Pharmacoepidemiology: Bias & Confounding

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 34

PHARMACOEPIDEMIOLOGY

Bias & Confounding

Prih Sarnianto
The Study Question
• A [pharmaco]epidemiologic
investigation:
>> Etiology of unintended outcome[s]
of drug-use[s] >>> Quantitative research.
Study hypothesis
>> A specific statement regarding the
relationship between two variables
—exposure and outcome.
Hypothesis, quantitative study
• A [pharmaco]epidemiologic study :
>> Test the hypothesis of association
between exposure and outcome.
>>> Hypothesis must be quantitative and
specific, must predict a relationship of a
specific size.
Hypothesis, example
• For example:
“Babies who are breast-fed have less illness than
babies who are bottle-fed.”
>> Which illnesses? How is feeding type defined? How large
a difference in risk?
• A better example:
“Babies who are exclusively breast-fed for three
months or more will have a reduction in the
incidence of hospital admissions for
gastroenteritis of at least 30% over the first year
of life.”
Two type of error
Only specific prediction allows one to
draw legitimate conclusions from a
study which tests a hypothesis. But
even with the best formulated
hypothesis, two types of errors can
occur.

• Type-1 - observing a difference when in


truth there is none.

• Type-2 - failing to observe a difference


when there is one.
Hypothesis test, association
• A [pharmaco]epidemiologic study :
>> Test the hypothesis of association
between exposure and outcome.
>>> If there is an association, the exposure
called a risk factor of the outcome.
>>> A risk factor can be:
■ A predictor [marker or proxy]
■ A causal factor
Steps in the Study, Causation
Approaches in the study of the etiology
of drug’s unintended outcomes:
>>> Animal models
>>> In-vitro systems
>>> Observations in human populations
Observations in Human Populations
Sequence of Studies in Human Subjects
Observed Association
>>> If an association is observed, the first
question must always be…

“Is it real?” —accurate?


■ Could it be by chance [random
error]? >>> Precision
■ Could it be due to bias
[systematic error]?
>>> Internal validity
What is accurate result?

Precision >>> degree of random error

Accuracy = Precision + Validity

Validity >>> degree of systematic error


Random vs. Systematic Error
Random error
Deviation of results and inferences from the
truth, occurring only as a result of the
operation of chance.
>>> Produce type-1 or type-2 errors.
Random error applied to the
measurement of an exposure or outcome
is called random misclassification.
>>> Only produce type-2 errors, except if
applied to a confounder or to an
exposure gradient.
Random vs. Systematic Error

Random error
>>> will cancel each other out in the long run, so it
► Can be reduced by increasing the
number of study participants thereby
increasing the statistical precision.
► Dependent on methodological choices.
► For a given result, the degree of chance
variability is quantified by the
confidence interval.
Random vs. Systematic Error
Systematic error: ► Bias
► Confounding

Bias
Non-random deviation of results or interferences from
the truth due to systematic error in sampling or
measurement.
>>> May produce type-1 or type-2 errors, but we usually focus on
type-1 errors due to bias.

In bias, the focus is on an artifact of some part of


research process [assembling subjects, collecting and
analyzing data] that produces a spurious result.
>>> Bias can be either conscious or unconscious. In epidemiology,
bias does not imply prejudice or deliberate deviation from
the truth [as in common usage].
Random vs. Systematic Error

Systematic error: ► Bias


► Confounding

>>> Observational studies are particularly vulnerable


to bias, but such systematic error can occur in
any type of studies.

>>> Bias may be introduced at any point of a study.

>>> Improving study design is the main [or even]


only way to minimize or control bias.
Random vs. Systematic Error
Systematic error: ► Bias
► Confounding

Although has several types, bias usually subdivided


into:
■ Selection bias
>>> related to study subject recruitment or retention
procedures.

■ Information bias
>>> related to procedures used to measure the information
about study variables.

