8/12/2019 Code and Quasi Code
2/2
dataset then heuristics and analytics are called for3
.n many datasets there will be both types of variables3
#herefore in practice it will be necessary
to apply more than one set of techni*ues3 4or e'ample, an
adverse drug reaction report may haveidentifying information about
the patient and the reporter3 .t will also have basic
demographics
on the patient3 #he former are the identifying variables and
most of the patient demographics
would be *uasi1identifiers3
Dealing With Quasi-identifiers
ealing with directly identifying variables, as described in the
section above, is insufficient toensure that the data is truly
de1identified3 %s illustrated in #able 5, there are real e'amples
of
datasets that had the identifying variables suppressed or coded,
and the individuals were still
reidentified3
.f there are *uasi1identifiers in a dataset, then their
de1identification is necessary3 #he main*uestion to as) about the
*uasi1identifiers is whether we are ma)ing de1identification
decisions
before data is collected or after the data is collected3 .f
before, then a set of heuristics need to be
used3 #he heuristics are li)e 6rules of thumb7 that can inform
data collection activities to ensure
that data is collected anonymously, or that the data is
de1identified at the earliest opportunityafter collection3 4or
e'ample, assume that an EB is reviewing a research %n overview
of
techni*ues for de1identifying personal health information
protocol, the investigator has claimedthat the data to be collected
is de1identified, and is accordingly ma)ing a case for waiving
consent3 #he EB can then apply the heuristics to ma)e a 8udgment
on whether the data to be
collected is truly de1identified3 euristics are appro'imations
to the best decision regarding howto de1identify a dataset3 But in
the absence of actual data, they are a reasonable approach to
de1
identification3
.f data have already been collected, then both the heuristics as
well as analytics (statistical and
computational methods+ can be applied to the data3 #he analytics
techni*ues involve the analysisof the dataset itself and then
transforming it so that it is de1identified3
