Title: Introduction: Causal theories and interrelationships between measures of disease occurrence
1Introduction Causal theories and
interrelationships between measures of disease
occurrence
- Lydia B. Zablotska, MD, PhD
- Associate Professor
- Department of Epidemiology and Biostatistics
2Learning Objectives
- Discuss how causal inference is central to the
role of epidemiology - Brief history of causal thinking through the
years - Theories of causal inference
- Causal models
- Sufficient-component cause model
- Describe (and critique) Rothmans causal
heuristic - Counterfactual model
- Counterfactual effect measures rate ratios, risk
ratios and odds ratios - Effect measures vs. measures of association
- Measures of attributable risk
- Causal diagrams (eg., directed acyclic graphs)
- Discuss how epidemiologic thinking leads to
causal inference - Discuss and critique Bradford Hills causal
criteria
3Practice of Epidemiology
- Example
- Study of the association between fiber intake and
risk of colorectal cancer - SEER 2008
- Incidence rates of colorectal cancer per year in
the U.S. - Males 60 per 100,000
- Females 43 per 100,000
40
50
60
7 Saga continues
0
- Cancer Causes Control. 2005 Apr16(3)225-33.
Dietary intakes of fruit, vegetables, and fiber,
and risk of colorectal cancer in a prospective
cohort of women (United States). Lin J et al. - CONCLUSIONS Our data offer little support for
associations between intakes of fruit,
vegetables, and fiber, and colorectal cancer
risk. However, our data suggest that legume fiber
and/or other related sources may reduce risk of
colorectal cancer. - Int J Cancer. 2006 Oct119(12)2938-2942 Dietary
intake of calcium, fiber and other micronutrients
in relation to colorectal cancer risk Results
from the Shanghai Women's Health Study. Shin A et
al. - CONCLUSIONS No apparent associations were found
for fiber, total vitamin A, carotene, vitamins
B1, B2, B3, C and E with colorectal cancer risk.
Our results suggest that calcium may be
protective against colorectal cancer development
8 and continues
- Am J Clin Nutr 2007851353 60.Dietary fiber and
whole-grain consumption in relation to colorectal
cancer in the NIH-AARP Diet and Health Study15.
A. Schatzkin et al. - CONCLUSIONS Total dietary fiber intake was not
associated with colorectal cancer. In analyses of
fiber from different food sources, only fiber
from grains was associated with a lower risk of
colorectal cancer... Whole-grain intake was
inversely associated with colorectal cancer
risk...
9 and then continues some more
Scand J Gastroenterol. 2010 Oct45(10)1223-31.
Dietary fiber, source foods and colorectal cancer
risk the Fukuoka Colorectal Cancer Study. K.
Uchuda et al. Results Total, soluble and
insoluble dietary fibers were not measurably
associated with overall risk or subsite-specific
risk of colorectal cancer. By contrast, rice
consumption was associated with a decreased risk
of colorectal cancer (trend p 0.03),
particularly of distal colon and rectal cancer
(trend p 0.02), and high intake of non-rice
cereals tended to be related to an increased risk
of colon cancer (trend p 0.07). There was no
association between vegetable consumption and
colorectal cancer, whereas individuals with the
lowest intake of fruits tended to have an
increased risk of colorectal cancer.
CONCLUSIONS The present study did not
corroborate a protective association between
dietary fiber and colorectal cancer, but
suggested a decreased risk of distal colorectal
cancer associated with rice consumption.
