Description
Unformatted Attachment Preview
Use https://dagitty.net for the DAG
It is widely accepted that infants born to mothers who smoke have lower birth weights (LBW) and are at
higher risk of infant mortality compared to infants born to mothers who do not smoke. Assume that
smoking increases the risk of infant mortality through its effect on birth weight, and also through other
mechanisms not involving birth weight.
1.
Draw a DAG to illustrate the relationship between maternal smoking (MS), LBW, and infant
mortality (D) using only the knowledge described above.
2.
Now further assume that folate intake (F) is causally related to both LBW and infant mortality.
Add this information to your DAG from Q1.
3.
Now further assume that lifestyle choices (L) are associated with both maternal smoking and
folate intake. Add this information to your DAG from Q2.
4.
You wish to estimate the causal effect of maternal smoking on infant mortality. Use the DAG
from Q3 to:
a.
Identify any backdoor paths between maternal smoking and infant mortality and describe any
such paths as open or closed.
b.
Should you control for any variables in this analysis? If so, which ones (provide the minimal set
of variables)? If not, why not?
5.
Now further assume that there is an unmeasured common cause (U) of low birth weight (LBW)
and mortality. Add a variable U (and any necessary arrows) to the DAG from Q3. Will this impact your
causal effect measurement?
6.
You obtain information on date of death, cause of death, date of birth, birth weight, maternal
smoking, and other medical and sociodemographic factors for all live births in the US in 2005. You wish
to test the hypothesis that maternal smoking increases the rate of infant mortality. Compared with
nonsmokers, the unadjusted infant mortality rate ratio for smokers was 1.55. Upon adjustment for birth
weight, this rate ratio was 1.09.
a.
Assuming that the DAG you drew in Q2 is correct, which point estimate more closely represents
the overall causal effect of maternal smoking on infant mortality?
b.
Explain why the other point estimate is biased.
c.
What covariate(s) would you adjust for if you wanted to know the direct effect of maternal
smoking on infant mortality?
7.
A colleague suggests that your analysis is biased because you only included live births in your
study. That is, because you only had data on live births you implicitly excluded stillbirths and abortions
from your study. Assume that smoking increases the risk of stillbirths and that there are other,
unmeasured, common causes of stillbirths, LBW and infant mortality. Note that infant mortality is
measured on children who are born alive.
a.
Using the DAG you drew in Q1 as a starting point, draw a modified DAG that additionally
illustrates the role of restricting your study to live births.
b.
Is your study biased as suggested by your colleague? Why or why not?
c.
Assume that contrary to your expectations, maternal smoking is not a cause of stillbirths. Is your
answer to Q7b the same? Why or why not?
Purchase answer to see full
attachment
