11/20091 epi 5240: introduction to epidemiology cohort studies: a return november 23, 2009 dr. n....
TRANSCRIPT
11/2009 1
EPI 5240:Introduction to Epidemiology
Cohort studies: A returnNovember 23, 2009
Dr. N. Birkett,Department of Epidemiology & Community
Medicine,University of Ottawa
11/2009 2
John Snow, 1860:
According to a return which was made to Parliament, the Southwark and Vauxhall Company supplied 40,046 houses from January 1 to December 31, 1853, and the Lambeth Company supplied 26,107 houses during the same period; consequently, as 286 fatal attacks of cholera took place, in the first four weeks of the epidemic, in house supplied by the former company, and only 14 in houses supplied by the latter, the proportion of fatal attacks to each 10,000 houses was as follows: Southwark and Vauxhall 71, Lambeth 5. The cholera was therefore fourteen times as fatal at this period, amongst persons having the impure water of the Southwark and Vauxhall Company as amongst those having the purer water from Thames Ditton.
11/2009 3
Some issues with Snow’s report:
• Dynamic populations.• Sort of a ‘rate’ since the denominator is
‘family-years’.– Are families the same size in the two cohorts?– Lack of control for age differences.– Other confounders.– SES factors.– mortality vs. disease incidence.– Validity of exposure allocation
• people have S&V water linkage but use water from a well– this is happening now with ‘bottled water’.
11/2009 4
General types of Study Populations (1)
• Type #1– Recruit subjects at a fixed point in time (eg. people
exposed to the Hiroshima atomic bomb).– Population involve a fixed list of people– Follow-up is complete on everyone.
• Type #2– Recruit subjects at a fixed point in time (eg. people
exposed to the Hiroshima atomic bomb).– Population involve a fixed list of people– Some people are lost-to-follow-up (drop-outs,
censored)
11/2009 5
General types of Study Populations (2)
• Type #3– Recruit subjects over time (eg. recruit people newly diagnosed
with HIV starting in Jan, 2003 and ending in Dec. 2005)– Population involve a fixed list of people– Follow-up may or may not be complete on everyone.
• Type #4– Define a cohort based on criteria such geography, etc. (eg.
people living in the boundaries of Ottawa-Carleton).– Follow-up the group over time.– You DO NOT have a list of who is in the group.– People will move in and out of the group during follow-up.
11/2009 6
General types of Study Populations (3)
TERMINOLOGY
Fixed: The membership of the group is known at the ‘start’; in-migration is not allowed. This definition is sometimes extended to require people to remain in the same exposure group;
Closed: A ‘fixed’ group in which no-one is allowed to leave (ie. follow-up is 100% complete).
Open: A group in which people are allowed to both enter and leave during follow-up.
Dynamic: The same as ‘open’.
Cohort vs. population/group– The term ‘cohort’ is usually reserved for groups which are ‘fixed’ while
‘population’ is used when membership is open.
Needs definition
11/2009 7
General types of Study Populations (4)
Type #1: Closed cohort (fixed)
Type #2: Fixed cohort
Type #3: Fixed cohort
Type #4: Dynamic (open) population
11/2009 8
KEY ISSUES (1)
• Defining cohort membership– ALL cohorts will have losses and late entry. Therefore, model for cohort
is based on person-year rather than persons
• Logistics of retaining participation and obtaining the information• Exposure measurement, not only the practicalities but also the
theory of what to measure.– How to account for time variability of exposure?
– Simplifying assumption: use a simple numerical summary (eg. average, peak, cumulative, lagged).
• CAN BE BIASSED.
• Person-time allocation.• Outcome assessment
11/2009 9
EXPOSURE ISSUES (1)
• Depends on your hypothesis. Should be as SPECIFIC as possible.– Acute exposure to ionizing radiation will increase risk
of colon cancer, starting 5 years after exposure.
• Time for accumulating exposure VS Time at risk for outcomes.– Lags (induction period)– Changes in risk with time.
• Acute vs chronic exposure.
11/2009 10
EXPOSURE ISSUES (2)
Atomic bomb exposure
• Acute exposure (almost a point source in time)
• No risk of cancer in first few months/years
• A simple ID will give average effect and thus will under-estimate effect.– Look at risk in various post-exposure time
windows.
11/2009 11
ID
Time
11/2009 12
EXPOSURE ISSUES (4)
CHRONIC EXPOSURES (e.g. smoking)• Exposures can be:
– Fixed• Occur once
– Atomic bomb exposure• Have a single onset but then remain as an exposure
– Fluoride added to drinking water (fixed levels)– Dixon from a spill (decaying levels)
– Time varying• Start/stop
– Smoking, drugs, work exposures• Continuous but level varies
– Diet– Background & continuous
• radiation
11/2009 13
EXPOSURE ISSUES (5)
CHRONIC EXPOSURES (e.g. smoking)• Fixed exposures
– Can be measured once at baseline– Assumed to not change during follow-up– Defines exposure groups– Suitable for count data but also person-time data.
• Time varying exposures– Requires multiple measurements (at baseline and during follow-
up) to determine change in exposure status– Only usable in person-time studies– Need a rule to determine whether person-time/outcomes
between measures accumulates to ‘exposed’ or ‘non-exposed’ group.
