Cohort Studies & Randomized Clinical Trials

advertisement
Cohort Studies
Hanna E. Bloomfield, MD, MPH
Professor of Medicine
Associate Chief of Staff, Research
Minneapolis VA Medical Center
Disclosure
• I have no financial relationships to
disclose.
• I will not discuss off label use and/or
investigational use in my presentation
Empowering Evidence 2014
Learning Objectives
By the end of this session participants should understand
• The difference between a prospective and retrospective
cohort study
• The difference between a cohort and a case control
study
• The concept of bias
• The concept of confounding
Empowering Evidence 2014
Evidence Pyramid
Increasing strength
of evidence for
clinical application
Clinical,
Epidemiologic,
Health
Services
Basic
Science
Empowering Evidence 2014
Cohort Studies
• Overview
• How they differ from Case Control
Studies
• Bias
• Confounding
• Characteristics of a GOOD cohort study
Empowering Evidence 2014
Cohort Studies
May be used to study…
• Etiology/ Risk Factors/Prognosis
• Effect of Treatments
– Hypothesis generating!
• May be either
– Prospective
– Retrospective
Empowering Evidence 2014
Prospective Cohort Study
To evaluate Etiology/Risk Factors/Prognosis
Risk Factor
population
Follow-up
1000
High blood pressure
Outcome
60 Heart
Attacks
sample
1000
Normal blood pressure
20 Heart
Attacks
Study begins here
PRESENT, 2014
Empowering Evidence 2014
FUTURE, 2014-18
Retrospective Cohort Study
To evaluate Etiology/Risk Factors/Prognosis
Risk Factor
population
Follow-up
1000
High blood pressure
Outcome
60 Heart
Attacks
sample
1000
Normal blood pressure
20 Heart
Attacks
You act as if study begins here
PAST, 2008
Empowering Evidence 2014
PRESENT, 2014
Cohort Study
To evaluate Treatment
Hypothesis generating only
Risk Factor
Population
of middle
sample
age
people
with
high blood pressure
Follow-up
1000
on treatment
1000
not on treatment
Outcome
60 Heart
Attacks
20 Heart
Attacks
Study begins here
PRESENT, 2014
Empowering Evidence 2014
FUTURE, 2014-18
Cohort Studies
•
•
•
•
•
Overview
How they differ from Case Control Studies
Bias
Confounding
Characteristics of a GOOD cohort study
Empowering Evidence 2014
Case Control Studies
To evaluate Etiology/Risk Factors
60%
High blood pressure
20%
High blood pressure
1000
Prior Heart
Attack
1000
No Prior Heart
Attack
Study begins here
PAST
Empowering Evidence 2014
PRESENT, 2014
Case Control Studies
To evaluate Treatment Efficacy
20%
On Aspirin
60%
On Aspirin
1000
Prior
heart attack
1000
No prior
heart attack
Study begins here
PAST
Empowering Evidence 2014
PRESENT, 2014
Cohort v. Case Control
• Cohort (either prospective or retrospective)
– Subjects are defined by risk factor/treatment
status
– Disease occurrence in the future is then
assessed and compared
• Case Control
– Subjects are defined by disease status
– Past history of risk factor/treatment are then
assessed and compared
Empowering Evidence 2014
Cohort Studies
•
•
•
•
•
Overview
How they differ from Case Control Studies
Bias
Confounding
Characteristics of a GOOD cohort study
Empowering Evidence 2014
Bias and Confounding
• Two problems that can undermine validity
of cohort studies
• Bias
– Systematic error in the design, conduct, or
analysis of a study
• Confounding
– It looks like Factor A causes Disease X but in
fact it is Factor B
Empowering Evidence 2014
Bias
• There are a million types of bias!!
• Some common ones to look for…
– Selection bias
– Information bias
Empowering Evidence 2014
Selection Bias: example
To evaluate Etiology/Risk Factors
population
1000
High blood pressure
60 Heart
Attacks
sample
1000
Normal blood pressure
• HBP recruited from a cardiology clinic
• Normal BP from a primary care clinic
• What’s wrong with this picture??
Empowering Evidence 2014
20 Heart
Attacks
Selection Bias
• Systematic difference in prognostic or
treatment factors between the 2 groups
• In our example….
– One group is more likely to have more cardiac
risk factors or history than the other
– One group is more likely to be aggressively
treated than the other (eg lipids)
Empowering Evidence 2014
Information Bias
To evaluate Etiology/Risk Factors
population
1000
High blood pressure
60 Heart
Attacks
sample
1000
Normal blood pressure
20 Heart
Attacks
• Heart Attack incidence is measured from hospital records
in one group and from patient recall in another
• What’s wrong with this picture??
