Introduction to pharmacoepidemiology 1 Measuring...
Transcript of Introduction to pharmacoepidemiology 1 Measuring...
Introduction to pharmacoepidemiology 1 Measuring occurrence and associationThe cohort study
Morten AndersenCentre for PharmacoepidemiologyKarolinska Institutet
May 11, 2011Morten Andersen 2
May 11, 2011Morten Andersen 3
Clinical pharmacology Epidemiology
Pharmaco-epidemiology
Pharmacoepidemiology
The study of the use of and the effects of drugsin large numbers of people
Brian L Strom, 1994
May 11, 2011Morten Andersen 4
Pharmacoepidemiological methods
Case reports Case series, clusters Spontaneous reports Ecologic studies Specialised surveillance Case-control studies Self-controlled designs Cohort studies Randomised trials Evidence
Signal or hypothesisgeneration
Risk or effectivenessassessment
May 11, 2011Morten Andersen 5
Overview
What is a cohort? Closed and open cohorts Measures of incidence Measures of prevalence Introduction to the cohort study Measures of association Time varying exposure
May 11, 2011Morten Andersen 6
What is a cohort?
A population followed during a period and observed for one or more outcomes
Can be divided into closed and open/dynamic populations
May 11, 2011Morten Andersen 7
The closed cohort
A group of persons defined at a certain point in time Each person is followed until an event occurs or to the end of
the observation period No entry of new individuals
Note: Cohorts can be identified both prospectively and retrospectively (historically)
Example: The Thule workers who participated in the clean-up after the B-52 crash in 1968
May 11, 2011Morten Andersen 8
The closed cohort
Time
Person
1234
May 11, 2011Morten Andersen 9
The closed cohort
Time
Person
1234
Outcomeevent
End of observation,censoring
May 11, 2011Morten Andersen 10
The closed cohortProblems
Persons censored(observation ends because of incomplete follow-up, migration, death – if not an outcome)
Decreasing population size
Ageing of cohort
May 11, 2011Morten Andersen 11
The open/dynamic cohort
A population changing over time
Persons can enter or exit during the observation period
Example: Drug users during a certain observation period
May 11, 2011Morten Andersen 12
The open/dynamic cohort
Time
Person
1234
May 11, 2011Morten Andersen 13
Person – time – event
Persons under observation (population at risk) Time during observation (risk time)
how long observation period(number of events)
which time axis(calendar time, time from start of study, age)
end of observation – censoring(death, migration, other outcomes)
Event according to defined criteria
May 11, 2011Morten Andersen 14
Measures of occurrence, which one to choose?
Prevalence rate
May 11, 2011Morten Andersen 15
Cumulative incidence proportion
Cumulative incidence proportion at time t =
number of new cases until time tnumber of persons at the start of the period
Dimensionless, range 0-1
Example: The 30-day mortality for persons admitted to hospital with MI is 20%
May 11, 2011Morten Andersen 16
French-Russian war 1811-1812
May 11, 2011Morten Andersen 17
Cumulative incidence proportion
Example: The cumulative incidence of dying in the French-Russian war after one year was
(422,000-10,000) / 422,000 = 0.98 = 98%
What do we usually call the cumulative incidence proportion?
May 11, 2011Morten Andersen 18
French-Russian war 1811-1812
May 11, 2011Morten Andersen 19
Cumulative incidence proportion
Time, years0 1 2 3
Cumulative incidence proportion
at 3 years =
May 11, 2011Morten Andersen 20
Cumulative incidence proportion
0 1 2 3
Cumulative incidence proportion
at 3 years =
3/8 = 37.5%
Time, years
May 11, 2011Morten Andersen 21
Cumulative incidence proportion
0 1 2 3
Cumulative incidence proportion
at 3 years =
3/8 = 37.5%
Time, years
May 11, 2011Morten Andersen 22
Cumulative incidence proportion
0 1 2 3
Cumulative incidence proportion
at 3 years =
3/8 = 37.5%
Lost to follow-up
Time, years
May 11, 2011Morten Andersen 23
Incidence rate
Incidence rate =
number of new eventstime during which events are observed
Dimension time-1, e.g. per year, range 0-∞
Example: The incidence of upper GI bleeding in DK is 50 per 100,000 person years
May 11, 2011Morten Andersen 24
One person year is
One person followed for one year Two persons each followed for 6 months Three persons each followed for 4 months 100 persons each followed for 3.65 days 10 persons followed for 1 month and 60 persons followed for 1
day …
May 11, 2011Morten Andersen 25
Incidence rate
0 1 2 3Time, years
May 11, 2011Morten Andersen 26
Incidence rate
0 1 2 3
Events Time
1 1.5
1 2.5
0 2
0 3
1 2
0 2
0 3
0 2
IR = 1 per 6 person yearsTime, years
May 11, 2011Morten Andersen 27
Prevalence proportion
The proportion of a population who have the disease at a certain point in time (point prevalence, cross-sectional)
Dimensionless, range 0-1
Example: The prevalence of beta-blocker use among persons with a previous MI is 30%
May 11, 2011Morten Andersen 28
Prevalence proportion – Illustration
0 1 2 3
Use ofbeta-blocker
MI
Prevalence proportion at
1 year =
Time, years
May 11, 2011Morten Andersen 29
Prevalence proportion – Illustration
0 1 2 3
Use ofbeta-blocker
Prevalence proportion at
1 year =
2/8 = 25%
Time, years
MI
May 11, 2011Morten Andersen 30
Period prevalence
The proportion of a population who are affected by the disease during the observation period (e.g. 1-year prevalence)
Dimensionless, range 0-1
Example: 4% of the population used insulin during 1993 Note: Mixes up prevalent and incident users, the result strongly
