Screening and diagnostic testing are not primary activities of
pharmacoepidemiology, but they are important functions of both public
health and epidemiology. Assessing the usefulness of a screening
or diagnostic test is based on the sensitivity, the specificity,
and the predictive value of the test’s results compared
with what is actually occurring. One example of testing that is
very pertinent to drug use are tests (e.g., urine, hair, breath)
designed to assess whether a person has been using certain psychoactive
drugs. But, first, there must be a discussion of data quality, including the
concepts of validity and reliability with regard to research results.
One of the most important aspects of research results pertains
to their validity. Interpretation of research results begins with
judgments about their accuracy. The validity of
a measure refers to the degree to which it actually measures what
it is designed to measure. Internal validity is
the extent to which the results of a study accurately reflect the
situation in reality, whereas external validity is
the extent to which the study’s results are applicable
to other populations.
One way of appraising validity is to compare a set of criteria
known or believed to be close to reality. In the absence of this
kind of criteria, it would be helpful to know the results of any
follow-up study showing association between the results of the test
and subsequent events (e.g., diseases, drug use problems). Associations
between known criteria and other variables are one way of appraising
validity, but there are others: developing a consensus of experts’ opinions;
using a set of questions that covers all of the essential components
of what they purport to measure (i.e., content validity); and using
measures that give the same results when repeated (i.e., reliability
of the measure).
Reliability is defined as the degree
of stability exhibited when a measurement is repeated under identical conditions.
In other words, reliability refers to the degree to which a measure
or result can be replicated. Lack of reliability may arise from
divergence among observers or instruments of measurement or from
the instability of the attribute being measured. Reliability is
not a guarantee of validity. It is usually measured by performing
two or more independent measurements and comparing the findings.
The goals of such independent measurements can include determination
of whether the observers vary on their measurements (interrater
variation), there are differences between measurements made by the
same observer at different times (intrarater variation), the measurement
instruments differ, or the attribute being measured is itself unstable.
Bias is systematic error in a study
that leads to distortion of the results. When bias occurs, the study’s results
do not accurately reflect reality. Bias can result during the selection
of a study sample or during information and data collection, or
it can result from the influence of a confounding variable. Selection
bias is an especially important problem in case-control studies.
In cohort studies ...