10 and just does not stop
BMJ. 2011 Nov 10343d6617. doi
10.1136/bmj.d6617. Dietary fibre, whole grains,
and risk of colorectal cancer systematic review
and dose-response meta-analysis of prospective
studies. D. Aune et al. Results 25 prospective
studies were included in the analysis. The
summary relative risk of developing colorectal
cancer for 10 g daily of total dietary fibre (16
studies) was 0.90 (95 confidence interval 0.86
to 0.94, I(2) 0), for fruit fibre (n 9) was
0.93 (0.82 to 1.05, I(2) 23), for vegetable
fibre (n 9) was 0.98 (0.91 to 1.06, I(2) 0),
for legume fibre (n 4) was 0.62 (0.27 to 1.42,
I(2) 58), and for cereal fibre (n 8) was
0.90 (0.83 to 0.97, I(2) 0). CONCLUSIONS A
high intake of dietary fibre, in particular
cereal fibre and whole grains, was associated
with a reduced risk of colorectal cancer. Further
studies should report more detailed results,
including those for subtypes of fibre and be
stratified by other risk factors to rule out
residual confounding. Further assessment of the
impact of measurement errors on the risk
estimates is also warranted.
11 genetic studies are on the horizon
12Epidemiology in the news
- Jennifer Kelsey on diet and nutrition articles in
The New York Times, week after week of cause
after cause.
13Why worry about causes?
- So that we can intervene
- So that we can reduce or prevent disease
14What is a cause?
- A cause is something that makes a difference.
Insofar as epidemiology is a science...that
aims to discover the cause of health states, the
search includes all determinants of health
outcomes. These may be both active agents... and
static conditions such as the attributes of
persons and places. - Mervyn Susser
15What is a cause?
- A cause is something that makes a difference.
Insofar as epidemiology is a science...that
aims to discover the cause of health states, the
search includes all determinants of health
outcomes. These may be both active agents... and
static conditions such as the attributes of
persons and places. - Mervyn Susser
160
- Back to basics
- Epidemiology is
- science that focuses on the occurrence of
disease rather than on the natural history or
some other aspect of the disease - K. Rothman
-
17 the study of the distribution and determinants
of disease frequency in human populations
0
MacMahon and Pugh (1970)
18 the study of the distribution and determinants
of disease frequency in human populations
0
MacMahon and Pugh (1970)
We also add
- AND the application of this study to
- control health problems
- improve public health
19Epidemiology defined
- Aims to find causes of diseases and to explain
varying patterns of disease occurrence across
populations and groups - The basic science or one of the pillars of public
health - Way of thinking and logically structuring
scientific inquiry in public health - Scientific discipline with roots in biology,
medicine, logic, and the philosophy of science
20Societal origins of epidemiology
0
- Epidemiology affects the daily lives of most
people - Comes from the Greek words epi and demos, meaning
the study of people - Originated in the Sanitary Era (XIX century) out
of necessity to improve the economic productivity
by decreasing squalor of the industrial slums - Epidemiology is the result of the evolution of
progressive thinking and our understanding of the
basic human rights
21And since it is the purpose of epidemiology to
- Identify factors that cause the distribution of
disease
22And since it is the purpose of epidemiology to
- Identify factors that cause the distribution of
disease - This must be the most important lecture of the
course
23Historical developments in the understanding of
causes of diseases
- 1. Sanitary era (paradigm miasma)
- Miasma theory of Sydenham
- foul emanations from soil, water and air cause
all diseases - poverty is at the core of all ills, it is a cause
rather then a consequence of disease - The Public Health Act of 1848
- Decaying organic matter ?insanitation ? foul
emanations ?diseases ?poverty ? high birth
rates among poor
Edwin Chadwick
24Historical developments in the understanding of
causes of diseases
- 2. Infectious disease era (paradigm germ theory)
- Discovery of causal agents of anthrax,
tuberculosis and cholera by R. Koch - Bacillus anthracis (1877)
- Mycobacterium tuberculosis (1882)
- Vibrio cholerae (1883)
Robert Koch
25Causal Inference Henle-Koch postulates for
causation
- The organism is always found with the disease
- The organism is not found with any other disease
- The organism, isolated from one who has the
disease, and cultured through several generates,
produces the disease (in experimental animals)
26Historical developments in the understanding of
causes of diseases
- 3. Risk factor epidemiology or chronic disease
era(paradigm black box) - Web of causation (MacMahon 1960)
- All factors are at the same level
- Diseases can be prevented by cutting a few
strands of the web - Does not elucidate societal forces or their
relation to health - too much statistics takes away all the
pleasure and the message of epidemiology.
