introduction to cohort studyhlm.tzuchi.com.tw/.../2019/20190412_cohort-study.pdf · a cohort study...
TRANSCRIPT
Learning objectives
• To know what is cohort
• To know the characteristics of cohort study
– Incidence
– Person-year
– Relative risk/risk ratio/rate ratio
• To know what is the ideal comparison
group
• To know the bias in cohort study
• Advantages & disadvantages
Idea
(hypothesis)
Study
design
Statistical
Analysis
Experimental design
• Randomized controlled trial
Non-experimental (observational) design
•Cohort study (世代追蹤研究)
•Case-cohort study
•Nested case-control study
•Case-control study (病例對照研究)
•Case-crossover study
•Case-time control study
•Cross sectional study (橫斷性研究)
資料收集
From idea to research
Test the hypothesis
Environmental exposure
or host characteristics
Disease or other health
outcome
1. An association is
observed2. Is the observed
association causal?
Population
Not
Exposed
Exposed
Randomly
Allocated
Population
Not
Exposed
Exposed
Self-
Selection
RCT vs. Cohort study
What is cohort?
Cohort studies:
marching towards outcomes
Cohort Study
Follow up
Exposed Not
Exposed
Disease No
disease
Disease No
disease
Defined population
Self selection
Retro.
1992
2018
Prosp.
2018
2032
Exposure Outcome
1. 有A暴露的人,未來得B疾病的發生率(incidence rate)或危險性 (risk)?
2. 有A暴露的人,未來罹患B疾病的發生率(或危險性)是沒有A暴露的人的幾倍,也就是說,有A暴露的人相對於沒有A暴露的人,其相對危險性(relative risk)為何?
HBsAg status PHC
The framingham study
Exposed Not Exposed
Disease No
diseaseDisease No
disease
Defines population
(a town in Massacusetts, 30~62 year-old)
Self selection
Prosp.
1948
Blood pressure, smoking, body
weight, diabetes, exercise, etc.
Coronary heart disease, stroke,
congestive heart failure, peripheral
arterial disease, etc.
Cohort Study
Exposed
people
Develop
DiseaseDo Not
Develop
Disease
Non-Exposed
people
Develop
DiseaseDo Not
Develop
Disease
If exposure is associated with disease:
A cohort study can provides..
• Incidence: the number of newly diagnosed cases of a disease.
• Risk is the number of new cases of a disease divided by the
number of persons at risk for the disease.
• Incidence rate is the occurrence rate of new cases of a disease
in defined population during a specific time period. (person time
at risk)
# of new eventspersons at risk at begining of study
# of new eventsperson-time at risk
Person-years (人年)
Subjects Time
at risk
A 8.3
B 11.0
C 14.0
D 14.0
E 10.2
F 3.0
G 12.0
H 7.0
I 10.0
J 3.0
K 9.0
L 6.2
81 95
Loss to f/u
disease developed
Incidence ratesubjects Time
at risk
A 8.3
B 11.0
C 14.0
D 14.0
E 10.2
F 3.0
G 12.0
H 7.0
I 10.0
J 3.0
K 9.0
L 6.2
Total time at risk =107.7 person-yrs
Incidence density (0~infinity) :
= 3/107.7
=0.028/person-yr
= 28/1000 person-yr
= 2.33/1000 person-month
= 0.54/ 1000 person-weeks
Unit depends on investigators, frequency of
event.
Cumulative incidence (risk) (0~1):
= # of new disease/ initial population
=3/12
Design of a cohort Study
Disease
developed
Disease
Not
developed
Total
Person-yr
At risk
Total
subjects
Incidence
rates of
Disease
Exposes a b T1 a+ba/(a+b) or
a/T1
Non-
exposed c d T2 c+dc/(c+d) or
c/T2
Association measure in cohort study
exp
non-exp
risk
risk
a
a bRRc
c d
當每人追蹤時間相同:
exp 1
non-exp
2
I
I
a
TRR
c
T
當每個人追蹤時間不同:
1
2
Rate difference:Iexp –Inon-expRisk difference: riskexp-risknon-exp
Interpretation RR of a Disease
RR Interpretation of RR
=1 Risk in exposed equal to risk in non-exposed
(no association)
>1 Risk in exposed greater than risk in non-
exposed (possibly causal)
<1 Risk in exposed smaller than risk in non-
exposed (possibly protective)
Incidence rate & person years
BMI #
MIs
Person-years
at risk
Rate of MI per
100,000 person-Yrs
(incidence rate)
Crude
RR
<21 41 177,356 23.1 1.0(ref)
21-23 57 194,243 29.3 1.3
23-25 56 155,717 36.0 1.6
25-29 67 148,541 45.1 2.0
> 29 85 99,573 85.4 3.7
Type of cohort• Dynamic cohort (Open cohort)
– Exposure status may change during follow up
– Subjects may enter the study at any time
– The cohort is defined by person-times rather
than on persons.
• Fixed cohort
– No new subjects enter the study after study
follow-up date
– Exposure status is consistent
– If there is no loss to follow-up: closed cohort
CHOICE OF STUDY POPULATION
CHOICE OF COMPARISON GROUP
Internal and external
The ideal comparison group
臨床角度問問題: Do patients who receive an atypical
antipsychotic drug have an increased risk of hip fracture?
Cohort study角度問問題: What would have happened to
these patients if they had not received the atypical
antipsychotic drug?
Ideally, the comparison group in the cohort study
should be identical to the intervention group, apart
from the fact that they did not receive the intervention
Internal comparison group
• That is, the experience of those members
of the cohort who are either unexposed or
exposed to low levels can be used as the
comparison group.
