
• Biostatistical Consultant

2
Outline of Talk

• A clue as to the method
• Simple Example Verbal Handwaving
• Combing with usual sequential monitoring
• The How is clear
• Practicalities
• Introduction into the middle of a study without
  prior planning
• Back to the basic idea
• Practicalities

3
Outline Continued

• Survival Data
• PhD nominally working on this
• One method (somewhat like group sequential)
• Comments on the way we do business
• Pharmaceutical, Biologics and Devices
  statisticians dont bother to fasten seatbelts?
  Vaccinate children? Like to stand under trees in
  lightening storm?
• Long past time for a change, good for society and
  also industry as well as in accord with our
  statistical traditions

4
Example - 1
5
Example - 2
6
Example - 3
7
Example - 4
8
Example - 5
9
Example - 6
10
Example - 7
11
Example - 8
12
Example - 9
13
Example 10
14
The Principle

• From an hypothesis testing point of view at any
  point in the RCT experiment can modify the
  design, data used for testing, etc. as long as
  the conditional probability of rejection under
  the null hypothesis is the same or larger as for
  the original design using all the past data
  (including by treatment arm).

15
Combining Self-Designing Randomized Clinical
Trials, SDRCTs, with a Lan-DeMets Spending
Function, OBrien-Fleming Boundary, Etc.

• In many, perhaps most, situations the interim
  looks at data require extreme data so as to
  preserve the power of the study.
• If this is so the using for example, the
  Bonferroni inequality, can use SDRCT with the
  more classical looks.

16
SDRCT with Lan-DeMets

• Practicalities
• Extending studies involves resources in a
  commercial or NIH type setting someone with
  authority needs to have access to data as well as
  resources
• At beginning get a statistical idea of the length
  of a study
• Uncertainty is superior to beginning from
  scratch with a barely negative study

17
SDRCT with Lan-DeMets - 2

• Possible DSMB/C set-up
• Open meeting with blinding intact
• Restricted unblinded meeting with sponsor and
  DSMB
• Restricted unblinded meeting for DSMB only
• Industry/FDA interactions have improved
  tremendously over the last 20 years
• Need for even more during ongoing studies

18
Introducing SDRCTs Without Preplanning During a
Trial

• Remember the basic principle!

19
Introducing SDRCTs Without Preplanning During a
Trial - 2

• Remember the basic principle!
• Practicalities
• Worry about mischief
• Clear documentation, e.g. dated, witnessed and
  securely archived files, material and decisions
• Another reason agency involvement would be useful
• Clearly a totally independent DSMB cannot do this
• For IRB review and informed consent need to have
  the possibility in the protocol

20
Time to Event Endpoints

• Comparing two groups the log-rank test compares
  observed and expected numbers of events
• By dividing the exposure and events up into
  disjoint groups we can weight new exposure
  depending upon the past experience thus
  effectively changing the sample size

21
A Few Limitations

• Designs are (as expected) not as efficient as if
  have correct alternative initially
• Dont you believe in the sufficiency principle?

22
Comments on the way we do business

• The essence of science is to learn from past
  experience rocket ship to mars analog
• Why use seat-belts? Vaccinate our children?
• Why not reduce risk in an acceptable way
• As you have and will hear there are multiple ways
  now to approach things more appropriately. Now
  is the time.

23
References

• Fisher LD. Self-Designing Clinical Trials.
  Statist. Med. 171551-1562, 1998.
• Shen Y, Fisher LD Statistical Inference for
  Self-Designing Clinical Trials with a One-Sided
  Hypothesis. Journal of the International
  Biometric Society 55(1)190-197, 1999.
• Thach C, Fisher LD Self-designing two-stage
  trials to minimize expected costs. In press,
  Biometrics, 2002.