■ Confounding
Systematic Error
Systematic error: ► Bias — selection, information
► Confounding

■ Selection bias
>>> is a method of participant selection that distorts the
exposure-outcome relationship from that present
in the target population.
Case and control selected from hospital patients
>>> Alcoholics with pneumonia are more likely to be
admitted non-alcoholic with pneumonia.
>>> Risk of pneumonia associated with alcoholism
is biased upwards.
Selection Bias

In hospital-based case-control study, patients with alcoholism seemed to


have significantly higher risk, 2.25 times, contracting pneumonia but, in fact,
did not.
Selection Bias
■ Selection bias
>>>> occurs when the selection of participants in one group results in a
different outcome than the selection for the other group

► Select volunteers as exposed group and non-volunteers as


non-exposed group in a study of screening effectiveness
>>> Volunteers could be more health conscious than non-volunteers,
thus resulting in less disease.
>>> Volunteers could also be at higher risk, such as having a family
history of illness, thus resulting in more disease.
► Study health workers in a workplace exposed to some
occupational exposures comparing to general population
>>> Working individuals are likely to be healthier than general
population that includes unemployed people [Healthy Worker Effect].
► Use of prevalence cases instead of incidence cases.
Selection Bias
■ Controlling selection bias
► Define criteria of selection of affected and non-affected
participants independent of exposures in case-control
study.
► Define criteria of selection of exposed and non-exposed
participants independent of drug’s unintended effect
outcome in cohort study.
► Use randomized clinical trials [RCTs] .
Information Bias
■ Information bias
>>>> occurs when information is collected differently between
two groups, leading to an error in the conclusion of the
association.
► When information is incorrect, there is a misclassification:
>>> Differential misclassification occurs when the level of
misclassification differs between two groups.
>>> Non-differential misclassification occurs when the level of
misclassification does not differ between two groups.
Information Bias
■ Example of information bias
► Interviewer knows the status of the subject before the
interview process
>>> Interviewer may probe differently about exposures in the past
if s/he knows the subjects as cases.

► Subject may recall past exposure better or in more detail if


s/he has the disease [recall bias].
► Surrogates, such as relatives, provide exposure
information for dead cases, but living controls provide
exposure themselves.
Information Bias
■ Controlling information bias
► Have a standardized protocol data collection.
► Make sure sources and methods of data collection are
similar for all study groups.
► Make sure interviewers and study personnel are unaware
of exposure/unintended drug effect status.
► Adapt strategy to assess potential information bias.

>>> Minimize loss to follow up in cohort studies


Information Bias
Random vs. Systematic Error
Systematic error: ► Bias
► Confounding

Bias Confounding
is a systematic error in a may lead to errors in the
study and cannot be fixed. conclusion of a study, but,
when confounding
variables are known, the
effect may be fixed.
Systematic Error, Confounding
Systematic error: ► Bias
► Confounding

Confounding
• occurs when the observed result between exposure
and disease differs from the truth because of the
influence of the third variable.
• For example, crude mortality rate (crude effect) of City
A differs from the rate of City B—but after adjusting
for age, the adjusted rates do not differ
>>> Age distribution differs between the two cities.
► Age confounds the association.
Age as a Confounder
Age as a Confounder
Age as a Confounder
Systematic Error, Confounding
Systematic error: ► Bias
► Confounding

Approaches to the Problem of Confounding


• In designing and carrying out the study:
>>> Matching
• In data analysis:
>>> Stratification
>>> Adjustment
Systematic Error, Confounding
Bias and Confounding
Bias and Confounding
Systematic error: ► Bias
► Confounding

When an Association Does Exist


• To conclude that an association between exposure
and disease outcome exists:
>>> The study must have adequate sample size
>>> The study must be free of bias
>>> The study must be adjusted for possible confounder

• We can the pursue the original objective of whether the


exposure is the causal factor of the unintended drug effect.
THANK YOU!

prih1488@gmail.com

You might also like