Professional Documents
Culture Documents
Lec11CaseControlStudies (Revised07)
Lec11CaseControlStudies (Revised07)
(Retrospective Studies)
What is a cohort?
Cohort
A group of individuals who share a common
characteristic, e.g. all of the individuals born
in one year (a birth cohort) or a group of
individuals entered in a prospective study or
clinical trial.
The term always carries a connotation that:
individuals are observed over a period of time
and that summary statistics describe the
experience of real individuals rather than
mathematical abstractions, e.g., a cohort
study or a cohort life table.
Design of a Case-Control Study
WERE WERE NOT WERE WERE NOT
EXPOSED EXPOSED EXPOSED EXPOSED
“CASES” “CONTROLS”
When is a Case-Control
Study Warranted?
• Useful as a first step when searching
for an adverse health outcome
• Useful when disease being
investigated is rare
Retrospective Studies
Disadvantages
• information not easily available
• difficult to validate
• quality of survey
• bias in selection
• possible misconception of factors
causing disease
• recall may be flawed
Retrospective Studies
Advantages
• inexpensive
• easy
• requires fewer subjects
• useful for rate of disease or diseases
with long latency periods
• can check for more than one risk factor
• useful for drug induced illness
Selection of Cases and Controls
• Variety of sources
avoid bias
• Choose incidence or prevalent
cases?
• Controls must represent general
population
can be hospitalized or non-hospitalized
Problems in Control Selection
• Controls must be carefully selected
otherwise, results may be confounded
• Coffee and pancreatic cancer study
Apparent dose-response relationship between
coffee consumption and pancreatic cancer
• Cases = confirmed diagnosis of pancreatic cancer
• Controls = from all patients hospitalized at same
time and by same attending physicians as cases
Difficult to know if disease caused by coffee
drinking or some factor closely related to
coffee drinking (confounding factor)
Problems in Control Selection (cont.)
• Coffee and pancreatic cancer study (cont.)
One such confounding factor was cigarette
smoking --- most smokers are coffee drinkers
--- and cigarette smoking is a known risk
factor for pancreatic cancer
To rule out effect of smoking, the data was
stratified for smoking history and reanalyzed
• Current smokers
• Never smokers
and the dose-response relationship with
coffee consumption held for both groups
Problems in Control Selection (cont.)
• Coffee and pancreatic cancer study (cont.)
Objective in control group is to have level of
coffee consumption approximate that in
general population --- and thus, that cases
demonstrate excessive coffee consumption
Problem was that attending physicians
referring both cases and controls were
gastroenterologists
• Controls may have reduced coffee consumption
because of their gastrointestinal problems, e.g.,
esophagitis or peptic ulcer disease
• Controls coffee intake may be abnormally low
compared to general population
Problems in Control Selection (cont.)
• Coffee and pancreatic cancer study (cont.)
Therefore, the dose-response relationship
differences between cases and controls may
not be due to cases drinking more coffee, but
rather, controls drinking less coffee than
expected
• Conclusion: when difference in exposure
is observed between cases and controls,
controls must not have a level of
exposure that is significantly higher or
lower than the general population
Matching
• Concern that cases and controls may
differ in characteristics or exposures
other than that observed in the study
• To overcome this problem, we can
match cases in controls in regard to
potential factors of concern
• Matching selects controls that are
similar to cases in characteristics
such as age, race sex, socioeconomic
status, occupation, etc.
Matching
• Group matching (frequency matching)
proportion of controls with a given
characteristic (variable) is identical to
proportion of cases with the same
characteristic
• Individual matching (matched pairs)
for each case, a control is selected who
is similar to the case for a given
variable(s)
Problems with Matching
• Practical problems
attempting to match too many
characteristics (variables)
• Conceptual problems
once controls are matched to cases
according to a given characteristic
(variable), that characteristic can not be
studied, because matching artificially
establishes identical proportions for that
characteristic among cases and controls
Recall (Interview) Problems
• Limitations in recall
• Recall bias
One group (e.g., mothers with child with
birth defect) may clearly remember
(recall) an event (e.g., mild respiratory
infection)
Other group (e.g., mothers with healthy
child) may not recall any such event
Use of Multiple Controls
• Same type
2, 3, or even 4 controls for each case
(i.e., increasing the ratio of matched
controls to cases), increases the power
of the study
• Different types
where there is concern that the controls
may differ in some way from what is
“expected” in the general population