Biostatistics Analysis Center Center for Clinical Epidemiology and Biostatistics University of Pennsylvania School of Medicine

1
Biostatistics Analysis CenterCenter for
Clinical Epidemiology and BiostatisticsUniversity
of Pennsylvania School of Medicine

• Minimum Documentation Requirements
• Amy Praestgaard
• March 6, 2008

2
Project Directory File Structure

• Located on the network to ensure regular back-up.
• Name of directory usually includes project number
  (e.g., BIO999 or EPI888)
• Has (at least) the following subdirectories
• Documentation
• Data
• Original
• Analytic
• Programs
• Current
• Archived
• Results
• Current
• Archived

3
Documentation Subdirectory

• Contact sheet (REQUIRED)
• Must include names and contact information for
  PI, Biostatistics faculty members, and BAC staff
  members.
• Project registration is sufficient for many
  projects.
• Larger projects may include need to include
  project managers, research assistants, and CRCU
  contacts.
• BAC Project Plan (REQUIRED -- will overlap with
  overall project plan)
• Study objective(s), Outcome definition, Statement
  of analytic methodology (analysis plan). At
  minimum, a note in the issues log should contain
  the BAC project goal, even if as simple as this
  project converts a SAS file to an SPSS file
• Analytic efforts related to timeline for
  abstracts, meetings, manuscripts.
• Study information (STRONGLY RECOMMENDED)
• Study protocol or grant proposal, data collection
  forms, meeting minutes or summaries, To-do lists
  (if applicable), all abstract and manuscript
  drafts.
• Issues log (STRONGLY RECOMMENDED)
• Includes brief description, priority, and
  estimated resolution date.
• Keep resolved issues on the list, noted with date
  resolved.

4
Data SubdirectoryBIO999\Data

• Good idea to separate the original data from the
  created
• Original (Raw) Data BIO999\Data\Original
• Original versions of source data (e.g., DVD,
  external hard drive, Excel spreadsheet) must be
  stored in a secure location, like a locked desk
  drawer.
• An electronic gold copy of the source data must
  be retained in the data subdirectory. It may not
  hurt to include GOLD in the file name.
• Analytic Data BIO999\Data\Analytic
• Original extract from source data
• Cleaned data, in original format
• Incorporates data changes, deletions
• Does not include derived variables
• Final analytic data set, including derived
  variables

5
Data Dictionary

• At minimum, must contain
• Results from PROC CONTENTS on the analytic
  dataset containing labeled variables.
• Results from PROC MEANS that includes N for each
  variable.
• A separate document may be necessary for some
  projects.
• If the data base is very large, spend billable
  time creating labels only for the variables you
  will be using for the analysis unless the client
  specifically asks you to do it for all variables.
  

6
Program headers

• All programs have a header with
• program name,
• purpose,
• location,
• project name,
• faculty and BAC name,
• input, output,
• last modified,
• and other relevant details

7
Code Documentation

• Code should have a sufficient number of comments,
  especially on macros, complex merges, and code
  for analysis
• See the Code Complete book for a fuller
  discussion of good documentation
• Within a program, the following types of code
  must be annotated with a description of purpose
  and expected outcome
• All macros or functions
• Any code generating a data set that is to be
  retained for interim or final analysis
• Any code generating output (e.g., frequency
  tables, test statistics, p-values) for an interim
  or final deliverable.
• Any code used for data validation, data changes,
  and derived variable calculation.

8
Program Archiving

• During active analytic periods, log and listing
  files must be archived as follows
• With date clearly indicated in the file name, as
  in gmm_8_8Dec07.log and gmm_8_8Dec07.lst.
• In a separate archive or history section of
  the appropriate BAC project folder
• At least weekly, and more frequently if
  substantial changes are made.
• Log and list files can be run in Windows, no need
  to use unix to generate them
• Programs generally should have a date in their
  names in order to keep track of versions

9
Documenting Deliverables

• All interim and final deliverables must include
  the following information
• Analyst name
• Deliveree
• Date produced (good practice to include in file
  name)
• Input data, output data, and program location
• Statistical software (including version) used to
  produce results
• Once distributed, the deliverable will be
  retained in the appropriate project folder