Epidemiology Test Part 2
Nested Case Control Study
A hybrid design in which a case control study is nested in a cohort study in a cohort
study
Controls are a sample of individuals who are at risk for the disease at the time each case
of the disease develops
Specimen obtained for cohort are frozen or otherwise stored
After disease has developed in some subjects the study begins and specimens from a
relatively small number of people who are included in the case-control study are thawed
and tested
Cases and controls derived from same original cohort so likely to be comparability
between cases and controls
Advantages
Interviews occurred at baselines and data obtained b efore disease developed;
preventing the possibility of recall bias
Abnormalities in biologic characteristics found before the development of the disease
more likely represents risk factors or premorbid characteristics (temporal relationship)
More economical than a cohort study; cost are lower than with a cohort study because
lab tests need to be done only on specimens from subjects who are later chosen as
cases or controls.
Cohort Study
A cohort is defined as a population group, that is followed over a period of time.
Advantages
Permit direct determination of rates
Time sequencing of exposure and outcome
Can study multiple outcomes
Can study rare exposures
Disadvantages
Take a long time
Costly
Subjects lost to follow up
Larger studies are more demanding than smaller ones; challenges due to data collection and
data management.
Selection Bias
If the way in which cases and controls, or exposed and non-exposed individuals, were selected
in such that an apparent association is observed – even if in reality, exposure and disease are
not associated – the apparent association is the result of selection bias.
Information Bias
Means for obtaining subject information in a study are inadequate therefore info gathered
regarding risk factors, exposures, and/or disease outcome is incorrect.
Main information bias: recall bias
Misclassification bias: misclassifying subjects (cases as controls, controls as cases)
Reporting bias: Does not want to report
Non-differential: Problem with data collection
Differential: Differs in different study groups
Confounding Bias
We observe an association and are tempted to derive a causal inference when in fact, the
relationship may not be causal.
Systematic error of interpretation that results from making an unfair comparison, i.e.,
from a failure to take into account important differences between exposed and
unexposed.
Definition of Confounding
Effect of the exposure is mixed with the effect of another variable
Results in biased measure of effect (RR, OR, IRR)
Observed relationship of E O is due, completely or in part, to another factor
Properties of a Confounder
1. Must be a risk factor for the disease
2. Must be associated with the exposure
3. Must not be a result of the exposure
A variable is a confounder if:
It is a known risk factor for the disease
It is associated with the exposure but is not a result of the exposure
A variable is NOT considered a confounder if it is an intermediate in the causal pathway
between exposure and disease.
Consent
Informed consent is the investigators obligation to study subjects
Assurance of privacy and confidentiality
HIPAA, federal law which protects the confidentiality and security of healthcare
information
Matching
Process of selecting the controls so that they are similar to the cases in certain
characteristics, such as age, race, sex, socioeconomic status, and occupation.
Concern in case-control studies is that cases and controls may differ in characteristics
other than the exposure of interest
We control for confounding in study designs by matching
Group matching: Proportion of controls with characteristics = proportion of cases with
characteristics (25% of cases married, controls will be selected so 25% are married)
Individual matching: For each case, select a match with similar characteristic (if case
enrolled in study is 45-year-old white woman, need a 45-year-old white women female
control)
Problems with matching:
Practical: The more characteristics selected, the harder it may be to find a suitable
control for each case.
Conceptual: We cannot study characteristics that we have matched on because we have
artificially established identical proportions in cases and controls.
We can only match on variables that we are convinced are risk factors for the disease,
characteristics we are not interested in investigating in the study.
Causality Criteria
Strength of association
Time sequence
Analogy
Plausibility
Consistency upon repetition
Specificity
Coherence of explanation
Experiment
Biologic gradient
Randomization
Randomized Trials
Generally considered the best of all clinical research designs
Strongest evidence for concluding causation
Provides best evidence results due to intervention and not something else
Why Randomization?
We cannot determine a causal relationship without any comparisons, therefore two
comparison groups are created.
Randomization is similar to tossing a coin to decide assignment of a person to a study
group
Randomization increases the likelihood that (on the average) the groups will be
comparable not only in terms of variables we recognize and measure but also those who
do not recognize and measure.
Important because?
Best assurance that control group (unexposed) is a valid substitute population
Only way to control for unknown factors
Facilities masking of exposure status
Avoids ambiguity of time order of exposure and outcome (most intervention studies
achieve this)
Provides foundation for statistical tests – valid quantification of uncertainty
Blinding
Way in which either subjects and/or observers are not told which treatment group they
have been randomized
Double blind: participants and study personnel are masked
Triple blind: treatment or intervention is unknown to the research participant, the
individuals who administer treatment or intervention, and the individuals who assess
the outcomes.
Effect Measure Modification
When the incidence rate of disease in the presence of 2 or more risk factors differs from the
incidence rate expected to result from their individual effects.
First question: is there an association between exposure and disease?
Second: Is it due to confounding?
Third: If we decide it is not due to confounding (that it is causal) then we ask whether
the association is equally strong in each of the strata that are formed on the basis of
some third variable.
Ex: There is an association between the use of aspirin for treatment of viral illnesses and Reye’s
syndrome. It can affect all people but is distinctly more common in children 4-14 years old. The
effect is modified by age.
A factor that changes the value of an effect measure (OR, RR, RD)
The presence and interpretation of EMM depends on the measure of effect used (OR,
RR, RD)
EMM is not a bias – it is a finding to be reported!
If EMM present, report stratified estimates separately, as they estimate different
underlying effects… a pooled estimate is not possible (or meaningful)
COMPARING EMM AND CONFOUNDING
Characteristics EMM Confounding
Nature Effects that should be Bias that should be
identified and reported controlled
Specific Strata
Relation to Crude Crude reflects “average”; lies Crude lies outside
between
Measure May differ by measure of Typically present for all
association measures
Incidence Rate
Rate at which new events occur in a population
Incidence rate = # of new events in a time period x person time
# of population at risk of disease during specified time
Risk Ratio
Disease risk in exposed
Disease risk in unexposed
RR=(a/a+b)/(c/c+d)
Incidence Rate Ratio
IR exposed
IR unexposed
Relative Risk/Risk Ratio
If RR = 1: we say that no evidence for any increased risk in exposed vs. unexposed individuals
If RR > 1: we say that the risk in exposed persons is greater than the risk in unexposed persons
If RR < 1: we say that the risk in the exposed persons is less than that in the unexposed persons
Risk Difference
Disease risk in exposed – disease risk in non-exposed
(A/A+B)-(C/C+D)
Odds Ratio
(A*D)/(B*C)
Prevalence
# of existing (and new) cases at a given period in time
total population
Incidence Proportion (CI)
# of new cases during period of time
total population at risk
PT 2
Understand what a nested case control study is
Compare it to a conventional cohort (normal) or case control study
Calculate incidence density (incident rate)
Relative risk based on two by two table
Calculate absolute difference
Understand why randomization in intervention studies
Describe confounding (draw the diagram and explain what it means)
Information bias
Selection bias
Confounding bias
Effect measure modification
Guideline for causality
Data set (come up with one odds ratio)
