Advanced Research Methodes Workgroups HCM

Advanced Research Methods
Quantitative methods
Workgroup 1: DAGs
Case 1: The costs of COPD treatment
1. What is the meaning of the following terms?
- Path: route between exposure and outcome
o 4 paths:
 ABC
 ADC
 AC
 AEC
- Backdoor path: does not follow the directions of the arrows
o ABC
o AEC
- Causal path: follows the direction of the arrows
o ADC
o AC
- Confounding: bias caused by common cause of exposure and outcome
o Grey hair  Age  Death
 Age (exposure) is both associated with grey hair and death, but there is
no causal path between grey hair and death
o Always adjust
- Collider: two arrows point to the same variable
o AEC
o Never adjust
- Intermediate variable: E  Z  O
o Adjusting depends on research question
- Blocking: removing a backdoor path from an association by adjusting for a variable on
that path (the variable should not be a collider)
o Adjust a variable so that the association changes
- Unblocking: opening a backdoor path by adjusting for a collider on that path
o Adjust a variable so that there is an association and possibly a causal effect

2. In what circumstances do you draw an arrow in a DAG? When should you decide not to draw
one?
You draw an arrow if there is a possible causal effect. When you assume there might be an
association (causal) between exposure variables and outcome variables.
You do not draw an arrow if there is certainly no possible causal effect. You are absolutely
certain that one variable does not affect another variable. It is stronger to not draw an arrow,
because this means that you are sure that there is no association.

3. Explain the difference between selection bias (collider bias) and confounding.
A collider is when both variables lead to the same outcome. When
you adjust these variables it leads to a selection bias, because a
collider should never be adjusted. Collider bias means that you’ve
opened a path that shouldn’t have been opened, because of the
fact that you’re dealing with a collider. In this case, X is the
collider. Prevent selection bias by not adjusting.

, A confounder is a variable that is associated with both the exposure as the outcome, but
there is no causal path between those. In this case, B is the confounder. There is a bias if you
don’t adjust for confounders. You need to adjust for this by blocking the variable.

The difference between the biases is that with selection bias you are opening a path and with
a confounder you are blocking a path.

Collider: common effect of exposure and outcome or a common effect of predecessors of
exposure and outcome

Confounding: exposure and outcome share a common cause, for which you have not
adjusted

4. For this case, formulate a research question that represents the goal of the researchers.
P: Patients with COPD stadium GOLD IV who are experiencing COPD exacerbation
I: Treatment at home
C: Treatment at hospital
O: Lower healthcare costs after 6 weeks start treatment

Does COPD exacerbation treatment at home to patients with COPD GOLD IV lower the
healthcare costs?

Don’t formulate a yes/no question, but be more specific. What is the difference between
costs etc.

5. As a preparation for your DAG, identify the exposure(s) and outcome of this case.
The exposure is something that you observe having an effect on the outcome.

The study is not about the effect of age on outcome. It is also not about the effect or severity
on outcome.
Exposure: treatment
Outcome: healthcare costs during 6 weeks after the start of treatment

6. Draw a DAG that represents the causal structure of this case.
Step 1: Write your exposure on the left and the outcome on the right side of a sheet of
paper.
Step 2: Draw an arrow from exposure to outcome.
Step 3: Add the other elements from the case (age, sex, severity of COPD at baseline).
Step 4: Draw all appropriate arrows between the items in your DAG.
Step 5: Consider whether you can make your DAG clearer by drawing a second version in
which some items are rearranged.

7. Identify the paths in the DAG. How many do you see?
There are 10 paths to be identified.
- TC
- TAC
- T  Sev  C
- T  Sex  C
- T  A  Sev  C
- T  Sev  A  C
- T  Sev  Sex  C
- T  Sex  Sev  C

, - T  A  Sev  Sex  C
- T  Sex  Sev  A  C

8. Which of the paths are open? Which are closed?
All paths are open unless arrows collide somewhere along the path. In this case, there are 8
open paths and 2 closed paths.
Closed path:
- T  A  Sev  Sex  C
- T  Sex  Sev  A  C

9. Which of the paths are causal paths? Which are backdoor paths?
There is 1 causal path and there are 9 backdoor paths.
Causal path:
- TC

10. Identify the confounders and colliders in your DAG.
Confounders: age, severity, sex
Colliders: severity

11. For which variables would you adjust in the statistical analysis? Would you expect this to lead
to an unbiased estimate?
Adjust for age and sex by blocking them (draw a block around it). They are the confounders
and if you block them, then there is no confounder bias. Adjusting for age and sex leads to
another open path without collider (T  Sev  C).

Age, Sex, Severity. This would lead to an unbiased estimate, since the combination of
adjustments would isolate the causal path from exposure to outcome.
It is true that Severity is a collider in some paths. These backdoor paths are opened by
adjusting for Severity. However, if we also adjust for confounders on those paths (Age, Sex),
we close them again.

Case 2: Fox News
1. The research question of this study is not stated explicitly in the report. How would you
formulate it, based on what you have read?
P: New Jersians
I: Different types of news sources
C: No news source
O: Knowledge of current events at home and abroad

What is the effect of different types of news sources on the knowledge of current events at
home and abroad of New Jersians?

2. How does the author of this study interpret the results?
People who watch FOX News have less knowledge on current events abroad compared to
people who do not watch the news at all. The Daily Show and the Sunday Show leads to
more knowledge on current events abroad. He implies its due to Fox News and not to the
people who watch Fox News. There is a causal effect between Fox News and less knowledge.
They think they have isolated the causal effect by adjusting for other factors.

3. Draw the DAG that describes the effect of media
consumption on knowledge of news and current
events. This DAG does not have to include or be

, limited to the variables that were collected in this particular survey. Start by writing down
what the exposure and the outcome are.




4. The author of this study did not provide a DAG. Draw the DAG that is consistent with the
design of this study. (Consider what variables were included and if they were used as
confounders, intermediate variables or colliders).




This DAG does not contain a factor such as ‘being interested’. However, it is essential for
explaining the implausible result. How could people know less by watching certain news
channels when there are no indications that these broadcasters lied to their viewers?

5. Compare the two DAGs. What do you think now about the conclusion of the study?
The conclusion of the study is not totally fair/correct, because they did not adjust for
something like whether a person is interested in foreign news. The researchers couldn’t
make those statements yet, because they didn’t adjust for everything nor did they do it
correctly. Because when you are adjusting for that variable, your interpretations are better.
There may be much unmeasured confounding. Furthermore, there no reason is given why
Fox viewers would be misinformed about these topics.

6. In chapter 7 of Naked Statistics, Wheelan discusses a number of types of bias. Which type do
you see in the news study?
Selection bias: If each member of the relevant population does not have an equal chance of
ending up in the same sample, we are going to have a problem with whatever results emerge
from the sample.

Publication bias: Positive findings are more likely to be published than negative findings,
which can skew the results we see.

Recall bias: Memory is not always a great source of good data. Our memories turn out to be
systematically fragile when we are trying to explain some particularly good or bad outcome
in the present.

Survivorship bias: Occurs when some or many of the observations are failing out of the
sample, chancing the composition of the observations that are left and therefore affecting
the results of any analysis.

Healthy user bias: People who faithfully engage in activities that are good for them (e.g.
prescribed drugs of having healthy diet), are fundamentally from those who don’t. This effect
can potentially confound any study trying to evaluate the real effect of activities perceived to
be healthful, such as exercising regularly or eating kale. This is also the type you see in the