Cohort Studies

1
Cohort Studies

• Introduction to Epidemiology
• Sharon Cooper, Ph.D.
• June 12, 2001

2
Cohort Studies

RR (cumulative incidence)a/(ab)/c/(cd) cum.
inc. exposed/cum. inc. nonexposed
Presentation of data from a cohort study with
person-time denominators. RR Incidence Rate
Ratio a/PY1/c/PY0
3
Cohort Studies

• What is the measure of the strength of
  association in a cohort study?
• Relative risk based on cumulative incidence (CI)
  
• CIE/CINE a/(ab) / c(cd) risk ratio
• Relative risk based on incidence density or rate
  (IR)
• IRE/IRNE a/PY1 / c/PY0 risk ratio
• If RR 1 ? no association
• If RR gt 1 ? increased risk
• If RR lt 1 ? decreased risk

4
Example

• Cohort study of oral contraceptive use (OC) in
  relation to subsequent development of
  bacteriuria
• Information
• Initial survey for exposure in 1973
• 2,390 women aged 16-49 free of bacteriuria
• 482 of these were OC users, 1,908 were not
• Second survey in 1976
• 27 of OC users developed bacteriuria
• 77 of nonusers developed bacteriuria
• Construct 2x2 table, and calculate and interpret
  Relative Risk
• If you want to calculate 95 CI for RR
• 95 CI RR (11.96/?)
• In this example, ?22.27 and 95 CI (0.9-2.2)

5
Cohort--Definition

• A group or aggregate of persons (disease-free)
  who have presumed antecedent characteristics or
  experiences in common and who are followed
  throughout their experience to observe the
  development or nondevelopment of a given health
  outcome.
• Examples
• Persons born in the same year (birth cohort)
• Cohort exposed to radiation in 1945
• Persons working at Dow Chemical Co. (1940-89)
• Cohorts exposed to a particular drug (DES)
• Entire population (cross-classified on e.g.,
  levels of blood pressure, cholesterol,
  smokingFramingham Cohort)

6
Distinguishing CharacteristicsCohort Studies

• Measures the incidence of a disease in a group
• Looks for an association between two variables
  comparing the incidence of a disease between two
  or more groups. The groups are formed based upon
  the level of the exposure variable observed by
  the investigator.
• (can also look at change in a variable within
  each group)

7
Two Types of Study Populations

• 1. The study participants may represent a real
  population, which is subdivided into levels of
  exposure.
• Example Framingham
• OR
• The study participants may be selected because
  they represent exposed individuals and unexposed
  individuals
• Example Radiologists versus other internists

8
Cumulative Incidence vs. Incidence Density Study

• If you were conducting a 10-year study to
  determine the incidence of stroke due to
  hypertension, how would you account for the
  following
• The individual who is killed in a car crash in
  Year 1?
• The individual who is killed in a car crash in
  Year 9?
• The individual who moved away and left no
  forwarding address?

9
Design of a Cohort Study
Time
Direction of inquiry
disease
Exposed
no disease
People without the disease
Population
disease
Not Exposed
no disease
10

• Some important characteristics of a study
  population for a cohort study are
• Adequate variation in exposure
• Adequate incidence of disease
• Accessible
• Cooperative
• Capable of being followed for occurrence of
  disease
• DIRECTIONALITY
• Forward Always looks from exposure forward to
  disease

11
The Population

• The population consists of those individuals who
  are free of the disease of interest at the start
  of the follow-up period for which incidence will
  be calculated.
• Usually, this is at the start of the study.
• However, occasionally a run-in period will be
  placed at the start of the study to eliminate
  individuals who have the disease, but which is
  not yet detectable.

12
Design Issues in Cohort StudiesExposed Group

• Exposed group can be selected because of
• A higher probability of exposure to agent of
  interest, e.g., semiconductor workers and
  solvents
• Logistic advantages to the researcher, i.e.,
  easier to recruit and follow over long time
  period (e.g., union memberscommon exposure,
  large groups)
• Sources of exposure information
• Pre-existing records factory records medical
  records
• Direct measure environmental sampling (air,
  water, soil) biomarkers (serum, urine, hair)
• Laboratory/clinical measures (cholesterol/blood
  pressure)
• Interviews (smoking)

13
The ConductProspective Cohort Studies

• Exposures are defined at the start of the study
  and data on exposure is collected prior to the
  occurrence of disease.
• Diseases occur as the study progresses.
• ADVANTAGES Allows the investigator to determine
  how to measure and collect information on
  exposure and possible confounders.

