Ecologic studies JF Boivin Version 8 October 2010
Jan 19, 2016
Ecologic studies
JF BoivinVersion 8 October 2010
2
Outline
1. Examples
2. Definition
3. Ecologic fallacy
• Definition
• Valid ecologic study
• Rate difference varies
• Reference rate varies
4. Contextual effects
5. Rationale for ecologic studies
6. Ecologic confounders
3
4
5
6
Figure 1. Statistical (empirical Bayes) estimates of county-specific adjusted annual suicide rates in the United States. A Bayes estimate of 1.0 indicates that the rate for the county was equal to the national rate of 12 per 100 000 population, a Bayes estimate of 2.0 represents a doubling of the national rate, and a Bayes estimate of 0.5 represents half the national rate. The estimates are based on all data from 1996 to 1998, adjusted for age, sex, and race.
Example: Time-trend studies
Time Trends in Autism and in MMR Immunization Coverage in California
7
Percentage of Children Receiving Measles-Mumps-Rubella (MMR) Immunization in Second Year of Life and Caseload of Children With Autism, by Year of Birth, California, 1980-1994
JAMA. 2001;285:1183-1185
Émile Durkheim• Émile Durkheim (1858 –1917) was a famous French
sociologist and pioneer in the development of modern sociology and anthropology.
• In a groundbreaking book published in 1897, entitled Le Suicide, Durkheim explored the differing suicide rates among Protestants and Catholics.
• In 19th century Europe, suicide rates were higher in countries that were more heavily Protestant. Durkheim found that suicide rates were highest in provinces that were heavily Protestant.
• He concluded that stronger social control among Catholics resulted in lower suicide rates.
• However, Durkheim's study of suicide was criticized as an example of the logical error termed the "ecological fallacy."
8
9
10
Using ordinary least-squares linear regression on Durkheim's data, Morgenstern (1995) found a strong positive correlation (Figure below) between proportion protestant and suicide rates. The estimated rate ratio, comparing Protestants with other religions, was 7.6 (i.e. suicide rates among protestants was about 8 fold higher than other religions).
11
Outline
1. Examples
2. Definition
3. Ecologic fallacy
• Definition
• Valid ecologic study
• Rate difference varies
• Reference rate varies
4. Contextual effects
5. Rationale for ecologic studies
6. Ecologic confounders
12
Ecologic study
A study in which the units of analysis are populations or groups of people, rather than individuals.
(Last. 2001)
13
Structure of an ecologic study: Counts
E+ E-
M1+
M1-
N1-N1+
D+
D-
?
?
?
?
Group 1
E+ E-
M2+
M2-
N2-N2+
D+
D-
?
?
?
?
Group 2
14
Person-years
E+ E-
M1+
PY1T
D+
PY
?
PY1+
?
PY1-
Group 1
E+ E-
M2+
PY2T
D+
PY
?
PY2+
?
PY2-
Group 2
15
Gibbons’ study
20
1,000,000
Suicide
PY
?
200,000
?
800,000
County i
Yes No
10
1,000,000
Suicide
PY
?
100,000
?
900,000
County 1 Antidepressant
16
17
Outline
1. Examples
2. Definition
3. Ecologic fallacy
• Definition
• Valid ecologic study
• Rate difference varies
• Reference rate varies
4. Contextual effects
5. Rationale for ecologic studies
6. Ecologic confounders
18
Ecologic fallacy
“…the mistaken assumption that a statistical association observed between two ecologic (group-level) variables… is equal to the association between the corresponding variables at the individual level…”
(Encyclopedia of epidemiologic methods. 2000)
19
Ecologic fallacy
“…arises when the disease rate in the unexposed (reference) population is correlated with exposure prevalence across groups or when the difference in rates between exposed and unexposed populations (biologic effect) varies across groups.”