11/2009 14
EXPOSURE ISSUES (6)
• PY’s accumulate in exposure group at the time. Assign outcome to exposure group which is accumulating PY’s at time of event.
• Need to consider induction time.– Delay from exposure to earliest onset of any disease due to
exposure.• Atomic bomb exposure on January 1, 2000• Leukaemia diagnosed on January 2, 2000• CAN NOT be causally linked.
– Common to assume induction time is ‘0’ in the absence of detailed information
• True in most situations.
• Analogous thing happens when exposure is stopped– Risk continues for some time
• and may not decline to ‘0’ (another issue)
11/2009 15
EXPOSURE ISSUES (6)
• ‘Unexposed’ time accumulates in exposed group.– prior to end of induction period (after 1st exposure)– after end of ‘at risk’ period (when exposure stopped).– during induction/risk period for exposure changes during follow-
up• Example
– ‘moderate’ smoking is 20-40 pack-years with 5 year induction time.
• Person remains in ‘mild’ smoking group until 5 years after acquiring 20 pack-years.
• Immortal Person Years– Follow-up time during which a subject can never get the
outcome (because of the study design)• subjects were recruited two years after starting to work• Cohort starts at time of first employment• Everyone must have survived at least two years
11/2009 16
EXPOSURE ISSUES (7)
CHRONIC EXPOSURES (cont.)• Types of exposure measures
– Ever/never– Maximum level– Average level– Cumulative Exposure
• Pack-years (smoking cumulative exp measure)– Add up number of cigarettes smoked in life-time;– Divide by 7305 (1 P-Y = 20 cigs/pack * 365.25 days)– Composite measure of duration and intensity.– Assumes:
0.5 packs/day x 20 years = 4 packs/day x 5 years • Consider separate duration and intensity variables.
11/2009 17
Change to overhead file
11/2009 18
EXPOSURE ISSUES (8)
Post-exposure events• Objective: Effect of quitting smoking on CHD risk.
– What to do with someone who quits and then starts smoking again?
– Censor at time of recidivism• Objective: Effect of having stopped smoking for five (or
more) years)?– Include only people with 5+ years of time after stopping
smoking.– Treat first five years post-stopping as Immortal PY’s.
Criteria for exposure grouping for continuous exposure• Quintiles or quartiles are common• Can base on distribution in cohort or on external data
about exposure risk.
11/2009 19
Outcome issues
• Timing of event
• Definition of outcome
• Logistics of determining if (and when) an outcome occurred.
11/2009 20
Logistic Issues
Expense• Related to large size and induction period.• Worse with rare outcomes and long induction periods.• Try to reduce cost by:
– Using administrative system to detect outcome events;– Historical designs;– Use the ‘general population’ as the unexposed cohort;– Nested case-control study.
• Tracking subjects– Need at LEAST 60% follow-up with 80% (or higher) being the
target.• Should you change measurement methods during follow-
up?
11/2009 21
Keeping participation up
• Subject’s need to develop committment with study (identify with it)– Regular contacts– Newsletters– Web site– Paraphernalia– Highlight key information, papers, awards,
contributions to policy• Contact name, address, e-mail,etc.
– Up-date regularly– Obtain contact info for at least one friend/relative– Professional organization, church, etc.
11/2009 22
Estimating ‘Incidence’ (1)
• Fixed Population (cohort)– Select all members at a Pre-determined point
• Jan 1, 2010• 1st diagnosis of disease
• Type A:– Equal, known follow-up in ALL subjects
• Type B– Variable but known follow-up
• Some lost-to-follow-ups (censored)• Variable time of entry
• Type C– Variable F/U and some people have unknown exact length of
follow-up
11/2009 23
Estimating ‘Incidence’ (2)
• Dynamic Population– Does not start with a pre-determined, named, fixed group
• MONICA study– People living in Halifax between 1980 and 1990– Not the same people each year.
• Type A:– Variable but known follow-up times
• Really is a type C fixed cohort
• Type B– Variable unknown exact length of follow-up for individual
subjects• The ‘classic’ dynamic population• Cancer incidence studies
11/2009 24
Switch to overhead slides
11/2009 25
• How much lung cancer can be attributed to smoking?
• Measure of exposure IMPACT rather than strength.
• There are many AR measures, often with similar names. Makes things confusing. One book used the same abbreviation for 4 different measures in five pages!
Attributable risks (1)
11/2009 26
• In exposed subjects
Attributable risks (2)
ExpUnexp
RD or Attributable Risk
Iexp
Iunexp
RD = AR = Iexp - Iunexp
Iexp – Iunexp
AR(%)=AF= -----------------------
Iexp
11/2009 27
• In the exposed group, the impact of the exposure on outcome depends on RR only:
1– AR(%) = 1 - -------- RR
• A value of ‘0’ shows no impact.
• Shouldn’t compute AR’s unless causation has been established.
Attributable risks (3)
11/2009 28
• My actual question was: in the general population, how much lung cancer was due to smoking?
• Depends on two factors:– Strength of the smoking/lung cancer
relationship (RR).– How common smoking is in the population
(exposure prevalence).
Attributable risks (4)
11/2009 29
Attributable risks (5)
ExpUnexp
Attributable Risk,population
Iexp
Iunexp
Population
Ipop
11/2009 30
Switch to overhead slides