Transcript John Snow

Epidemiology Kept Simple
Sections 11.1–11.3
Ecological & Cross-Sectional Studies
Gerstman
Ecological and Cross-Sectional
1
Basic Design
• Ecological and cross-sectional studies involve no followup of individuals, so are often grouped together
• In addition, these studies depend on a full accounting or
random cross-section of the population
• This design is capable of measuring prevalences and
open population incidence rates:
Prevalence or rate,
group 1
Random sample of
population divided
into exposure groups
Prevalence or rate,
group 2
Compare
prevalence or rates
:
:
Prevalence or rate,
group k
Gerstman
Ecological and Cross-Sectional
2
Illustrative Example #1
Regional Cigarette Consumption & Lung Cancer
Each line of data represents a
geographic aggregate → this
is an ecological design
The variables name cig1930
refers to “cigarette
consumption per capital in
1930.” The variable mortalit
represents “lung cancer
mortality per 100,000 personyears in 1950”
Gerstman
Ecological and Cross-Sectional
3
Illustrative Example #1 (cont.)
Regional Cigarette Consumption & Lung Cancer
Per capita cigarette
consumption and lung
cancer mortality are highly
correlated, r = 0.74
Gerstman
Ecological and Cross-Sectional
4
Illustrative Example #2
% calories from fat & heart disease
Studies in the 1950s showed an
ecological correlation between high
fat diet and cardiovascular disease
mortality (see pp. 194–5)
Gerstman
Ecological and Cross-Sectional
5
Illustrative Example #3
Demonstration of Confounding
•
•
•
•
•
•
•
Gerstman
Confounding = bias due to an
extraneous variable
This historical study by Farr (1852)
reveals how ecological studies are
susceptible to confounding.
Explanatory variable = elevation
above sea level by neighborhood
Outcome variable = cholera mortality
This strong correlation was used to
support the erroneous miasma
theory (see Chapter 1!)
In fact, elevation plays no part in
cholera transmission
Confounding variable = proximity to
Thames River.
Ecological and Cross-Sectional
6
Illustrative Example #4
Psychosis, Neurosis, & Social Class
Here
are data from a 1964 field study of mental disorders
Note the negative correlation between high SES and psychosis
Note the positive correlation between high SES and neurosis
Can you predict biases in this study? (see next slide)
Prevalence of psychosis and neurosis by social
class, per 100,000 (Hollingshead & Redlich, 1964)
Social class
Psychosis
Neurosis
High
188
349
Moderate
291
250
Low
518
114
1505
97
Very low
Gerstman
Ecological and Cross-Sectional
7
Illustrative example #4 (cont.)
Psychosis, Neurosis, & Social Class
• Detection bias: Different diagnostic practices create
artificial differences in incidence or prevalence
– e.g., Poor people labeled psychotic; rich people labeled neurotic
• Reverse-causality bias: “Disease” causes the “exposure”
– e.g., Psychosis causes low SES
• Prevalence-incidence bias: Difference in prevalence but
not incidence
– wealthy people no more likely to be diagnosed with neurosis but
more persistent diagnoses (due to different type of health care)
• During later half of 20th century, epidemiologists became
increasingly aware of the limitations of cross-sectional
surveys, prompting development of cohort and casecontrol methods (see next set of slides…)
Gerstman
Ecological and Cross-Sectional
8
The remaining slides in this
presentation are optional
Gerstman
Ecological and Cross-Sectional
9
The Ecological Fallacy
(aggregation bias)
•
•
The ecological fallacy occurs when an
association seen in aggregate does not hold
for individuals
Illustrative example: There is a negative
ecological association between high foreign
birth and illiteracy rate (r = −0.62)
– When data are disaggregated, there is a positive
association high foreign birth and literacy (as one
would expect)
– Reason: high immigration states had better
public education
Gerstman
Ecological and Cross-Sectional
10
“Logic of the Ecological”
• Renewed interest in ecological measures
• Studies that mix aggregate observations and
individual-level observations are called multilevel designs
• Multi-level analysis useful in elucidating :
– causal webs
– interdependence between upstream factors and
downstream factors
Gerstman
Ecological and Cross-Sectional
11
Types of aggregate-level risk
factors (Susser, 1994)
• Integral variables – factors that effect all
community members (e.g., the local
economy)
• Contextual variables – summary of
individual attributes (e.g., % of calories from
fat)
• Contagion variables – a property that
involves a group outcome (e.g., prevalence of
HIV effects risk of exposure)
Gerstman
Ecological and Cross-Sectional
12
Illustrative Example
Goldberger on Pellagra
• Pellagra epidemics of
early 1900s initially
thought to be of infectious
origin
• Joseph Goldberger used
epidemiologic studies to
demonstrate nutritional
basis of pellagra (niacin
deficiency)
Gerstman
Ecological and Cross-Sectional
13
Goldberger’s (1918) Field Study of Food Intake (Average
Calories by Food Group) pp. 200 - 201
Nonpellagrous Households
Pellagrous Households
Groups of Foods
With Two or More
Cases (Mostly Low
Income
Households)
With Highest
Income
With Lowest
Income
Meats (exclusive of
salt pork), eggs,
milk, butter, cheese
762
639
338
270
Dried and canned
peas and beans
(exclusive of
canned string
beans)
126
113
115
123
Wheaten flour,
bread, cakes and
crackers, cornmeal,
grits, canned corn,
rice
2162
2082
1752
1840
Salt pork, lard and
lard substitutes
741
673
748
745
Green and canned
vegetables
(exclusive of corn),
green and canned
string beans, fruits
of all kinds
131
71
60
69
55
53
53
46
250
205
222
217
4267
3836
3288
3310
Irish and sweet
potatoes
Sugar, syrup, jellies
and jams
All foods . . . . .
Gerstman
With Lowest
Income and One or
More Cases
Ecological and Cross-Sectional
14