14
Retrospective Cohort Studies(Historical Cohort
Studies)

• Both exposure and disease have occurred before
  study is initiated.
• Requires existence of records.
• Direction is still forward, starting with exposed
  who are not diseased and following for a given
  time period to see if they develop disease.
• More quickly done than a prospective cohort
  study.
• However, information on exposure or confounding
  variables not often available (e.g., smoking
  history, nutrient intake)

15
The AnalysisTest of Hypotheses

• The formal hypothesis under investigation can be
  stated as follows
• There is no association between the exposure and
  disease (i.e., the relative risk 1.0)
• The relative risk (risk or rate ratio) is
  calculated as the ratio of the disease incidence
  in the exposed to the disease incidence in the
  unexposed.
• Interpretation of the RR Exposed individuals
  have RR times the risk (rate) of developing
  disease Y (over a given period) than unexposed
  individuals

16
Cohort Study Strengths

• Is of particular value when the exposure is rare.
• Can examine multiple effects of a single
  exposure.
• Can elucidate the temporal relationship between
  exposure and disease.
• If prospective, minimizes bias in the
  ascertainment of exposure.
• Allows direct measurement of incidence of disease
  in the exposed and nonexposed groups.

17
Cohort Study Limitations

• Is inefficient for the evaluation of rare
  diseases.
• If prospective, can be extremely expensive and
  time consuming.
• If retrospective, requires the availability of
  adequate records.
• Validity of the results can be seriously affected
  by losses to follow-up.
• Validity of the results can be seriously affected
  by bias in the ascertainment of disease.
• Can be impractical when the exposure varies over
  time and is short-acting.

18
Measures of Public Health Impact

• If the exposure is associated with the disease
  and there is reason to think it is causal, the
  public health importance of the risk factor is
  measured by the risk difference and etiologic
  fraction.
• GENERAL QUESTION
• How many cases of Y (disease/outcome) are
  attributable to X (exposure)?
• OR
• How many cases of Y can be eliminated by
  eliminating exposure X?

19
Measures of Public Health Impact

• RISK (RATE DIFFERENCE) (also known as
  attributable risk) is a measure that quantifies
  the risk (rate) of disease in the exposed group
  considered attributable to exposure (removes risk
  due to other causes, i.e., the risk in the
  nonexposed)
• RD IE INE
• (IF RR 1, what is RD)?
• In cohort study of OC use and risk of
  bacteriuria
• RD 27/482 77/1908 0.01566 or
  1566/100,000/3 years
• Interpretation
• Excess occurrence of bacteriuria among OC users
  attributable to their OC use is 1566 per 100,000
  (3 years).

20
Measures of Public Health Impact (cont)

• ETIOLOGIC FRACTION (also known as attributable
  risk , attributable proportion, attributable
  fraction) estimates percentage of disease among
  the exposed group attributable to exposure, or
  the percentage prevented by removing exposure.
• EF RD/IE x 100
• 1566/100,000 / 27/482 27.96 or
• If OC use causes bacteriuria, approximately 28
  of bacteriuria cases among OC users can be
  attributed to OC use.

21
Measures of Public Health Impact (cont)

• POPULATION RISK DIFFERENCE (PRD) excess risk of
  disease in total study population attributable to
  exposure (also known as population attributable
  risk)
• PRD IP INE (incidence rate (risk) in total
  population minus incidence rate in unexposed)
• 104/2390 77/1908 316/100,000/3 years
• Algebraically equivalent and useful,
• PRD RD x PE (where PE proportion of exposed
  individuals in the population)
• 1566/100,000 x 482/2390 316/100,000/3 years
• Therefore, if OC stopped, excess incidence of 316
  per 100,000/3 years would be eliminated.
• NOTE Prevalence of exposure has to represent
  correct prevalence in some target population,
  i.e., if select 100 exposed and 100 unexposed,
  PRD using this would be meaningless.

22
Measures of Public Health Impact (cont)

• POPULATION ETIOLOGIC FRACTION - of cases in
  total study population attributable to exposure
  (also known as population attributable fraction,
  population attributable risk )
• PEF PRD/IP x 100
• (IP INE) / IP x 100 (more formulas to come)
• IP 104/2390 or 4351.5 per 100,000 ?
• PEF 316/100,000/ 4351.5 per 100,000 x 100
  7.3
• Therefore, if OC use causes bacteriuria,
  approximately 7 of all bacteriuria in the study
  population could be prevented if OC use were
  eliminated.

23
Measures of Public Health Impact (cont)

• EXPOSED ONLY --- RISK (RATE) DIFFERENCE (RD)
• Among smokers, how much of the mortality they
  experience is due to their smoking?
• TOTAL POPULATION POPULATION RISK (RATE)
  DIFFERENCE (PRD) (Exposed Non-Exposed)
• How much of the mortality in the general
  population is attributed to smoking?
• PRD is usually used for public health
  implications, and intervention planning.

24
Measures of Public Health Impact (cont)

• Relative risk and risk difference of mortality
  from lung cancer and coronary heart disease among
  cigarette smokers in a cohort study of British
  male physicians

R. Doll and R. Peto, Mortality in relation to
smoking Twenty years observations on male
British doctors. Br Med J 1976 21525.
25
MeasuresCohort Studies
26
Pop Quiz

• When would you want to use Relative Risk?
• When would you want to use Risk Difference
  (Attributable Risk)?