Epidemiology – Cohort studies I March 2010 Jan Wohlfahrt Afdeling for Epidemiologisk Forskning Statens Serum Institut.

Epidemiology – Cohort studies IEpidemiology – Cohort studies I

March 2010March 2010

Jan WohlfahrtJan WohlfahrtAfdeling for Epidemiologisk ForskningAfdeling for Epidemiologisk Forskning

Statens Serum InstitutStatens Serum Institut

EPIDEMIOLOGYEPIDEMIOLOGY

COHORT STUDIES ICOHORT STUDIES I

March 2009 (modified)March 2009 (modified)

Søren FriisSøren FriisInstitut for Epidemiologisk KræftforskningInstitut for Epidemiologisk Kræftforskning

Kræftens BekæmpelseKræftens Bekæmpelse

”While the individual man is an insoluble puzzle, in the

aggregate he becomes a mathematical certainty. You

can, for example, never foretell what any one man will

do, but you can say with precision what an average

number will be up to”

Arthur Conan Doyle

Sherlock Holmes: The Sign of four

Ideal study of a causal effectIdeal study of a causal effect

”The experience of exposed people is compared

with their experience when not exposed, while

everything else is held constant”

Kenneth Rothman, Modern Epidemiology, 1998

Assignment of exposure

Non-experimentalstudies

Randomised/intervention

trials

Sampling accordingto exposure status

Sampling according to outcome status

Cohort studies

Case-control studies

Yes No

no yes

Analytic epidemiological studies Analytic epidemiological studies

Experimental studies

Non-experimentalstudies

Random allocation

Community intervention

trials

Cohort studiesCohort studies

Classical definitionClassical definition

”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 morbidity or mortality to be recognised”

Richard Doll 1964


Recent definitionRecent definition

Experiments Randomised clinical trials

two (or multiple)-arm, cross-over

Field trials intervention on single-person level

Community intervention trials intervention on community level

Non-experimental cohort studies

Past Present Future

Identify study subjects and assess exposure characteristics

Follow-up

Population at risk

Exposed

Non-exposed

Udfald

+

-+

-

Censored

Censored

Population at riskPopulation at risk

Individuals at risk of developing the outcome(s) of interest Basis for computation of measures of diseases frequency and effect

measures

Classified according to exposure characteristics At baseline During follow-up

Censoring at First outcome (typically) Death Migration Upper age limit, if age restriction Other criteria, e.g. exposure shift

CohortCohort

”Any designated group of individuals who are followed or traced over a period of time” Kenneth Rothman, Modern Epidemiology, 1998

Can be divided into closed and open populations

Closed and Open PopulationsClosed and Open Populations

Closed population

A population that adds no new members over time

Open/dynamic population

A population that may gain members over time or lose members who are still alive e.g. drug users within a specific observation period

Closed populationClosed population

limitationslimitations

Loss to follow-up (censoring)

Decreasing cohort size

Aging of cohort members

Depletion of susceptibles

Selection of the exposed populationSelection of the exposed population

General population Diet, Cancer & Health cohort, Danish Cancer Society

Individuals aged 50 to 64 years, follow-up from 1994 (n 57,000)

Occupational exposure groups Nurses Health Study, USA

Nurses aged 30 to 55 years, follow-up from 1976 (n 120,000)

Exposure ”Special exposure groups”

Ex.: Workers at the Thule base, Epileptics at Dianalund, individuals exposed to thorotrast

Drug users

Registers General Practice Research Database, UK Danish health and administrative registers

Selection of the comparison groupSelection of the comparison group

Ideally identical to the exposed group with respect to all other factors that may be related to the disease except the outcome(s) under study

”Internal” comparison general population/large occupational cohort frequent exposure

”External” comparison

General population (rates) Standardised incidence rate ratio (SIR) Standardised mortality rate ratio (SMR)

