Missing Data: What do you want us to do with missing data?
Is there anything different about those with missing data? Can you tell me what influence those differences might make on my results. I don’t think we have enough information to do anything about the missing values. I suspect you will tell me that they will be dropped from the analysis. