EPID 620/PUBH 801: Epidemiologic Methods I Take Home Midterm Exam Due Wednesday, October 22, 2019 at 11:59pm Wingila Mpamila Total Points: 20
EPID 620/PUBH 801: Epidemiologic Methods I Take Home Midterm Exam Due Wednesday, October 22, 2019 at 11:59pm Wingila Mpamila Total Points: 20
Please make sure to read each question carefully and be clear and comprehensive in your
answer.
Please use complete sentences when providing explanations and interpretations.
1
EPID 620/PUBH 801: Epidemiologic Methods I
1) You are interested in whether eating apple pie causes insomnia. You find 50 people suffering
from insomnia at a sleep disorders clinic, and you find 150 people who do not have insomnia in a
nearby dermatology clinic. You interview each participant to determine their history of apple pie
consumption.
a) What kind of study is this (cross-sectional, case-control, prospective cohort,
retrospective cohort, randomized controlled trial)? [1 point]
The study is Case control
b) What is an appropriate measure of association for the apple pie – insomnia relationship
using data from this study? [1 point]
The appropriate measure of association is Odds ratio
2) You are conducting a study to identify whether BMI (treated as a categorical variable: <20,
20-25, >25) is associated with experiencing a myocardial infarction (treated as yes or no) after
age 50. You have the exact contribution of person-time at risk for each participant (including the
timing of each MI).
a) What determines the appropriate kind of regression model for an analysis, the exposure or
the outcome? [1 point]
The Variables
b) What type of regression analysis (e.g., linear regression, logistic regression, Cox
proportional hazards regression) would you use, and why? [1 point]
I would use Logistic regression because I have binary/two dependent variables and an
independent measurable variable (BMI)
II. LONGER (BUT STILL SHORT) ANSWERS (one sentence to one short paragraph
answers; 10 points total)use :
2
EPID 620/PUBH 801: Epidemiologic Methods I
CVD within the follow-up period. Based on these data, calculate the appropriate measure of
association for CVD and its 95% confidence interval. Interpret your results. [3 points]
CVD+ CVD- Total
E+ 389 8291 8,680
E- 266 8515 8,781
Total 655 16,806 17,461
Z=1.96
At least 95% confident interval of CVD among alcohol consumption compared to CVD among
non alcohol users is between 6.758 and 3.095 if there’s no bias. Since 95% CI includes null
value of 1 this is statistically insignificant.
[Show your work for full/partial credit. You may either do the calculations by hand, or you may
use SAS. If you choose to do them by hand, refer to slides for lecture 5 for the formula for the
standard errors. If you choose to use SAS, please provide syntax and output. You may adapt the
past syntax to enter the data.]
4) The randomized controlled trial (RCT) is often considered the “gold standard” study design in
epidemiology [3 points total].
a) What is the primary purpose of randomization in RCTs? Why does this often confer a
significant advantage over observational studies? [1 point]
The major purpose of this study is for human studies as it removes selection bias
And because they are randomized the selection bias is removed since the researcher is
not aware of the statues of the exposure groups. Thus important over the
3
EPID 620/PUBH 801: Epidemiologic Methods I
observational studies which are more prone to selection bias since the sample is
usually not randomized like the RCTs
b) Name and explain one threat to validity in an RCT other than non-adherence to
treatment. [1 point]
Drop out of study is another threat to validity because when more samples/people
over the years wither die or loose contact then there will be an effect in the study.
c) Describe why an analyst might want to use intent-to-treat analysis in an RCT. [1 point]
Intent to treat analysis is mainly used to maintain random selection and rid of bias because
when it is used even after dropping out, the sample are assigned the same treatment that
were assigned before dropping out.
5) You are interested in the effect of “Exposure” on “Disease.” Below is a DAG that describes
“true” causal mechanisms related to your question of interest, including measured variables V,
W, X, Y, and Z.
V
X Exposure Disease
a) Please list all pathways from Exposure to Disease before conditioning (i.e., before you
control for any variables), and indicate if they are closed (blocked) or open (unblocked).
For example, in a different DAG, you might say “OPEN: Exposure –> Disease” and
“CLOSED: Exposure –> A <– B –> C –> Disease.” [1 pt]
Exposure –>Y Z –> Disease
Exposure W –> Disease
Exposure X
X–>Exposure–>Disease
X–>Exposure–>YZ–>Disease
V–>Disease
X–>ExposureW–>Disease
4
EPID 620/PUBH 801: Epidemiologic Methods I
b) If you estimate the effect of Exposure on Disease without conditioning on any variables,
will your estimate be biased or unbiased? Why? [1 pt]
The estimate will be biased because the results will be non-random, however the
conditioning only leads to a non-random result if the conditioning was non-random at
the first place.
d) What is the minimum set of variable(s) that you can condition on to get an unbiased
estimate of the effect of Exposure on Disease? Hint: you may need to use the dagitty.net
site to answer this question [1 pt]
ANS- 3
6) Answer the following questions related to Curhan et al. 2012, which is posted on Blackboard
along with the midterm. Please make your answers fairly brief (1-4 sentences).
a) What are the exposures, and what is the outcome? What are the advantages and
disadvantages of the way these were ascertained? [1 point]
b) Note: In this paper, “RR” is not the same as the “risk ratio” that we have learned: the
authors use the term “relative risk,” abbreviated “RR,” to broadly encompass any
relative measure of disease frequency (e.g., odds ratio, risk ratio, incidence rate
ratio, hazard ratio). Summarize the main findings, based on Table 4, for any one of
the analgesics. The summary should represent your assessment of the results, and
not necessarily the authors’ interpretation. [1 point]
5
EPID 620/PUBH 801: Epidemiologic Methods I
NB. The reason behind this could be because old adults have passed the menopause and also do
not read/work or concentrate as much in comparison to the young adults which could be the
reason for the use of the analgesics (headache, back pain and menstrual pain)
Additionally, the reason why aspirin was reported to not be positively associated with the
hearing loss could be because only a very small amount of sample were reported to be using
aspirin compared to the other two.
In the Discussion, the authors discuss their findings and provide interpretation and explanation.
Please give your assessment: could any of this paper’s findings be affected by the following
threats to validity? Why or why not? [3 points]
i. Confounding, it could not be affected by confounding because the research
was adjusted both for age and for potential confounders. On the other side,
since the study only used one demographic group (non-Hispanic white
women), and hence did not adjust for race, we are sure whether race could
also be one of the confounders.
ii. Selection bias- thou it was adjusted, selection bias in this study can be
slightly observed because the research only reported those who “reported
loss of hearing”
iii. Measurement error- these are de-facto errors that tend to occur randomly
in any study so it possibly might have happened either randomly or
otherwise.
In this study, exchangeability is more common because there is only one sex which are the
women and also there is only one demographic group which are White non-Hispanic group and
hence the exchangeability is quite possible.