Reference no: EM132279167

Assignment -

Using the Jackson Heart Study customized teaching dataset; please select two variables of interest for an Exposure - Outcome research question of interest for you, variables that are both continuous (as opposed to categorical variables).

From the same dataset, select additional variables as potential confounders as a combination of continuous and categorical variables.

Justify your inclusion or exclusion of variables as potential confounders. Referencing the literature or constructing a DAG may be helpful.

For Bivariate Associations between Continuous Variables, construct a table incorporating all the correlation coefficients (Pearson's correlation coefficients 'r'; as discussed in class) with their p-value level of significance among all the variables (main exposure, outcome, and covariates).

For Bivariate Associations between Continuous Variables and Categorical Variables, construct two tables: one between exposure and your potential confounders and one between your outcome and potential confounders. Report means, SD (or SE, your choice), and appropriate p values in your table.

Run a linear model (using 'proc glm' or 'proc reg'; as detailed during our computer lab session) with only the main exposure and the outcome variables - this will constitute the 'crude' (unadjusted model);

- Run additional linear models (using the same SAS procedure) in which the other covariates are sequentially and separately added one-by-one, in the order indicated by the p values of each covariate with both the exposure and the outcome (use the average of the two values as a guide).

• Add covariates to the model starting from most significant to least significant.
• Specify if the regression coefficients of the main exposure variable change by more than 10% after the addition of each of variable.
• If the change is less than 10% do not include the variable when you add the next most significant variable.
• If you have more than 5 potential confounders, you can stop after attempting to sequentially add 5 confounders to your models.
• Report your last model.

Contrast your previous model with the automated selection regression procedures discussed in class.

- Use only those confounders that you attempted to sequentially add in your model above.

- Perform a forward selection procedure.

• Does the automated procedure add variables in an order that differs from your sequentially added models? If so, why?

- Perform a stepwise selection procedure.

• What has changed/not changed from your sequentially added model and forward selection? Why? Make sure to explain how this procedure differs from forward selection.

- Report the last models for each procedure.

Using the crude model as the referent, create separate crude interaction models.

- Create separate models of your cofounders and the interaction term formed by the cross-product of the (main) exposure variable and your confounders

Example

• outcome = intercept + β1 x exposure + β2 x covariate1 + β3 x exposure*covariate1
• outcome = intercept + β1 x exposure + β2 x covariate2 + β3 x exposure*covariate2

- If you have more than 5 potential confounders, you can stop after assessing 5 interaction models.

- Report which of the five covariates appears to interact with your main exposure variable;

Based on all of the results above, create a "final" model and assess it.

• Does your final model need to be revised after examining the results?

Please use the same paper/manuscript writing format; not as sentences under the questions. In addition, please attach your SAS codes as an appendix, with the main lines annotated.

Attachment:- Assignment Files.rar