Keep it simple. Only do advanced analysis if you know it is needed. In this class we don’t expect things beyond longitudinal data analysis (when needed).
The coding for control and treatment is confusing. Can we group treatments for analysis?
A: There are two groups that receive no active ingredient on the gums. The control is literally no intervention. The placebo is a gel that is rubbed on the gums but has no active ingredient. We want to know the effect of each treatment (and placebo) group.
Are you interested in difference or where the person ends up at year 1?
A: We are interested in change, but remember that where you start out can influence how much you can change AND that this is a randomized trial. This question can lead to knowing what graphs you may want to make after clarifying with the investigator that this is the interest.
Missing data (no follow-up for year 1)
A: Yes, there may be missing follow-up data. Will it be an issue in the analysis?
Compliance (assume intent to treat)
A: Intent to treat uses the assigned treatments regardless of whether the individual was compliant. We do not have compliance data, so cannot do an analysis of efficacy, which relies on who actually got treatment.
What do we do with smoking–relationship between smoking and outcomes
A: This is a randomized trial, which means that in theory there is no association between smoking and treatment group. However, smoking is likely related to the outcome. We are not sure if smoking will influence the treatment.
How do we handle two outcomes? Are both of equal interest?
A: Both are of primary interest to us because we think both have clinical importance.
Are these other variables confounders? (Project 1)
A: This is a randomized trial, so confounders are not of primary concern for this analysis.
How can we create code that doesn’t confuse the variable labels?
A: You can code variable labels in SAS and in R.
How does treatment effect differ between the groups?
A: That is the primary question of interest.
How were patients recruited (Answer: convenience sample)?
Balanced design
A: The randomization was balanced (we think); that does not necessarily mean that the end of the study resulted in balanced data in all treatment groups at both baseline and follow-up.
Does it matter how many different sites were used to create the average?
A: We tried to get same locations and took average; this is not something we can assess in this project since we do not have the locations recorded. There were also so many sites measured on each subject that we don’t think the number of sites matters.