Epid 600 Class 5 Cohort Studies

EPID 600; Class 5 Cohort studies University of Michigan School of Public Health

Drug Abuse: A workshop on behavioral and economic research October 18-20, 2004 1

Three key dimensions to epidemiologic studies Measures of association Relative measures (relative risks, rates, and odds) Absolute measures (risk and rate differences) Study design Observational Cohort Case-control Cross-sectional Experimental Randomized trial Field trials Group randomized trials Units of analysis Individual Group 2

Three key dimensions to epidemiologic studies Measures of association Relative measures (relative risks, rates, and odds) Absolute measures (risk and rate differences) Study design Observational Cohort Case-control Cross-sectional Experimental Randomized trial Field trials Group randomized trials Units of analysis Individual Group 3

The world

persons “exposed”

persons “unexposed”

4

The cohort study

persons “exposed”

persons “unexposed”

5

The cohort study

persons “exposed” with disease

persons “unexposed” with disease

6

What is a cohort? 1. 

A place to play tennis

2. 

The tenth part of a Roman legion

3. 

A population that is surveyed at a given moment in time

4. 

People born a hundred years apart

5. 

Equivalent to a trohoc

7

Why epidemiologic studies Why epidemiologic studies? To determine whether there is an association between “exposure” and “outcome” Example Does aspirin prevent myocardial infarctions? Is eating carrots associated with increased risk of skin cancer?

8

Types of cohort studies

time

i Concurrent Prospective

i Mixed

i Non-concurrent Retrospective

It is where the INVESTIGATOR sits that determines the type of cohort

9

Types of cohort studies

time

i

Concurrent Prospective

i

Mixed

i

Non-concurrent Retrospective

It is where the INVESTIGATOR sits that determines the type of cohort

10

Prospective vs. retrospective In prospective cohort studies, exposure and non-exposure ascertainment happens in present then study groups followed over time to measure disease In retrospective cohort studies, exposure and non-exposure are ascertained in the past In “mixed” cohort studies, we have a bit of each approach

11

Retrospective vs. prospective Advantages of retrospective studies Less expensive Less time consuming Efficient for study of diseases with long latency periods Disadvantages of retrospective studies Introduces possible error in the form of recall of past information, challenges in collecting data retrospectively, primarily information bias (to be covered in a future lecture)

12

Fixed vs. open cohorts Fixed cohort Has fixed membership Once group is defined and follow-up begins, no-one is added Open cohort Also called a “dynamic cohort”, can take on new members during study period

13

Fixed cohort Fixed cohort

14

Fixed cohort Fixed cohort

cohort is set; no new participants

there may be withdrawals from cohort

size of cohort gets smaller over time 15

Closed fixed cohort Fixed cohort

there are no withdrawals, all persons are followed until end of follow-up period of until they get disease

cohort is set; no new participants 16

Open cohort Open cohort

17

Open Open (dynamic) cohort

cohort may be replenished; size of cohort does not necessarily shrink

new cohort members 18

Rothman KJ. Epidemiology: An Introduction. Oxford, 2002.

19

Cohort studies

Exposed

D+ DD+

Unexposed

D-

20

Cohort studies

Exposed

D+ DD+

Unexposed

D-

Disease

No disease

Exposed

a

b

Not Exposed

c

d 21

Cohort studies

Exposed

D+ DD+

Unexposed

D-

Disease

No disease

Exposed

a

b

Not Exposed

c

d 22

Cohort studies

Exposed

D+ DD+

Unexposed

D-

Disease

No disease

Exposed

a

b

Not Exposed

c

d 23

Cohort studies

Exposed

D+ DD+

Unexposed

D-

Disease

No disease

Exposed

a

b

Not Exposed

c

d 24

Cohort studies

Exposed

D+ DD+

Unexposed

D-

Disease

No disease

Exposed

a

b

Not Exposed

c

d 25

Advantages of cohort studies Maintains temporal sequence, i.e., assesses exposure before outcome Good for assessing rare exposures and rapidly fatal diseases Can study multiple diseases/outcomes from a given exposure Can calculate incidence among exposed and unexposed Minimizes error in ascertainment of exposure (at least if prospective) Provides complete description of experience subsequent to exposure, including rate of progression and natural history of disease 26

Disadvantages of cohort studies Expensive Inefficient for rare diseases Potentially long duration for follow-up Secular trends in technology, behaviors, and changes that may influence behavior and study characteristics over time

27

An example Population: A cadmium factory in South Dakota Exposure: Exposure to cadmium production (which involves the gaseous decomposition of cadmium compounds); exposure assessed by information on jobs at high risk of exposure between 1950 and 1970 Outcome: respiratory cancers, mostly lung and nasal cancer

28

What kind of study is this?

Exposure interval, 1950-1970

End of follow-up, 2000

You (the investigator) are here

29

Therefore... This study aims to identify association between cadmium exposure and subsequent development of respiratory cancer Who is at risk of respiratory cancer? Persons with lungs and noses Persons who do not already have respiratory cancer at baseline

30

What did we find?

