Is the association causal, or are there alternative explanations?

Is the association causal, or are there alternative explanations?

Epidemiology matters: a new introduction to methodological foundations

Chapter 8

2Epidemiology Matters – Chapter 1

Seven steps

1. Define the population of interest2. Conceptualize and create measures of exposures and health

indicators3. Take a sample of the population4. Estimate measures of association between exposures and health

indicators of interest

5. Rigorously evaluate whether the association observed suggests a causal association

6. Assess the evidence for causes working together7. Assess the extent to which the result matters, is externally valid, to

other populations


Inferential thinking, chapter 7

In Chapter 7 we asked a conceptual (counterfactual) question:

Would the disease have occurred when and how it did without the exposure, or without the amount of exposure that occurred, the timing of exposure, or within the context of multiple exposures?


Inferential thinking, chapter 8

In Chapter 8 we ask a pragmatic question:

Does the association that we measure in our data reflect the amount of excess disease that occurred due to the effects of the exposure, or could there be alternative explanations for the study findings other than a causal explanation?


1. Exposure causes disease

2. Complicating causes

3. Causal thinking in populations

4. Epidemiologic studies and assessing causes

5. Summary






5. Summary


When does exposure cause disease?

A counterfactual test to see if an exposure is a cause would require us to:1. Take the same person observed over the same time period,

once with the exposure and once without the exposure2. Hold all other characteristics of the person, place and time

constant3. Change only the exposure and observe then if the health

indicator changesThis is, of course, impossible


Non-diseased Diseased

Non-exposed Exposed


Observing individualsunder simultaneous conditions

Epidemiology Matters – Chapter 8 10

Observing individualsunder simultaneous conditions

Person 1: exposure causal

Person 2: exposure not causal






5. Summary


Why would an exposure be causal for Person 1 but not

causal for Person 2?


Complicating causes

Many sufficient cause sets can produce particular health indicators

The exposure of interest may be part of only one particular sufficient cause set; there are other sufficient causes that also produce the health indicator of interest


Complicating causes, an example

Disease X has two sufficient causes

1. A, B, and C

2. E, F, and G

Individual exposed to A, B, C, F, and G

Will get the disease

Completes sufficient cause 1 (A, B, and C)

Now exposed to E

Completes sufficient cause 2 (E, F, and G)

Exposure to E is not causal for this individual because she would have gotten the disease regardless given

exposure to A, B, and C

Therefore if E is exposure of interest we need to consider A, B, and C as other causes of disease

How can we visualize individuals with component causes not included in sufficient causal structure of E?


Previous exampleExposure of interest E

Component causes of sufficient cause A,B,C - without E


Previous exampleExposure of interest E

Component causes of sufficient cause A,B,C - without E

Person 2 gets disease regardless of exposure EThese additional causes complicate causal inference






5. Summary


Causal thinking in populations

Remember that epidemiological studies investigate groups of people

Therefore, our causal thinking applies to groups of individuals with multiple sufficient causes

We are interested in understanding the number of excess cases of disease that can be removed if we remove a particular cause


Group comparison, example





Excess cases of disease due to causal effect of the exposure on the outcome


Causal association?





5. Summary


Epidemiologic study design

It is impossible to observe the same people over the same period with and without

exposure

Instead we use group comparison of exposed and unexposed groups, often observed

in parallel over a similar time period

Ideally we want the unexposed group in an epidemiologic study to represent the

experience of exposed group had they not been exposed

However, what can complicate this approach is if there are imbalances in the

comparability of these groups allowing there to be different causes in each group

It is therefore essential to know how comparable these groups are to each other,

i.e., how close is the unexposed group to what we would expect the exposed group

to resemble if they were not exposed?


Distribution of additional causes

To assess comparability we need to know about the distribution

of other causes of disease between exposed and unexposed

groups


Comparing groupsEpidemiologic study #1


Comparing groupsEpidemiologic study #1 Epidemiologic study #2



Even distribution of dots across exposure conditions

Exposure conditions are comparable



Uneven distribution of dots across exposure conditions

These exposure conditions are not comparable Even distribution of dots across exposure conditions

Exposure conditions are comparable


Causal association?





5. Summary


Non-comparability

To replicate a counterfactual paradigm we want to observe the same

group at same time with the only variable changing being exposure

This is infeasible. Instead we compare groups of people and aim to keep

the distribution of all other variables equal between the groups

Failure to achieve this results in group ‘non-comparability’


Seven steps

1. Define the population of interest2. Conceptualize and create measures of exposures and health

indicators3. Take a sample of the population4. Estimate measures of association between exposures and health

indicators of interest

5. Rigorously evaluate whether the association observed suggests a causal association

6. Assess the evidence for causes working together7. Assess the extent to which the result matters, is externally valid, to

other populations


epidemiologymatters.org

Is the association causal, or are there alternative explanations?

Documents

exposure causal person

exposure cause disease

timing of exposure

result matters

association causal

causal associationassess

causal explanation

sufficient causesa