depends on length of the period
May 11, 2011Morten Andersen 31
Period prevalence – Illustration
0 1 2 3
Use ofbeta-blocker
Period prevalence during the 1st year =
Time, years
MI
May 11, 2011Morten Andersen 32
Period prevalence – Illustration
0 1 2 3
Use ofbeta-blocker
Period prevalence during the 1st year =
4/8 = 50%
Time, years
MI
May 11, 2011Morten Andersen 33
Exercise 1Measures of occurrence Which epidemiological measures are/can be used? During the influenza epidemic in 1997, 25% of the pupils in one
Copenhagen school were absent on February 1st 15% of pregraduate medical students left the study during the
1st year During 2003, 300 new cases of breast cancer were diagnosed
among 100,000 women aged 50-54 years in DK
May 11, 2011Morten Andersen 34
The cohort study
”the delineation of a group of persons who are distinguished in some specific way from the majority of the population and observation of them for long enough to allow any unusual morbidityor mortality to be recognised”
Richard Doll 1964
May 11, 2011Morten Andersen 35
Cohort studies
Individuals are included based on exposure status, e.g. +/- drug use
The occurrence of outcome events among the exposed individuals is compared to the occurrence of events among the unexposed individuals
Overall aim to assess the association between exposure and outcome
May 11, 2011Morten Andersen 36
The cohort study
Time, years0 1 2 3
Exposed
Unexposed
May 11, 2011Morten Andersen 37
Risk ratio
Risk among exposed = 2/4 = 0.5Risk among unexposed = 1/4 = 0.25Risk ratio (“relative risk”) = 2.0Risk difference = 0.25
DiseaseExposure + - Total+ 2 2 4- 1 3 4Total 3 5 8
May 11, 2011Morten Andersen 38
The cohort study
Exposed
Unexposed
0 1 2 3Time, years
May 11, 2011Morten Andersen 39
Incidence rate ratio
Rate among exposed = 0.2 per yearRate among unexposed = 0.1 per yearRate ratio = 2.0Rate difference = 0.1 per year
DiseaseExposure + - Person time+ 2 2 10- 1 3 10Total 3 5 20
May 11, 2011Morten Andersen 40
Time varying exposure
Time, years0 1 2 3
Use ofNSAID
Person timeExposed Unexposed
IRR = =
Non-use
May 11, 2011Morten Andersen 41
Time varying exposure
0 1 2 3
Use ofNSAID
Person timeExposed Unexposed
0.5 1
0 2
1.5 1.5
0.5 1
0 1.5
1.5 1
0 1.5
0 2.5
IRR = = 3.02/42/12
Non-use
Time, years
May 11, 2011Morten Andersen 42
Exercise 2Measures of associationThe incidence of hospital admission for ulcer per 1,000 person yearsadjusted for baseline demographic characteristics, co-morbidity, previousulcer and medication use
Based on Ray et al. Gastroenterology 2007;133:790-8
Person years Incidence rate
NSAIDs without ulcer prophylaxis 57,032 5.65
NSAID + proton pump inhibitor 6,227 2.57
Coxib without ulcer prophylaxis 13,962 3.38
May 11, 2011Morten Andersen 43
Exercise 2Measures of association Which study design is used? Which measures of frequency and association are relevant? How much is the risk of an ulcer increased using traditional
NSAIDs compared to coxibs, both without ulcer prophylaxis? How many hospital admissions due to ulcer can be avoided
By changing from NSAID to coxib? By supplemeting NSAID with a proton pump inhibitor?
May 11, 2011Morten Andersen 44
NSAIDs, coxibs and ulcer(Ray et al. Gastroenterology 2007;133:790-8)
May 11, 2011Morten Andersen 45
Further considerations related to exposureand outcome Point or continuous exposure Acute or chronic effects Dose-related effect Cumulative effect Latency period Biological evidence (or plausibility) for
exposure-outcome relation Outcome includes first event only or multiple events Time required for defining ”new” incident event
May 11, 2011Morten Andersen 46
Experimental versus observational research
Experimental research (randomised clinical trial) Randomisation ensures comparability of patients Study the effect of exposure isolated from other causal factors Procedures to ensure complete and valid data collection
Observational research (epidemiological study) Patient groups are fundamentally incomparable Multiple causal factors are acting simultaneously Often incomplete and imperfect data collected for other purposes
May 11, 2011Morten Andersen 47
False adverse effects
May 11, 2011Morten Andersen 48
Confounding – mixing of effects
Patient factors become confounders if they are associated with the exposure and also independent predictors of the outcome
Confounder
OutcomeExposure
May 11, 2011Morten Andersen 49
Confounder kontrol
DESIGN Randomisation Cross-over Restriction Matching
ANALYSIS Stratification Standardisation Multivariable analysis Propensity scores Instrumental variables
May 11, 2011Morten Andersen 50
Ensuring that differently treated patients arecomparableCausal experiment (rewind time, perfectly comparable)
Patient 1: Drug Change in outcome
Patient 1: Placebo No change
Clinical trial (randomize, on average comparable)
Patient 1: Drug Change in outcome
Patient 2: Placebo No change
May 11, 2011Morten Andersen 51
Ensuring that differently treated patients arecomparableObservational study
Patient 1: Drug Change in outcome
Patient 2: No drug No change
How do we select patient 2 to be comparable with patient 1?• A patient that looks exactly like patient 1 with regard to
predictors of outcome (confounders, disease risk score)• A patient that looks exactly like patient 1 with regard to
choice of treatment (propensity score)
May 11, 2011Morten Andersen 52
Two ways to address confounding in the analysis
Confounder
OutcomeTreatment