Lec 9

Answer 1

Types of populations
Open or dynamic population
- Indiv can enter or leave anytime
- Membership is defined by characteristic that can change (eg smoking habits/status,
- Measure of association: INCIDENCE RATE (measure disease freq w/ PY denom)
Fixed cohort
- Has IRREVOCABLE event
- Defined start and end point
- Potential for loss to follow up
- Measure of association: incidence rate

Closed cohort
- Observation period is SHORT
o Eg: GI issues after the party
- Measure of association: cumulative incidence
o Why – no loss to follow up

Answer 2

Types of cohort studies
- Prospective: look forward in time
o Exposure happened; outcome – not happened
- Retrospective: look backward in time
o Exposure and outcome already happened
- Ambi-directional: look both ways

What type of study is needed
- Retrospective
o Diseases that take long time to dev and diagnose
o Disease and exposure happened
o Rely on records not intended for rs
- Prospective
o Collect detailed data on exposure and can control how data is collected
o Easier to follow-up
o Less prone to bias b/c the outcome has not occurred

Answer 3

Selection of study population
Special cohort
- Selected based on exposure
- Used when exposure is RARE
General cohort
- Population selected NOT RELATED TO EXPOSURE
- COMMON exposures (eg cigarettes, OH)
- Select from general group of ppl (healthy)

Types of Comparison group (internal > gen pop > external)
Internal
- unexposed members of the same cohort
- Procedure
o #1: select cohort (eg residents in town A)
o #2: determine who is exposed vs unexposed
o #3: then divide the cohort
General population
- Use existing data on outcome variable (ex disease, deaths) in the population
- Used when a comparison group CANNOT be assembled
- Used in occ studies of mortality
- Healthy worker effect
External
- If everyone in the cohort is EXPOSED (eg workers in the industry); look for an unexposed cohort that is very similar (eg workers in a neighboring industry)

Answer 4

Definition of exposure and outcome
Assessing exposure
- CLEAR definition of exposure
- Collect info from multiple sources
- Biological specimens (can be used to confirm self-reported exposure status)

Assessing outcomes
- Clear definition of OUTCOME
- Can use medical record to confirm self-reported outcomes

Answer 5

2: incidence rate

Analysis
Measures of disease frequency
- In cohort studies, we are generally interested in incidence (NOT prevalence)
#1: cumulative incidence (aka risk)
- # new cases of disease (over a period of time)/ # in study pop
- Measure of association (exposed vs unexposed)
o RR: Re/Ru
o RD: Re – Ru (aka AR)

• # new cases/P-time
• FIXED and DYNAMIC cohorts
• NOTE: CI = IR x t
• Measure of association
  o RR, RD (aka AR)
• 95% CI
  o Wider CI = less precise
  o Larger sample = narrower CI
  o Smaller sample = wider CI
  o Interpretation: 95 out of 100 times the CI will contain the true measure of association (or 95% confident)
  o NOTE: if CI contains “1”, result is NOT stat sig

Answer 6

Advantages of cohort studies
- Great to study rare EXPOSURES
Examine multiple outcomes
- Inefficient to study rare OUTCOMES
o To study a rare cancer, it may take 30 yrs to develop
- Prospective: high cost, time consuming
- Retrospective: need good records, and patient recall
- Potentially lose to follow up

Answer 7

Potential biases in cohort studies
Selection bias: When loss to follow up is related to both the exposure and disease
- Types of selection bias in a cohort study
o Loss to follow up
 Non-differential: equally likely to loss indiv in exposed and unexposed gp
 Differential: losses are more or less likely among exposed than unexposed
o Healthy worker effect: workers (mostly healthy) is compared to general pop (healthy & unhealthy ppl)
- How to avoid selection bias
o Use same methods to follow up exposed and unexposed ppl
o Minimize loss to follow up
o Collect info to locate subject
o Regular contact
o Send additional emails to non-respondents

Answer 8

Potential biases in cohort studies
Information bias
- Measurement flaw in exposure, covariate or outcome variables that leads to biases
- Leads to incorrect classification as exposed vs unexposed or disease vs not
- Types of info bias
o Recall bias: accuracy of info recalled differ b/w groups
 Common in case control studies, retrospective cohort
o Interviewer/detection bias
 Systematic difference in soliciting, recording, interpreting info
o Misclassification/ measurement bias
 Error in measuring exposure and or outcome -> misclassification of exposure or disease
 Differential (systematic), nondifferential (random)
- How to avoid info bias
o Study deisgn
 Mask interviewers, subjects
 Trained interviewers; have interview scripts
 Careful design of questionnaire
* Eg closed end qs
* Understandable
 Use multiple measurements and data sources
 Clear definitions of exposures and outcome