Brian MacMahon
27Historical developments in the understanding of
causes of diseases
- 4. Ecoepidemiology (paradigm Chinese (nesting)
boxes) - Eras in Epidemiology The Evolution of Ideas
(Susser 2009) - Conceptual approach combining molecular,
societal, and population-based aspects to study a
health-related problem. - People are not only individuals but also members
of communities (social context) - Helps to recognize broad dynamic patterns and
disease in its social context - Places exposure, outcome and risk in societal
context.
Mervyn Susser
28Causal inference
- Goal of epidemiology
- learn causes of diseases and factors that could
prevent or delay disease development - Causal inference
- a process of determining causal and preventive
factors
29Theories of causal inference
- Deductive reasoning
- Inductivism
- Bayesianism
300
- George Simenons Inspector Maigret
- Arthur Conan Doyles Sherlock Holmes
- Agatha Christie's Hercule Poirot
Pull the clues together, arrive at
generalization, i.e. deduct the answer
310
In Epidemiology we use inductive
reasoning Francis Bacon (XVI century) suggested
the conditional inductive tree
formulate laws based on limited observations of
recurring phenomenal patterns
Specification of alternative hypotheses
Design of crucial experiments to test these
hypotheses
Exclusion of some alternatives
Adoption of what is left (for the time being)
32Deductive vs. inductive reasoning
- Deductive reasoning applies general principles to
reach specific conclusions - Wikipedia
33Deductive vs. inductive reasoning
- Deductive reasoning applies general principles to
reach specific conclusions, whereas inductive
reasoning examines specific information, perhaps
many pieces of specific information, to derive a
general principle. - Wikipedia
34Causal models
35What is a cause? (Rothman)
- A cause of a specific disease event is an
antecedent event, condition or characteristic
that was necessary for the disease at the moment
it occurred, given that other conditions are
fixed. - A cause of a disease is an event, condition, or
characteristic that preceded the disease event
and without which the disease event would not
have occurred at all or would not have occurred
until some later time.
36What is a cause? (Rothman)
- A cause of a specific disease event is an
antecedent event, condition or characteristic
that was necessary for the disease at the moment
it occurred, given that other conditions are
fixed. - A cause of a disease is an event, condition, or
characteristic that preceded the disease event
and without which the disease event would not
have occurred at all or would not have occurred
until some later time.
37What is a cause? (Rothmans sufficient-component
cause model)
- A cause of a specific disease event is an
antecedent event, condition or characteristic
that was necessary for the disease at the moment
it occurred, given that other conditions are
fixed. - A cause of a disease is an event, condition, or
characteristic that preceded the disease event
and without which the disease event would not
have occurred at all or would not have occurred
until some later time.
38Types of causal relationships(Rothmans
sufficient-component cause model)
- If a relationships is indeed causal, then
- Necessary and sufficient
- E.g., rabies, HIV exposure in AIDS
- Necessary but not sufficient
- Multiple factors acting in a specific temporal
sequence - E.g., multistage carcinogenesis
- Sufficient but not necessary
- E.g., both ionizing radiation and benzene
exposure cause leukemia independently - Neither sufficient nor necessary
- Many different pathways of getting the same
disease
39Sufficient and component causes
A sufficient cause is a set of minimal conditions
or events that inevitably produce disease
40Sufficient and component causes
Component causes
Sufficient Cause 1
Sufficient Cause 2
A sufficient cause is a set of minimal conditions
or events that inevitably produce disease
41Sufficient and component causes
A component cause is any one of a set of
conditions which are necessary for the completion
of a sufficient cause
Component causes