Data sourcesData sources

Exposure Existing data

registers medical records bio-banks

Questionnaires interview self-administered

Ad hoc measurements clinical parametes

biological samples

Outcome Registers

Clinical examination

Information from study subjects interview

questionnaire

Information from next-of-kin

Mortality data


Can examine multiple effects of a

single exposure rare exposures

Exposures with certainty precede outcomes (if prospective)

Allows direct measurement of incidence (IR, IP) of outcomes

Can elucidate temporal relationship between exposure and outcome

Allow study subjects to contribute person-time to multiple exposure categories

Biological material can be collected prior to outcome

If prospective, minimizes bias in the ascertainment of exposure

AdvantagesAdvantages


Is inefficient for the evaluation of rare diseases

If prospective, can be very expensive and time consuming

If retrospective, requires the availability of adequate records for both exposure and outcome

If prospective, cannot provide quick answers

If retrospective, precise classification of exposure and outcome may be difficult

Validity of the results can be seriously affected by losses to follow-up

DisadvantagesDisadvantages

Historical cohort studies

Comparison with general population (rates)

Nested case-control studies

Register studies


Methods for reduction of costs and time Methods for reduction of costs and time

Register studies in DKRegister studies in DK


Frank L. Science 2000;287: 2398-9Frank L. Science 2000;287: 2398-9


CPR Register

National Death Files

National Hospital Register

Birth Register

Prescription Databases

IDA Register(socioeconomic

variables)

Cancer Registry

Register studiesRegister studies

Registers are highly valuable data sources, BUT

Difficulties in interpretation due to incomplete data on competing risk factors

Life-style factors, socioeconomic factors, comorbidity, medical treatment

Other potential biases

Misclassification, non-compliance, etc.

Measures of disease frequencyMeasures of disease frequency

Definitions

What is the case? What is the study period?What is the population at risk?

Measures of disease frequency, summaryMeasures of disease frequency, summary

Incidence proportion (IP) Proportion of population that develops the outcome of interest during a

specified time Can be measured only in closed populations ”Average risk” for a population

Incidence rate (IR) Number of new cases of the outcome of interest divided by the amount of

person-time in the base population Can be measured in both open and closed populations Most often restricted to include a maximum of one event per person

Prevalence proportion (PP) Proportion of population that has the outcome of interest at given instant

IP+ = a/a+b IP- = c/c+d

RR = IP+/IP-

Attributable risk (AR) = IP+ - IP-

Attributable proportion (AP) = AR/IP+ = (RR-1)/RR

Effect measures in cohort studiesEffect measures in cohort studies

Incidence proportion Incidence proportion

ConditionsConditions

All persons should be followed-up from start of study (t0) until end of study with respect to the outcome(s) of interest

Problems: Open/dynamic population (t0?)

Competing risks of death Censoring

Is usually not directly observable, solution:

Computation of incidence rates

Relation between rate (IR) og risk (IP)

IP = 1 - exp(-IR x t) (IR constant)

IP = 1 - exp(- IRí x tí ) (IR variable)

IR small and/or short t:

IP IR x t

Person-time in study

cases

cases

Non-exposed

Exposed

Problem: Exposure status changes over time (episodical, sporadical)

Solution: Allow persons to contribute person-time to multiple exposure categories

Time dimensionTime dimension

Calendar time

1970

19901975 1980 1985 1995

Age

30

35

40

45

50

55

Y

X

Non-X

Contribution f rom the two study subjects

Exp. to drug X Non-exp. to drug X

Age PY Disease Y PY Disease Y

30-34 y 0 0 5 0

35-39 y 5 0 5 0

40-44 y 10 0 0 0

45-49 y 8 1 0 0

50-54 y 0 0 5 0

”Crude” 23 1 15 0

30-year-old man is enrolled in a cohort study of drug X in relation to disease Y in 1970 and followed free of Y through 1995