Empowering Evidence 2014
Bias
• Systematic error in the design, conduct, or
analysis of a study
• The question to ask yourself when reading
a study: Did they do things differently
between the 2 groups?
– Recruitment? Treatment? Follow-up?
Ascertainment of Endpoints? Analysis of
Data?
Empowering Evidence 2014
Confounding
• This is the main problem in ALL
observational studies
• Bias is under control of investigators
– Did they do things differently between the
2 groups?
• Confounding is NOT under the control
of the investigators
– It is endemic to observational studies
– But it can be mitigated
Empowering Evidence 2014
We do this cohort study…
To evaluate Etiology/Risk Factors
population
1000
High blood pressure
sample
1000
Normal blood pressure
60 Heart
Attacks
P<0.001
20 Heart
Attacks
• We have done a good job controlling for bias
• We find a significant association between a
history of HBP and risk of heart attack
• Is that the end of the story?
Empowering Evidence 2014
We still don’t know if…
• Its the high BP that increases the risk of
heart attack or something else (that frequently
accompanies HBP) that is actually the culprit
• In other words, is there “confounding”?
Empowering Evidence 2014
Confounding
in Risk Factor/Etiology Studies
High Blood
Pressure
Risk factor
Heart
Attacks
Outcome
Before we can definitively say that high blood
pressure is a risk factor for heart attacks, we
need to rule out confounding
Empowering Evidence 2014
Population: Middle Aged People in the US
sample
No HBP
20 heart
attacks
Empowering Evidence 2014
HBP
60 heart
attacks
Population: Middle Aged People in the US
sample
No HBP
Don’t Smoke
Exercise
Have normal
cholesterol
20 heart
attacks
Empowering Evidence 2014
HBP
Smoke
Don’t Exercise
Have high
cholesterol
60 heart
attacks
Confounding
in Risk Factor/Etiology Studies
High Blood
Pressure
Risk factor
Heart
Attacks
Outcome
High
Cholesterol
Confounding Variable
A variable that is associated with both
the risk factor and the disease
Empowering Evidence 2014
Confounding
in Treatment Studies
Treatment of
High Blood
Pressure
Empowering Evidence 2014
Heart
Attacks
Population: Middle Aged People in the with HBP
sample
HBP
treated
HBP
Not treated
20 heart
attacks
60 heart
attacks
Empowering Evidence 2014
Population: Middle Aged People in the with HBP
sample
HBP
treated
Get other
Interventions
e.g. aspirin
20 heart
attacks
Empowering Evidence 2014
HBP
Not treated
Don’t get
other
interventions
60 heart
attacks
Confounding
• A problem in even the most meticulously
conducted cohort study
• There are ways to mitigate its effects
– Have all the likely confounders been
identified?
– Have the authors used appropriate statistical
techniques for dealing with potential
confounders?
Empowering Evidence 2014
Confounding
• But it can never be totally ruled out in an
observational study (cohort, case-control)
• You can deal with the known confounders
– “control for”, “adjust for” them
• But you can’t deal with the unknown,
unmeasured ones
Empowering Evidence 2014
Confounding
• The only way to avoid confounding is to do
a randomized trial
– Randomization balances the known and
unknown risk factors evenly between the two
groups
• Treatment decisions (especially for
prevention) should be based on
randomized trial data
Empowering Evidence 2014
Cohort Studies
•
•
•
•
•
Overview
How they differ from Case Control Studies
Bias
Confounding
Characteristics of a GOOD cohort study
Empowering Evidence 2014
Features of Good Cohort Studies
• Sample representative and assembled at a
common point in time
• Follow-up sufficiently long and complete
• Outcome criteria objective or applied in a blinded
fashion
• Adjustment for possible confounding (prognostic
factors)
• Results reported with time to event curves (if f/u
longer than a few months)
• Precision of the effect size reported (CI)
Empowering Evidence 2014
Small Group Exercise
Glucose Levels and Risk of Dementia
Crane et al NEJM 2013; 369(6):540-548
Start by reading the abstract …Then try to answer these questions
•
•
•
•
•
•
•
What kind of study was this?
Was follow-up sufficiently long and complete?
Were the outcome criteria objective or applied in a blind fashion?
Was adjustment for prognostic factors (confounding) done?
Were outcomes reported over time (ie time to event analysis)?
How precise were the estimates of prognosis?
How would you apply the results of this study in your practice?
Empowering Evidence 2014
Download