Sufficient Cause 1
Sufficient Cause 2
A sufficient cause is a set of minimal conditions
or events that inevitably produce disease
42Sufficient and component causes
A necessary component cause is a component cause
that is a member of every sufficient cause
Sufficient Cause 1
Sufficient Cause 2
43For example Tuberculosis
M. tuberculosis
M. tuberculosis
Immuno-suppression
Poornutrition
Sufficient Cause 1
Sufficient Cause 2
44Causing a myocardial infarction
Potato chips
Y
W
No exercise
45Causing a myocardial infarction
Potato chips
Y
W
Obesity
No exercise
A
46Causing a myocardial infarction
Potato chips
Y
W
Obesity
No exercise
A
NO EFFECT
47Causing a myocardial infarction
Potato chips
Y
W
Obesity
No exercise
A
C
Genes
48Causing a myocardial infarction
Potato chips
Y
W
Obesity
No exercise
High cholesterol
A
C
Genes
49Causing a myocardial infarction
Potato chips
Y
W
Obesity
No exercise
High cholesterol
A
C
Genes
NO EFFECT
50Causing a myocardial infarction
Potato chips
Y
W
Obesity
No exercise
High cholesterol
A
C
Genes
Smoking
Stress
51Causing a myocardial infarction
Potato chips
Y
W
Obesity
No exercise
High cholesterol
A
C
Genes
Smoking
Stress
52The trouble with Rothman
- Omits discussion of origins of causes, focuses on
proximal causes and ignores induction period - Specific components but not linkages among them
- Ignores indirect effects (effects of some
component causes mediated by other component
causes in the model) - Causes of disease in individuals but not in
populations - Does not consider factors that control
distribution of risk factors - Ignores dynamic non-linear relations
53Causing a myocardial infarction
Potato chips
Y
W
Obesity
No exercise
High cholesterol
A
C
Genes
Smoking
Stress
54Causing a myocardial infarction
Potato chips
Y
W
Obesity
No exercise
High cholesterol
A
C
Genes
Smoking
Stress
55Causing a myocardial infarction
Potato chips
Y
W
Obesity
No exercise
High cholesterol
A
C
Genes
Smoking
Stress
56Causing a myocardial infarction
Potato chips
Y
W
Obesity
No exercise
High cholesterol
A
C
Genes
Smoking
Stress
57Counterfactual model (potential-outcome)
- Ideal comparison to obtain a measure of effect
would be of study subjects with themselves in
both an exposed and an unexposed state - One of the two conditions in the definitions of
the effect measures must be contrary to fact
exposures or treatment vs. a reference condition
58Effect measures vs. measures of association
- Effect is
- The endpoint of the causal mechanism
- Change in a population characteristic that is
caused by the factor being at one level versus
another - Effect measures
- Can never achieve counterfactual ideal
- Logically impossible to observe the population
under both conditions - Measures of association
- Compares what happens in two distinct populations
- Constructed to equal the effect measure of
interest - Absolute differences in occurrence measures
(rate or risk difference) - Relative ratios of occurrence measures (rate or
risk ratio, relative risk, odds ratio)
59Comparison of absolute and relative effect
measures (Rothman 2002)
Measure Numerical Range Dimensionality
Risk difference -1, 1 None
Risk ratio 0, ? None
Incidence rate difference - ?, ? 1/Time
Incidence rate ratio 0, ? None
60Interrelationships between relative measures of
disease occurrence
- Rare disease assumption (Cornfield 1951)
- If disease is rare, the odds ratio approximates
the risk ratio - ORodds of disease among exposed/odds of disease
among unexposed - (A/B) / (C/D)
- RRrisk in exposed/ risk in unexposed
- (A/(AB)) / (C/(CD))
Disease Disease
-
Exposure A B
Exposure - C D
61Interrelationships between relative measures of
disease occurrence
- Exposure only negligibly affects the person-time
at risk (T1?T0) - IRRincidence rate among exposed/ incidence rate
among unexposed - (IR1xT1) / (IR0xT0)
- RRrisk in exposed/ risk in unexposed
- (A/(AB)) / (C/(CD))R1/R0
62Interrelationships between relative measures of
disease occurrence
- If R1gtR0, then A/Cgt1, i.e. OR overestimates
association and is larger than RR and further
away from the null - If R1gtR0, then T1ltT0 and T1/T0lt1, i.e.
