Week 4 - Observational Studies I
Observational studies
● “Natural experiments” where the investigator does not manipulate treatment or exposure
to disease but rather observes natural variation
● Advantages:
○ Some risk factor/disease associations are not appropriate for experimental study
(ethically)
○ Can have high external validity (real-world)
● Dis-advantages:
○ Prone to bias
Case Reports
● Consist of one person (patient)
● Typically describe unusual symptoms or syndromes observed in this one person
● May contain speculation about exposure
Case Series
● A group of case reports with similar symptoms/common exposures
● Cohort study without comparison group
○ Essentially a case series, but involving a group of people with the same level of
exposure who are followed up over time
○ Involves a defined study protocol and inclusion/exclusion criteria
● Case series often seen in surgery realm
● Hypothesis generating
● What’s missing in a case series? → a control group!
Control group
● Sets what you’re studying into comparative context
● Example:
○ Case series → 10% of B-average students who take vitamin X before an exam
will get an A+
○ Is this good?
○ Hypothesis generation → vitamin X improves average students’ scores on exams
○ How to test hypothesis? → compare students taking vitamin X to students taking
another vitamin or no vitamin (control group)
○ If 50% of B-average students who take vitamin Y get an A= on the exam, then
vitamin X doesn't look good anymore!
○ To assess the impact of something (e.g., test a hypothesis about vitamin X), we
must compare it to sometime else
Case-control Study
● Observational epidemiologic study with a comparison
group
, ● Subjects are grouped according to disease status
○ Diseased
○ Not diseased
● Connect exposure information (historical) on subjects
● Classify subjects according to exposure status
○ Exposed
○ Not exposed
Selecting cases
● Representative of all possible cases?
○ If interest is in ‘average’ patient, make sure you’re not selecting the sickest
patients
● Specify clear criteria for inclusion of cases
○ E.g. age ≧ 65, DSM-IV criteria for dementia, MMSE score ≨ 18, not receiving
cholinesterase inhibitor
● Incident preferable to prevalent cases
○ Incident → newly diagnosed cases during time period
○ Prevalent → cases who are known to have disease
● Prevalent cases likely to have disease for a long time (e.g. chronic illness)
○ Harder to assess exposure prior to such disease → poor memory, bad
record-keeping, lost records
○ Not only measuring exposure, but also variables associated with duration of
disease → survivor bias: healthier people last longer with disease (prevalent
cases likely to be healthier)
Selecting Controls
● Controls must come from the same source population as cases
○ If cases come from a hospital, then select controls from the population served by
the hospital
○ Controls must be people who do not have the disease under investigation but
who would be included in the study if they got the disease
■ If a control got sick, then she would go to the same hospital from where
cases were selected
■ Controls are drawn from the population “at risk” of becoming cases
Matching
● What happens if you want to ensure that cases and controls are balanced on certain
factors?
● For example, you want the same number of males/females in your study
● Match controls to cases
Individual matching
● For every male case, select a male control (same for females), one to one matching
● For every male case, select four male controls (same for females), one to one matching
Observational studies
● “Natural experiments” where the investigator does not manipulate treatment or exposure
to disease but rather observes natural variation
● Advantages:
○ Some risk factor/disease associations are not appropriate for experimental study
(ethically)
○ Can have high external validity (real-world)
● Dis-advantages:
○ Prone to bias
Case Reports
● Consist of one person (patient)
● Typically describe unusual symptoms or syndromes observed in this one person
● May contain speculation about exposure
Case Series
● A group of case reports with similar symptoms/common exposures
● Cohort study without comparison group
○ Essentially a case series, but involving a group of people with the same level of
exposure who are followed up over time
○ Involves a defined study protocol and inclusion/exclusion criteria
● Case series often seen in surgery realm
● Hypothesis generating
● What’s missing in a case series? → a control group!
Control group
● Sets what you’re studying into comparative context
● Example:
○ Case series → 10% of B-average students who take vitamin X before an exam
will get an A+
○ Is this good?
○ Hypothesis generation → vitamin X improves average students’ scores on exams
○ How to test hypothesis? → compare students taking vitamin X to students taking
another vitamin or no vitamin (control group)
○ If 50% of B-average students who take vitamin Y get an A= on the exam, then
vitamin X doesn't look good anymore!
○ To assess the impact of something (e.g., test a hypothesis about vitamin X), we
must compare it to sometime else
Case-control Study
● Observational epidemiologic study with a comparison
group
, ● Subjects are grouped according to disease status
○ Diseased
○ Not diseased
● Connect exposure information (historical) on subjects
● Classify subjects according to exposure status
○ Exposed
○ Not exposed
Selecting cases
● Representative of all possible cases?
○ If interest is in ‘average’ patient, make sure you’re not selecting the sickest
patients
● Specify clear criteria for inclusion of cases
○ E.g. age ≧ 65, DSM-IV criteria for dementia, MMSE score ≨ 18, not receiving
cholinesterase inhibitor
● Incident preferable to prevalent cases
○ Incident → newly diagnosed cases during time period
○ Prevalent → cases who are known to have disease
● Prevalent cases likely to have disease for a long time (e.g. chronic illness)
○ Harder to assess exposure prior to such disease → poor memory, bad
record-keeping, lost records
○ Not only measuring exposure, but also variables associated with duration of
disease → survivor bias: healthier people last longer with disease (prevalent
cases likely to be healthier)
Selecting Controls
● Controls must come from the same source population as cases
○ If cases come from a hospital, then select controls from the population served by
the hospital
○ Controls must be people who do not have the disease under investigation but
who would be included in the study if they got the disease
■ If a control got sick, then she would go to the same hospital from where
cases were selected
■ Controls are drawn from the population “at risk” of becoming cases
Matching
● What happens if you want to ensure that cases and controls are balanced on certain
factors?
● For example, you want the same number of males/females in your study
● Match controls to cases
Individual matching
● For every male case, select a male control (same for females), one to one matching
● For every male case, select four male controls (same for females), one to one matching