Exposed

D+ DD+

Unexposed

D-

31

What did we find?

Exposed

Unexposed

250

450

D+

100

D-

150

D+

90

D-

360

32

2x2 table

Cancer

No cancer

Total

Exposed

100

150

250

Not Exposed

90

360

450

Total

190

510

700

33

How many of you eat breakfast? A cross-sectional association between skipping breakfast and obesity has been shown in adults. Is it that people with low SES skip breakfast so tend to be obese, or is it that skipping breakfast in itself makes people obese? A research team in Britain decided to investigate the association between proportion of calories consumed at breakfast and weight gain.

Purslow et al. Energy intake at breakfast and weight change: prospective study of 6,764 middle-aged men and women. Am J Epid. 2007; 34 167(2):188-192.

Breakfast study: cohort mechanics STEP 1: Indentify a group of people of interest. Participants recruited for the European Prospective Investigation into Cancer and Nutrition (Norfolk cohort); first (baseline) measurement on nutrition and weight STEP 2: Follow them through time and monitor outcome of interest During 1998-2000, a second measurement on nutrition and weight; include only those who did not report stroke, cancer, or heart attack at baseline STEP 3: Classify participants according to outcome across exposure categories. Classify people as breakfast eaters (exposed cohort) or noneaters (unexposed cohort); assess weight change in both cohorts Purslow et al. Energy intake at breakfast and weight change: prospective study of 6,764 middle-aged men and women. Am J Epid. 2007; 35 167(2):188-192.

Breakfast study: Find people and follow them through time 3) 2007: Analysis; Assess weight change in both cohorts 2) 1998-2000: 2nd measurement on nutrition and weight; include only those who did not report stroke, cancer, or heart attack at baseline 1) 1993-1997: 1st (baseline) measurement on nutrition and weight Purslow et al. Energy intake at breakfast and weight change: prospective study of 6,764 middle-aged men and women. Am J Epid. 2007; 36 167(2):188-192.

Breakfast study: set up What is the “exposure”? > Breakfast consumption How do the investigators measure the exposure? > They determine the percentage of total energy intake consumed at breakfast What is the outcome of interest? > Weight gain How do the investigators measure the outcome? > Change in weight in kg from time 1 to time 2 Purslow et al. Energy intake at breakfast and weight change: prospective study of 6,764 middle-aged men and women. Am J Epid. 2007; 37 167(2):188-192.

Breakfast study: findings Everyone gained weight over time…. But larger breakfasts are associated with lesser weight gain among the middle aged participants of this study This association is independent of the quantity of calories consumed in the day, social class, physical activity level, fruit and vegetable intake, fat/carb/ protein intake, smoking, BMI

Calorie Distribution

Avg weight gain in the follow up

Big breakfast (22-50% of total intake)

1.23 kg

Small Breakfast (0-11% of total intake)

0.79 kg

10 yrs x 1 lb/year = 10 lbs of potentially avoidable weight gain!

Purslow et al. Energy intake at breakfast and weight change: prospective study of 6,764 middle-aged men and women. Am J Epid. 2007; 38 167(2):188-192.

So, what measures of disease occurrence and what measures of association can we calculate? Risk (incidence proportion) Odds Prevalence at any point during the cohort (but making some assumptions about duration of disease) Incidence rate? And... Risk ratio Risk difference Odds ratio (relative odds) Incidence rate difference? 39

Relative risk (risk ratio) The ratio of risks for two populations

RR =

Rexp osed Run exp osed

100 Rexp osed = 250 90 Run exp osed = 450 100 100* 450 250 RR = = = 2.0 90 250*90 450

So, the risk of developing respiratory cancer among South Dakota miners exposed to cadmium was 2.0 higher than among miners not exposed to cadmium in a cohort study with a follow-up period of 30-50 years

40

Risk difference The additional risk among those exposed when compared to those unexposed

RD = Rexp osed − Run exp osed

100 90 (100* 450) − (90* 250) RD = − = = 0.2 250 450 (250* 450)

So, the difference in the risk of developing respiratory cancer among South Dakota miners exposed to cadmium compared to miners not exposed to cadmium was 0.2 in a cohort study with a follow-up period of 30-50 years 41

Can we calculate incidence rate ratio from this information? We could, if we made some assumptions Length of follow-up same for everyone No competing risks No loss to follow-up Assume then that average follow-up was 40 years; i.e., we are assuming that everyone was followed for 40 years

PYOexp osed = 250* 40 = 10000 PYOun exp osed = 450* 40 = 18000 42

Relative rate (rate ratio) The ratio of rates for two populations

IRR =

IRexp osed IRun exp osed

100 10, 000 100*18, 000 IRR = = = 2.0 90 10, 000*90 18, 000 So, the rate of developing respiratory cancer among South Dakota miners exposed to cadmium was 2.0 higher than among miners not exposed to cadmium in a cohort study with a follow-up period of 30-50 years And, the relative rate and relative risk are the same assuming that time of follow-up is identical; of course this assumption is only valid if there is nothing else is competing as a 43 risk, that follow-up is complete, and that this is a closed cohort