IR1/IR0gtR1/R0, i.e. rate ratio fall between the
risk ratio and the odds ratio - 1 lt RR lt IRR lt OR
Disease Disease
-
Exposure A B
Exposure - C D
63Measures of attributable risk
- In Rothmans model, the fraction of disease
attributable to a single component cause cannot
exceed 100, but attributable fractions for
individuals could sum far more than 100 - Formula for attributable fraction
- For dichotomous exposure
- Risk difference/ Risk in exposed(RR-1)/RR
- For categorical (ngt2) exposure
- ? (AFi x Pi)
64Causal diagrams
- Provide a unified framework for evaluating design
and analysis strategies for any causal question
under any set of causal assumptions -
-
65ExampleComparison of mortality in Sweden and
Panama
- Our prediction
- Standard of living in Sweden is generally higher
than in Panama - Panama has more limited health care and higher
poverty rates compared to Sweden - Proportion of the population that dies each year
is higher in ?
66(No Transcript)
67ExampleComparison of mortality in Sweden and
Panama
- Our prediction
- Standard of living in Sweden is generally higher
than in Panama - Panama has more limited health care and higher
poverty rates compared to Sweden - Proportion of the population that dies each year
is higher in ? - Actual findings
- Death rates are lower for people of the same age
in Sweden - In both countries older people die at a greater
rate than younger people - Proportion of older people is greater in Sweden
68Causal diagrams
- Provide a unified framework for evaluating design
and analysis strategies for any causal question
under any set of causal assumptions -
- Older population
- SES Proportion of
dead each year
--
69Practice of causal inference
- Is the observed association valid? Is it true?
- Association appears causal but is due to
- Bias or systematic error (misclassification of E
or D) - Confounding (other variable causes the D and this
variable correlates with E) - Chance or random error (just this once)
- Did the exposure actually cause the disease?
- Use causal guidelines to decide if association is
truly causal - The most important is temporality
70Causal InferenceA. Bradford Hill Criteria for
Causal Inference (1965)
- Strength
- Consistency
- Specificity
- Temporality
- Biological gradient
- Plausibility
- Coherence
- Experiment
- Analogy
711. Strength of association
- Strong associations are less likely to be caused
by chance or bias - A strong association means a very high or very
low relative risk
CAVEAT Environmental associations with very low
relative risks
722. Consistency
- Replication of findings in different populations
under different circumstances, in different
times, with different study designs
- CAVEAT
- Lack of consistency does not rule out a causal
association, because some effects are produced by
their causes only under unusual circumstances - Publication bias
- Contradictory findings across different studies
are not unusual in studies of weak effects
733. Specificity of the association
- Specific exposure associated with only one
disease - Arises from old Henle-Koch postulates for
causation - Effect has one cause, not multiple causes
- CAVEATS
- Many exposures are linked to multiple diseases
- Many diseases have multiple causes
744. Temporality
- Exposure must precede disease (cause must precede
effect) - Levels of evidence
- Cross-sectional studies (exposure and disease
measured at the same time) - e.g., NHANES (National Health and Nutrition
Examination Survey) looking at the link between
obesity and coronary artery disease - Case-control studies (compare exposures and risk
factors among people with and without the
disease) - e.g., case-control study investigating the link
between radiation exposure among Chornobyl
clean-up workers and leukemia - Cohort studies (follow-up exposed and unexposed
to see who will develop the disease) - e.g., cohort study of children with thyroid
activity measurements taken within 6 weeks after
the Chornobyl accident and development of thyroid
cancer 15-22 years later) - In disease with long latency periods, exposures
must precede latency period - In chronic diseases, often need long-term
exposure for disease induction
755. Biologic gradient (dose-response relationship)
- Presence of a dose-response or exposure-response
curve with an expected shape - Changes in exposure are related to trend in risk
of disease - Strong evidence for causal relation suggesting
biologic relation
76Tronko et al. JNCI 2006
77Study of the survivors of atomic bombings in
Hiroshima (LSS)
Association between radiation dose received in
1945 and risk of developing cancer later in life.