35-year-old man is enrolled in 1970 and followed until occurrence of Y in 1983

Person-time in study

cases

cases

Non-exposed

Exposed

Incidence rate = cases / person-time

Incidens Rate Ratio (IRR) = IR+ / IR-

A

PYC

PY

Cases Person-time

Exp

osu

re

Yes

No

A = Exposed cases

C = Non-exposed cases


IR+ = a/PY+

IR- = c/PY-

Incidence rate ratio (IRR) = IR+/IR-

Incidence rate difference = IRD (≈AR) = IR+ - IR-

AP = IRD/IR+ = (IR+-IR-)/IR+ = (IRR-1)/IRR


”Relative risk” vs. incidence rate ratio

Given IP IR x t (IR small)

”Relative risk” is equivalent with the ratio of two incidence

rates when the disease is rare

2

1

2

1

2

1

IRIR

tIRtIR

IPIP

Indirect Standardisation

Do more outcomes occur in the studied population than would be expected if the risk prevailing was the same as in the general population?

Estimation of expected number of outcomes Number of person-years at risk x incidence rate

PYage,period,sex x incidenceage,period,sex

Observed number/expected number ≈ RR

Standardised incidence ratio (SIR)


Calendar time

SIR = Observed number of outcomes/ expected number of outcomes

= Obs/IRpop x PYexp

= (Obs/PYexp) / IRpop

= IRexp / IRpop

≈ IRexp / IR0

= IRR (RR)

Risk windowRisk window

Exposure

Often unknown

Relevant exposure?Relevant exposure?

Ex Ex Ex

ExExEx

Ex Ex Ex

1-3 days?

10-15 days?

100-150 days?

Ex Ex Ex years?

Hazard functionHazard function

Outcome

Exposure

Theoretical association

Hazard functionsHazard functions

Outcome

Exposure

NSAID cohort studyNSAID cohort study

Population: Saskatchewan – province in Canada with appr. 1.1 mill. inhabitants

A study of the association between use of NSAIDs and risk of gastrointestinal (GI) bleeding included all 228,392 individuals who had redeemed one og more prescriptions for NSAIDs. The study subjects were followed during the period 1982-1986 for hospitalization due to upper GI bleeding

From the paper: .. Entered our cohort upon the first receipt of a prescription for diclofenac, indomethacin, naproxen, piroxicam or sulindac. Person-time contributed by this person continued until the earliest of: 1) hospitalization due to UGB, 2) death, 3) departure from Saskatchewan or 4) end of study

Note!: No control group of ’non-exposed’

Garcia Rodriguez et al. NSAIDs and GI-hospitalizations in Saskatchewan: A cohort study. Epidemiology 1992;3:337-42

Current user Recent past user Old past user Non-user

Day 0 30 60 150

Day 0 30 30 30 30 60

The person time of the study subjects was categorized according to time since last prescription

Current user Current user Current user Recent past userCurrent user

# 1

# 2

1.Rx 4.Rx3.Rx2.Rx

1. Rx

Current user Recent past user Old past user Nonuser

Person 1 30 30 90 >90 Person 2 120 30 - -

Modified from Garcia Rodriguez et al. NSAIDS and GI-hospitalizations in Saskatchewan: A cohort study. Epidemiology 1992;3:337-42

I ncidence rate ratios of GI -hospitalisations of NSAI D users

Current users Recent past users Old past users (0-30 days) (30-60 days) (60-150 days)

Diclofenac 3.9 2.2 1.3

I ndomethacin 4.0 1.7 1.4

Naproxen 3.8 2.3 1.4

Nonusers

1.0

Absolute vs. relative disease measuresAbsolute vs. relative disease measures

Avoid confusing measures of frequency with

measures of association (effect measures)

Ex:

A RR=10 is described as a high risk, or a population for whom RR=10 is said to be at higher risk than a population in which RR=5

A RR=10 may be described as a high relative risk

Risk of deep vein thrombosis (DVT)Risk of deep vein thrombosis (DVT)Third vs. second generation oral contraceptives Third vs. second generation oral contraceptives

RR 1.7 (1.4-1.7)

AR 1.5 per 10 000 person-years

Mortality of DVT 3%

Kemmeren et al. BMJ 2001; 323: 131-4

Vioxx (rofecoxib) and cardiovascular diseaseVioxx (rofecoxib) and cardiovascular disease

