Reference no: EM133302096

Applied Epidemiological Data Analysis

This practice uses the CRC3 data that you are familiar with. Remember this is a cross-sectional study on cardio metabolic risk among Costa Rican adults. We are going to use abdominal obesity as our outcome and alpha-linolenic acid as our main exposure. The purpose of this evaluation is to assess the association between abdominal obesity and alpha-linolenic acid using the appropriate models.

Question 1: Create binary variables for your outcome, abdominal obesity, and your exposure, alpha-linolenic

Instructions:
a) Dichotomize the variable alpha-linolenic acid using the median
b) Create the abdominal obesity variable by using the metabolic syndrome cutoff (remember it is different by sex)
c) What is the prevalence of abdominal obesity in this population? And in women?

Question 2: Crude association between alpha-linolenic acid and abdominal obesity using a two by two table

Instructions:
a) Run a two by two table between alpha-linolenic acid (median variable) and abdominal obesity
b) Calculate an odds ratio
c) Calculate a prevalence ratio
d) Are they different? Why or why not?

Question 3: Run a logistic model to test the association between alpha-linolenic acid and abdominal obesity

Instructions:
a) Run a logistic regression model with PROC GENMOD with the median alpha-linolenic acid as the exposure and abdominal obesity as the outcome. Use low alpha-linolenic acid as the reference.
b) What is the OR and the 95% CI? Interpret this OR. Is it consistent with your biological hypothesis? Why or why not?

Question 4: Run a log-binomial model to test the association between alpha-linolenic acid and abdominal obesity

Instructions:
a) Run a log-binomial regression model with PROC GENMOD with the median alpha-linolenic acid as the exposure and abdominal obesity as the outcome. Use low alpha-linolenic acid as the reference.
b) What is the PR and the 95% CI? Interpret this PR.
c) What would be the PR and 95% CI of abdominal obesity for people below the median alpha-linolenic acid compared with people above the median alpha-linolenic acid?
d) Imagine your previous model did not converge. Run a robust Poisson model to estimate the prevalence ratio of abdominal obesity comparing people above the median alpha-linolenic acid with people below the median.
e) Interpret your measure of association. Do you get similar results? Why or why not?

Problem Set

Question 1:Create binary outcomes and report prevalence of the outcomes

Instructions:
a) Create binary variables for each continuous z-score (HAZ, WAZ, and WHZ) using the following cutoffs:

Indicator	Cutoff	Moderate Malnutrition
Height-for-age z-score (HAZ)	<-2	Stunting
Weight-for-age z-score (WAZ)	<-2	Underweight
Weight-for-height z-score (WHZ)	<-2	Wasting

b) Report the prevalence of stunting, underweight, and wasting with their corresponding 95% CI assuming this data comes from a complex sampling design. Please, round off to one or two decimal places at most.

c) Report the prevalence of stunting, underweight, and wasting with their corresponding 95% CI, that you would have obtained assuming that these data comes from a simple random sample where all children have the same probability of being selected. Please, round off to one or two decimal places at most.

d) Are there any differences in the prevalence estimates using the different assumptions in b and c? Independently of finding differences or not what would you expect and why?

e) Are there any differences in the 95% CI using the different assumptions in b and c? Please state why or why not differences may exist.

Question 2: Bivariate analysis

Instructions:

a) Create a descriptive Table 1 with the stunting variable as the outcome (stunted=yes vs. stunted=no) including the variables listed below. Fill in the descriptive Table 1 reporting means and standard errors for continuous variables for stunted and non-stunted groups. Fill in the descriptive Table 1 reporting percent for categorical variables for stunted and non-stunted groups. Unless otherwise specified binary variables are coded as 1=yes, 0=no. Ignore the fact that some variables have missing values and work with the data you have available.

Mother's age
Mother BMI
Mother had elementary school
Higher SES
No toilet in the house (0=there is toilet, 1=there is no toilet)
Child age in months
Sex of the child (0=girls, 1=boys)
Months breastfeeding
Child was bottle fed
Child received vitamin A supplements
Child had diarrhea last month
Child had fever last month
Child had malaria last month
Child had intestinal parasites last month

b) Add a column to your table 1 with p-values and a footnote explaining which statistical test you used for each variable. Assume normality for continuous variables (no need to check)

c)Based on your bivariate analysis (Table 1), is mother's education associated with stunting? Explain.

Question 3: Multivariate analysis

Instructions:
In the real world, you should assess for confounding by yourself, but to make your life easier and to make sure that we all work with the same models, we are going to tell you which variables to include in your models. Run two logistic regression models with stunting as the outcome variable and education of the mother as the exposure variable. The first model should be a crude model, and the second model should be an adjusted model, adjusting for age of the child in months, sex, and having a higher SES. (Note: it could be argued that SES is an intermediate variable but given that this is a cross-sectional study let's assume it is a confounder. Age and sex of the child are not really confounders since they do not affect education of the mother but for the purposes of this exercise we will include them in the model anyway).

a) Create a table to report odds ratios and 95% confidence intervals for the two models described above. Do not use weights for these models.

b) Is there clear evidence of confounding by the variables you adjusted for? Explain

c) Estimate and interpret the OR and 95% CI for age of the child in the fully adjusted model NOT for a 1 month change in age BUT for 10 months change in age (you can use a contrast statement or you can do it manually if you prefer. If you do it manually, please show your intermediate calculations. If you use SAS, please, paste your code below)