Source E. Hall, Radiobiology for the
Radiologist, 2000.
78Threshold effect?
795. Biologic gradient (dose-response relationship)
- Presence of a dose-response or exposure-response
curve with an expected shape - Changes in exposure are related to trend in risk
of disease - Strong evidence for causal relation suggesting
biologic relation
- CAVEAT
- Thresholds, i.e., no disease past a certain level
of exposure
806. Plausibility
- The proposed mechanism should be biologically
(etiologically) plausible - Reference to a coherent body of knowledge
- CAVEAT
- New diseases and new causes
- Theoretical plausibility
81Radiation-associated risks of chronic lymphocytic
leukemia vs. other types of leukemia
Romanenko et al. Rad Res 2008
827. Coherence with established facts
- A cause-and-effect interpretation for an
association does not conflict with what is known
of the natural history and biology of disease - Implications
- If a relation is causal, would expect observed
findings to be consistent with other data - Hypothesized causal relations need to be
consistent with epidemiologic and biologic
knowledge
- CAVEATS
- Data may not be available yet to directly support
proposed mechanism - Science must be prepared to reinterpret existing
understanding of disease process in the face of
new evidence
838. Experiment
- In Hills article, refers to cessation of
exposure, i.e., elimination of putative harmful
exposure results in the decrease of the frequency
of disease
- CAVEATS
- If the pathogenic process has already started,
removal of cause does not reduce disease risk - Reduction in disease frequency might not be for
etiologic reason hypothesized
849. Analogy
Zablotska et al. Environ Health Perspect 2008
859. Analogy
- Similar exposures can cause similar effects,
e.g., medications and infectious agents may cause
other birth defects
- CAVEAT
- Limited by the current knowledge
86Overall caveats to criteria
- None of my ... criteria can bring undisputable
evidence for or against the cause-and-effect
hypothesis and none can be required as a sine qua
non. - Sir Austin Bradford Hill (1965)
- Temporality?
87Summary When is an association causal?
88SummaryWhen is an association causal?
Smoking is a carcinogen
89SummaryWhen is an association causal?
Smoking is a carcinogen
Smoking causes lung cancer
90SummaryWhen is an association causal?
Smoking is a carcinogen
Smoking causes lung cancer
Prospective cohort study
91SummaryWhen is an association causal?
Smoking is a carcinogen
Smoking causes lung cancer
Prospective cohort study
Recruit 10,000 doctors, follow for 10 years
92SummaryWhen is an association causal?
Smoking is a carcinogen
Smoking causes lung cancer
Prospective cohort study
Recruit 10,000 doctors, follow for 10 years
High RR of lung cancer in smokers
93Practice of Epidemiology
- Review of the previous example
- Study of the association between fiber intake and
risk of colorectal cancer
94Application of A. Bradford Hills Guidelines for
Causal Inference
- Strength - Yes
- Consistency - Questionable
- Specificity - No
- Temporality - Yes
- Biological gradient - Yes
- Plausibility - Yes
- Coherence - Possible
- Experiment - Yes
- Analogy - Yes
95Summary When is an association causal?
96ExampleA few well known causes of disease
- Smoking
- High cholesterol
- M. tuberculosis
- S. viridans
- Head injury
- ? Poverty
97ExampleA few well known causes of disease
- Smoking Lung Cancer
- High cholesterol Cardiovascular Disease
- M. tuberculosis Tuberculosis
- S. viridans Endocarditis
- Head injury Subarachnoid hemorrhage
- ? Poverty All-cause mortality
98Ethics and the public health balance
- When is there enough evidence to say something is
a cause? - When should we decide that something is a cause
and act on it? - Does first do no harm always apply at the
population level? - Are there different guidelines for solutions
where we have to DO something vs. solutions where
we try to remove something?
99Therefore, causal inference
- Causal inference is not a simple (or quick)
process - No single study is sufficient in establishing
causal inference - Requires critical judgment and interpretation
- Can one prove causal associations?
1000
November 26, 2001, The New Yorker.