External comparison group
• When the cohort is essentially homogeneous in
terms of exposure to the suspected factor, a similar
but unexposed cohort, or some other standard of
comparison, is required to evaluate the experience
of the exposed group.
– People in employment from the same geographical area .
– General population of the geographical area in which the
exposed individuals reside.
Multiple comparison groups• When we can’t be sure that any single group
will be sufficiently similar to the exposed
group in terms of the distribution of potential
confounding variables.
• The study results may be more convincing if
a similar association were observed for a
number of different comparison groups.– an internal comparison groups (same factory but having different
job) and the experience of general population (national and local
rates) may be used.
Possible types of comparisons in
cohort study
Ex: association between antipsychotic drugs and hip
fracture.
General population (all elderly)
- intervention vs. alternative intervention
- intervention vs. no intervention
Restricted population (elderly people with dementia)
- intervention vs. alternative intervention
- intervention vs. no intervention
BIAS
Reliable (precise)
Lack of random error
Valid
Lack of systematic errorReliable
& valid
Bias is a systemic error, rather than the random variation or
lack of precision.
Bias occurs in the recruitment of participants, the measurement
of their risk factors and outcomes.
Bias threatens study validity
Internal validity
The internal validity of a study is defined as the
extent to which the observed difference in
outcomes between the two comparison groups can
be attributed to the intervention rather than other
factors.
Bias of Cohort Study
• Selection Bias
– Attrition bias
– Healthy entrant effect
– non-response bias
• Information Bias
– Response bias
– Acertainment bias (detection bias)
• Confounding
– The effect or association between an exposure
and outcome is distorted by the presence of
another variable
Selection bias
• Selection bias occurs when there is
something inherently different between the
groups being compared that could explain
differences in the observed outcomes.
• Loss to follow-up differs between exposed
and not exposed (or between disease and
no disease).
• Follow up is usually easier in people who
have been exposed to the exposure of
interest.
Information bias• Collect different quality of information from
exposed and not exposed groups
(from participants or investigators).
– Exposure ascertainment (response bias)
under-reported the exposure behavior because
they are aware that it can affect their health.
– Diseases ascertainment (ascertainment bias,
detection bias)
diagnosis could have been influenced by
knowledge of the study research hypothesis.
Confounding
Exposure
Confounder
Disease
The effect or association between an exposure and
outcome is distorted by the presence of another variable
ex: asthma and lung cancer (smoking is the confounder)
Confounding
• Features of medical history—for example,
stroke, osteoporosis
• Exposure to drugs—for example,
benzodiazepines,oestrogens
• Demographics—for example, age and sex
• Social and behavioural factors—for example,
exercise and diet.atypical
antipsychotic
C
Hip fracture
Differences in distribution of potential
confounders
To deal with confounding
Idea
(hypothesis)
Study
design
Statistical
Analysis
Matching
RestrictionRandomization
Stratification
Regression
Matching & Restriction
a 45-yr-old women with atypical antipsychotic
a 45-yr-old woman with no intervention
Matching
Restriction
ex: Only the elder are recruited
ex: Only the older with dementia are recruited
E
C
D
Patients taking atypical antipsychotics were over
12 times more likely (63.1% vs. 4.7%) to have
dementia.
Restriction
Regression &
stratification
ANALYSIS METHODS
Data preparation
Analysis methods• Cox proportional hazard regression (HR)
– Time to an event (ex. time to hip fracture)
– person-time information
– time-varying covariates
• Poisson regression
– Count data
• Logistic regression (RR)
– Binary outcome (ex. occurrence of hip fracture)
– No loss to follow up
– Rare outcome
– Person-time information not required
Interpretation of Hazard Ratio
• HR = 0.5: at any particular time, half as many
patients in the treatment group are
experiencing an event compared to the control
group.
• HR = 1: at any particular time, event rates are
the same in both groups,
HR = 2: at any particular time, twice as many
patients in the treatment group are experiencing
an event compared to the control group.
Examples
Exposed Not Exposed
Disease No
diseaseDisease No
disease
Defines population
(40 clinical centers in USA,
50-79 yr-old postmenopausal women)
Self selection
Prosp.
1993~
1998
Active or passive smoking
Invasive breast cancer
2009/8/14
Inclusion:
93679 women
aged 50~79 yr-
old were
recruited
Exclusion:
Had conditions
that were
predictive of
survival less
than 3 years
Final subjects
for analysis:
93676 -> 79990
1
2
Which bias might the above cohort
study have been prone to?
(a) Non-response bias
(b) Healthy entrant effect
(c) Attrition bias
(d) Response bias
(e) Confounding
(f) Allocation bias
93 676
12075 Cancer Hx.
(12.9 %).
(healthy entrant
effect)
443 loss to f/u
(0.5 %).
(attrition bias)
1168 smoking status
missing (1.25%).
Under-report the
exposure
(response bias)
79 990
Non-response bias
(a) Non-response bias
(b) Healthy entrant effect
(c) Attrition bias
(d) Response bias
(e) Confounding
(f) Allocation bias
Exposure
(smoking)
Confounder
(Alcohol intake)
Disease
(breast cancer)
Allocation bias is mainly of concern in clinical trials.
(to allocate people who they think would show the greatest
benefit to a particular intervention)
Which bias might the above cohort
study have been prone to?
Advantages
•Measure the effect of each variable on
the probability of developing the outcome
of interest (RR or HR).
• A single study can examine various
outcomes.
Example: smoking vs. lung, cardiovascular, and
cerebrovascular diseases.
Disadvantages
• Expensive and time-consuming.
• Inefficient when outcome is rare.
• Loss to follow up can be a serious
problem. The rarer the outcome the more
significant the effect. (Attrition bias)
Thank you