Relative odds (odds ratio)

pexp Oddsexp osed Oddsunexposed

1- pexp = punexp 1 − pun exp

100 100 250 250 100 150 100 250 100 1* 250 = 250 = 250 150 = 150 = 100 * 360 = 2.67 = 90 90 90 450 90 150 90 * 450 450 450 360 360 90 360 1450 450 Cancer

No cancer

Total

Exposed

100

150

250

Not Exposed

90

360

450

Total

190

510

700

44

Relative odds (odds ratio)

pexp Oddsexp osed Oddsunexposed

1- pexp = punexp 1 − pun exp

100 100 250 250 100 150 100 250 100 1* 250 = 250 = 250 150 = 150 = 100 * 360 = 2.67 = 90 90 90 450 90 150 90 * 450 450 450 360 360 90 360 1450 450 Cancer

No cancer

Total

Exposed

100

150

250

Not Exposed

90

360

450

Total

190

510

700

45

Relative odds (odds ratio)

pexp Oddsexp osed Oddsunexposed

1- pexp = punexp 1 − pun exp

100 100 250 250 100 150 100 250 100 1* 250 = 250 = 250 150 = 150 = 100 * 360 = 2.67 = 90 90 90 450 90 150 90 * 450 450 450 360 360 90 360 1450 450 Cancer

No cancer

Total

Exposed

100

150

250

Not Exposed

90

360

450

Total

190

510

700

46

Relative odds (odds ratio)

pexp Oddsexp osed Oddsunexposed

1- pexp = punexp 1 − pun exp

100 100 250 250 100 150 100 250 100 1* 250 = 250 = 250 150 = 150 = 100 * 360 = 2.67 = 90 90 90 450 90 150 90 * 450 450 450 360 360 90 360 1450 450

a*d therefore Odds Ratio = b*c

Cancer

No cancer

Total

Exposed

100

150

250

Not Exposed

90

360

450

Total

190

510

700

47

Interpretation of odds ratio The odds of respiratory cancer is 2.67 times greater in those exposed to cadmium compared to unexposed Note that RR < OR...remember why?

48

Reality...

Exposure interval, 1950-1970

End of follow-up, 2000 p1 p2 p3 p4 etc

49

So we can take into account PYO That is, by taking into consideration, the actual time of follow-up Cancer

PYO

Exposed

100

20,000

Not Exposed

90

30,000

Total

190

50,000

100 20, 000 100*30, 000 Re lative rate = = = 1.67 90 90* 20, 000 30, 000 50

But of course we can do better That is, by taking into consideration, the actual time of follow-up Cancer

PYO

Exposed

100

20,000

Not Exposed

90

30,000

Total

190

50,000

100 20, 000 100*30, 000 Re lative rate = = = 1.67 90 90* 20, 000 30, 000

Without knowing actual PYO we cannot know how RR is estimating 51 IRR

But of course we can do better That is, by taking into consideration, the actual time of follow-up Cancer

PYO

Exposed

100

20,000

Not Exposed

90

30,000

Total

190

50,000

100 20, 000 100*30, 000 Re lative rate = = = 1.67 90 90* 20, 000 30, 000

If PYOs were different, would it be possible that cadmium is actually protective? 52

Rate difference The additional incidence rate comparing those exposed vs. those unexposed

IRD = IRexp osed − IRun exp osed 100 90 2 IRD = − = 0.002 or 20, 000 30, 000 1000 person years

53

Attributable fraction among exposed Proportion of the disease burden among exposed people that is due to the exposure AFexp osed =

Rexp osed − Run exp osed

AFexp osed

Rexp osed

100 90 − 0.2 250 450 = = = 0.5 100 0.4 250

So, we say, that 50% of disease among exposed is attributable to exposure; or, if we removed all the exposure, we might expect to reduce disease by 50% among exposed

54

Attributable fraction in population Proportion of the disease burden among the whole population that is due to the exposure

AFpopulation =

AFpopulation

Rpopulation − Run exp osed Rpopulation

190 90 − 0.072 700 450 = = = 0.26 190 0.271 700

55

Or... AFpopulation

p * (RR − 1) = p * (RR − 1) + 1

AFpopulation

prevalence of exposure

250 * (2.0 − 1) 0.36 700 ≅ = = 0.26 250 1.36 * (2.0 − 1) + 1 700

So, if exposure is removed, we would expect that disease would be reduced by ~26% in the whole population Attributable fraction in population is always lower than attributable fraction among exposed, as long as exposure is associated with more disease. Why?

56

Summary Cohort studies follow participants forward in time Cohort studies allow us to calculate all cardinal measures of association Key limitations to cohort studies are unsuitability for rare disease and the fact that they take a long time (and are expensive!) to implement

57

Aside...cross-sectional studies Prevalence studies “Snapshot” look at exposure and outcome at a point in time Only provide relative odds; do not allow us to calculate either risk or rates, so no relative risks or relative rates Potential over-representation of diseases with long duration (P=ID) 58