Human Subjects Issues in the Analysis of Existing Data

1
Human Subjects Issues in the Analysis of
Existing Data

• Richard T. Campbell
• University of Illinois at Chicago

2
Public Use Files (PUFs)

• A public use file is made available by a data
  distributor under the assumption that persons
  supplying data can not be identified.
• Examples
• Census
• General Social Survey
• Health and Retirement Survey

3
Preventing Identification

• Remove direct identifiers
• Remove or fuzz indirect identifiers
• Coarsen income and other variables
• Remove detailed geographic information
• Add random noise to variables

4
Who Decides If a Datefile is Sufficiently
De-identified?

• Federal agencies must meet protection standards.
  These are evolving.
• Non-government agencies, such as NORC and ISR,
  meet local IRB criteria

5
Local IRBs Are Sometimes Confused by This Issue

• Investigators using PUFs frequently are required
  by local IRBs to demonstrate that data are
  non-identifiable
• This is a rather technical issue and usually
  neither the investigator or the IRB has the
  necessary expertise.

6
NHPRAC Recommendations http//ohrp.osophs.dhhs.go
v/nhrpac/

• A PUF, once certified by an IRB, does not contain
  human subjects data and is outside the
  regulations not exempt but not covered.
• Hence IRBs would not review such research.
• The onus of approval would be on the distributor
  of the data, not the recipient.
• These are recommendations to OHRP and they may be
  substantially modified.

7
Enhanced Data Files

• Investigators sometimes want to add data to
  public files such as geographical information
• The added data sometimes increases the risk
  (probability) of identification or raises the
  magnitude of harm which could result if a subject
  were identified
• In such cases, an IRB must review the research

8
Restricted Distribution of Enhanced Files

• Data distributors are increasingly likely to make
  two versions of a file available, a PUF and an
  enhanced version that contains, for example,
  more geographic detail.
• Investigators must meet more stringent criteria
  to use such files, such as filing a data security
  plan.
• IRBs need to review such research.

9
Data Enclaves

• Some organizations, particularly the Census, are
  creating secure environments where investigators
  can go to do analyses on unexpurgated data.
• Investigators have to file a research plan and
  meet strict criteria for what can be printed,
  saved and removed from the site.

10
Future Developments

• As usual, the internet has changed everything.
• In the old days, with data on round tapes and
  mainframes, one could rely on investigators to
  protect data
• Now, data can be analyzed at will by any high
  school student with a computer, e.g. the GSS.
• The research community is struggling to come to
  terms with these developments.