(Encyclopedia of epidemiologic methods. 2000)
20
No ecologic bias
E+ E-
32
20,000
D+
PY
24
12,000
8
8,000
Group 2 (Ontario)
IE = 200/100,000
Io = 100/100,000RD = 100/100,000RR = 2
Group rate = 32/20,000 = 160/100,000% exposure = 12,000/20,000 = 60%
Adapted from Rothman-Greenland Table 23-2
E+ E-
28
20,000
D+
PY
16
8,000
12
12,000
Group 1 (Québec)
IE = 200/100,000
Io = 100/100,000
RD = 100/100,000RR = 2
Group rate = 28/20,000 = 140/100,000% exposure = 8,000/20,000 = 40%
21
No ecologic bias
110
120
140
130
150
160
170
180
190
200
100908070605040302010
RATE
(per
100
,000
)
% EXPOSURE
IRR = = 2=IE
Io 100/100,000
200/100,000
Québec
Ontario
22
Ecologic bias(rate difference varies across groups)
E+ E-
27
20,000
D+
PY
20
13,000
7
7,000
Group 2 (Ontario)
IE = 154/100,000
Io = 100/100,000
RD = 54/100,000RR = 1.54
Group rate = 27/20,000 = 135/100,000% exposure = 13,000/20,000 = 65%
E+ E-
33
20,000
D+
PY
20
7,000
13
13,000
Group 1 (Québec)
IE = 286/100,000
Io = 100/100,000
RD = 186/100,000RR = 2.86
Group rate = 33/20,000 = 165/100,000% exposure = 7,000/20,000 = 35%
23
Ecologic bias
110
120
140
130
150
160
170
180
190
200
100908070605040302010
RATE
(per
100
,000
)
% EXPOSURE
IRR = = 0.5=IE
Io 200/100,000
100/100,000
24
Ecologic bias(reference rate varies across groups)
E+ E-
46
20,000
D+
PY
40
16,000
6
4,000
Group 2 (Ontario)
IE = 250/100,000Io = 150/100,000
RD = 100/100,000RR = 1.67
Group rate = 46/20,000 = 230/100,000% exposure = 16,000/20,000 = 80%
E+ E-
28
20,000
D+
PY
16
8,000
12
12,000
Group 1 (Québec)
IE = 200/100,000
Io = 100/100,000
RD = 100/100,000RR = 2
Group rate = 28/20,000 = 140/100,000% exposure = 8,000/20,000 = 40%
25
Ecologic bias
0
100
150
200
250
100908070605040302010
RATE
(per
100
,000
)
% EXPOSURE
IRR = = 5.5=IE
Io
275/100,000
50/100,000
26(Koepsell & Weiss)
27
Outline
1. Examples
2. Definition
3. Ecologic fallacy
• Definition
• Valid ecologic study
• Rate difference varies
• Reference rate varies
4. Contextual effects
5. Rationale for ecologic studies
6. Ecologic confounders
28
29
• Neighborhood social class as aggregate of individual social classes
Can differ from study subjects’ social class
• Neighborhood social class as contextual variable
Same contextual variable for all subjects
Example: ecologic effect
30
31
Outline
1. Examples
2. Definition
3. Ecologic fallacy
• Definition
• Valid ecologic study
• Rate difference varies
• Reference rate varies
4. Contextual effects
5. Rationale for ecologic studies
6. Ecologic confounders
32
1. Low cost and convenience
2. Measurement limitation of individual-level studies
3. Design limitations of individual-level studies• Koepsell and Weiss, Figure 12.1
4. Simplicity of analysis and presentation
5. Instrumental variables
Rationale for ecologic studies
33(Koepsell & Weiss)
34
Outline
1. Examples
2. Definition
3. Ecologic fallacy
• Definition
• Valid ecologic study
• Rate difference varies
• Reference rate varies
4. Contextual effects
5. Rationale for ecologic studies
6. Ecologic confounders
35
No ecologic bias
Country 1 Country 2 Country 3
Covariate Exposure PY Rate PY Rate PY Rate
Yes Yes 3000 600 4000 600 4000 600
No 12000 500 8000 500 6000 500
RR 1.2 1.2 1.2
No Yes 2000 600 4000 600 6000 600
No 8000 500 8000 500 9000 500
RR 1.2 1.2 1.2
Sum Yes 5000 600 8000 600 10000 600
No 20000 500 16000 500 15000 500
RR 1.2 1.2 1.2
(rates per 100,000 person-year)(Note: no individual-level confounding)
Ecologic analysis
% exposed 20% 33% 40%
% covariate 60% 50% 40%
Overall disease rate 520 533 540
Crude RR = 1.2
Adjusted RR = 1.2
Crude is valid!
36
Country 1 Country 2 Country 3
Covariate Exposure PY Rate PY Rate PY Rate
Yes Yes 3000 600 4000 600 4000 600
No 12000 500 8000 500 6000 500
RR 1.2 1.2 1.2
No Yes 2000 200 4000 200 6000 200
No 8000 100 8000 100 9000 100
RR 2 2 2
Sum Yes 5000 440 8000 400 10000 360
No 20000 340 16000 300 15000 260
RR 1.3 1.3 1.4
(rates per 100,000 person-year)(Note: no individual-level confounding)
Ecologic analysis
% exposed 20% 33% 40%
% covariate 60% 50% 40%
Overall disease rate 360 333 300
Crude RR = 0.3
Adjusted RR = 1.3
Adjusted is valid!
Crude ecologic biasNo stratum-specific ecologic bias
37
Country 1 Country 2 Country 3
Covariate Exposure PY Rate PY Rate PY Rate
Yes Yes 8000 500 13000 1500 14000 1000
No 12000 500 12000 1500 6000 1000
RR 1 1 1
No Yes 2000 100 2000 300 6000 200
No 28000 100 23000 300 24000 200
RR 1 1 1
Sum Yes 10000 420 15000 1340 20000 760
No 40000 220 35000 711 30000 360
RR 1.9 1.9 2.1
(rates per 100,000 person-year)
Ecologic analysis
% exposed 20% 30% 40%
% covariate 40% 50% 40%
Overall disease rate 260 900 520
Crude RR = 8.6
Adjusted RR = 8.6
No valid estimate available!
Ecologic bias (crude and stratum-specific)
38
Another example
39
40