1. An FDA Statistician’s Perspective on Standardized Data Sets
Steve Wilson & Cynthia Liu
CDER/OPaSS/OB/Division II
CDISC Initiatives – Impact on Electronic Data Submission
FDA/Industry Statistics Workshop
September , 2005
Disclaimer: The views expressed in this presentation are those of the speaker and do not necessarily represent those of the Food and Drug Administration

2. Outline Background and Regulatory Guidance
2004 FDA/Industry Workshop -- William J. Qubeck, Beyond the CDISC SDTM V3.1 Model: Statistical & Programming Considerations
Cynthia Liu -- Review Experience

3. Background: Why Standards?
Improve patient safety and reduced costs by reducing time to market for safe and effective treatments by
Improving efficiency of evaluation of safety and efficacy of investigational treatments
Facilitating communication between regulatory authority and applicant
Facilitating development of efficient review environment (e.g., access to data, orientation, redundancy, training, analysis tools)
Levin, 2005

4. Background: Why Standards?
Improve patient safety and reduced costs by reducing time to market for safe and effective treatments by:
Improving efficiency for clinical research
Facilitating design and conduct of clinical trials
Facilitating communication between researchers and study sponsor (e.g., between CRO and drug company)
Integration with the electronic health record
Levin, 2005

5. Background: Why Standards?
Facilitating communication between regulatory authority and applicant
Facilitating development of efficient review environment (e.g., access to data, orientation, redundancy, training, analysis tools)

6. Regulatory Guidance Guidance, 1999 Study Data Specifications, 2004
Providing Regulatory Submissions in Electronic Format — NDAs
Study Data Specifications, 2004
Submit Both SDTM and Analysis Data
“Study Data Tabulation Model (SDTM) developed by the Submission Data Standard working group of the Clinical Data Interchange Standard Consortium (CDISC)”

7. Study Data Specifications
Individual subject data listings
Data tabulations
Data tabulations datasets
Data definitions
Data listing
Data listing datasets
Analysis datasets
Analysis programs

8. William J. Qubeck, IV MS, MBA
2004 FDA/Industry Workshop -- Beyond the CDISC SDTM V3.1 Model: Statistical & Programming Considerations
William J. Qubeck, IV MS, MBA
Electronic Submissions Data Group Leader
Global Clinical Data Services, Pfizer Inc.
CDISC SDTM Version 3.1
Interventions
Events
Findings
Implementation for Pfizer Submission – First SDTM data submitted to CDER

9. Submission Data Processes
FDA
PFE
CRTs
(Raw)
Extraction
&/or
Transformations DB
Clinical
Algorithms
Analysis
eSub
CDISC Pilot
FDA
Extraction
&/or
Transformations
CRTs
(Raw) DB
Clinical
Algorithms
Analysis
eSub
SDTM
Qubeck, 2004

10. First FDA Statistical Review Experience of CDISC SDTM Submission: Cynthia Liu Division of Biometrics II / CDER

11. The first statistical reviewer at FDA to work with the new data format (Study Data Tabulation Model) and the new tool (Web Submission Data Manager). Terrifying or exhilarating?

12. Preparation for the Review
Understand the submission (indication, drug, study design, endpoints, analysis methods, etc.)
See if data are accessible
Usual format - through EDR (Electronic Document Room)
SDTM format - through WebSDM (Web Submission Data Manager)
See if data are sufficient (primary, secondary, LOCF indicator, etc.)

13. Preparation for the Review (continued)
Attend the first training class of WebSDM held by Lincoln Technologies
Receive a training guide
Receive an introductory training to fundamentals of the SDTM format
Receive a demonstration and hands-on training on WebSDM tool developed for FDA to review the standardized data
Helpful, simple, and straightforward

17. Comparison Between the Analysis Files (customized data sets) and SDTM Files
Use of Sponsor Supplied Analysis Files:
Find the data files containing efficacy variables
Read data definition files (define.pdf) to understand variable names, types, formats, and codes
Write SAS codes to perform statistical analyses for the primary and secondary efficacy variables (e.g., PROC MEANS, PROC MIXED, PROC FREQ, and PROC LOGISTIC, etc.)

18. Comparison Between the Analysis Files (customized data sets) and SDTM Files (continued)
Use of Downloaded SDTM Files from WebSDM:
Download the data file containing efficacy variables (ABC.XPT - huge file, 260K+ rows)
Do extensive data manipulation in the SAS Data Step (will explain in the next few slides)
Write SAS codes to perform statistical analyses for the primary and secondary efficacy variables (e.g., PROC MEANS, PROC MIXED, PROC FREQ, and PROC LOGISTIC, etc.)

19. The extensive data manipulation involved the following:
For the primary efficacy variable
Select the baseline, post-baseline, and % change from baseline for the primary efficacy variable
Select appropriate visit numbers for analyses
Delete duplicate or invalid data - measurement collected more than once for the same visit window and no indication of which data to be used for analyses
Transpose the vertical data format to the horizontal one by subject and visit

20. Downloaded SDTM Efficacy Data
Duplicate data
Four out five studies did not achieve the difference planned, which was supposed to be the minimum clinically meaningful difference.
Invalid data

21. The extensive data manipulation involved the following:
For the primary efficacy variable (continued)
Extract data for each visit
Create a new data set by merging data of each visit so that last valid observation was able to be carried forward (LOCF)
For the secondary efficacy variable
Create a categorical variable (z) based on % change from baseline (y) for responder analysis as defined by the sponsor (e.g., if y  0 then z = 1; else z = 2)

22. Summary of My Experience
If data files are structured and documented well by the sponsor, it is easier to use analysis files than SDTM files for statistical analyses, because less programming is required.
However, contents and numbers of analysis files submitted vary from sponsor to sponsor, or even within the same sponsor.
Because of standardized domain names, variable names, and formats, reviewers can find and understand data more easily; and therefore, minimize the learning curve.

23. Summary of My Experience (continued)
Reviewers can download data of interest through WebSDM for exploratory analyses, instead of waiting for the sponsor’s response (time saving).
Standards are a good thing - imagine no standards for traffic lights when we drive and walk!
One last note - please talk to statistical reviewers and submit analysis files, particularly the ones for the primary and important secondary variables.

24. Am I still terrified? Not anymore. Am I still exhilarated? Certainly!
Acknowledgements
Dr. Steve Wilson
Dr. Todd Sahlroot
Dr. Ed Nevius
Thank you very much for your attention.