APPROVe trialAPPROVe trial

2,586 patients randomised to rofecoxib (Vioxx) (25 mg daily; n=1287) or placebo (n=1299) during a 3-year study period

1.50 CVE per 100 py (46 events; 3,059 py) vs. 0.78 CVE per 100 py (26 events; 3,327 py)

RR = 1.92 (1.19-3.11)

AR 72 pr. 10 000 py

Bresalier et al. N Engl J Med 2005; 352: 1092-1102

Attributable proportionsAttributable proportions

What proportion of the disease among the exposed is attributable to the exposure (APexp)?

APexp = IR+-IR0 / IR+ = AR / IR+ = (RR-1)/RR

What proportion of the disease in the total study population of exposed and non-exposed individuals is attributable to the exposure (APpop)?

APpop = IRpop-IR0 / IRpop

= AR x pe / IRpop (pe = exp. prevalence in population)

= APexp x pc (pc = exp. prevalence among cases)

= [(RR-1) x pe] / [(RR-1) x pe - 1]

Attributable proportionAttributable proportionIncidence rates of head and neck cancer per 100,000 py Incidence rates of head and neck cancer per 100,000 py

Among drinking smokers, what proportion of head and neck cancer is caused by smoking?

Among drinking smokers, what proportion of head and neck cancer is caused by drinking?

”Non-smoker” ”Smoker”

”Non-drinker” 1 4

”Drinker” 3 12


Among drinking smokers, what proportion of HNC is caused by smoking?

AP = IRD/IR+S+A = (IR+S+A-IR-S+A)/IR+S+A = (12-3)/12 = 75%



”Drinker” 3 12


Among drinking smokers, what proportion of HNC is caused by drinking?

AP = IRD/IR+S+A = (IR+S+A-IR+S-A)/IR+S+A = (12-4)/12 ≈ 67%



”Drinker” 3 12

Study population N GI bleeding

NSAID users 20,000 100

Non-users of NSAID 100,000 100

In total 120,000 200

A hypothetical population consists of 20.000 users of non-steroid A hypothetical population consists of 20.000 users of non-steroid anti-inflammatory drugs (NSAIDs) og 100.000 non-users of NSAID. anti-inflammatory drugs (NSAIDs) og 100.000 non-users of NSAID. The study subjects are followed for one year for the occurrence of The study subjects are followed for one year for the occurrence of upper gastrointestinal (GI) bleedingupper gastrointestinal (GI) bleeding

Please calculate the following measures of frequency and risk:

1. Incidence rate (IR) for GI bleeding in each exposure group

2. Incidence rate ratio (IRR) for the association between NSAID and upper GI bleeding

3. Incidence rate difference (IRD≈AR) between NSAID users and non-users

4. Attributable proportion (APexp) among users of NSAIDs

5. Attributable proportion (APpop) in the total population

(Censoring in the risk population should be ignored)

Study population N GI bleeding

NSAID users 20,000 100

Non-users of NSAID 100,000 100

In total 120,000 200

IRNSAID = 100/20000 = 0.005 = 5 per 1000 person-years

IRo = 100/100000 = 0.001 = 1 per 1000 person-years

AR = IRD = IRNSAID–IRo = 5-1 = 4 per 1000 person-years

IRR = IRNSAID/IRo = 5/1 = 5

IRpop = 200/120000 = 0.00167 = 1.67 per 1000 person-years

APexp = AR/IRNSAID = 4 per 1000/5 per 1000 = 0.80 or 80%

ARpop = IRpop–IRo = 1.67 – 1 = 0.67 per 1000 person-years

APpop = ARpop /IRpop = 0.67/1.67 0.40 or 40%

Epidemiology – Cohort studies I March 2010 Jan Wohlfahrt Afdeling for Epidemiologisk Forskning Statens Serum Institut.

Documents

cancer health cohort

exposed group

experience of exposed

modern epidemiology

uk danish health

new members

group of persons

period of time kenneth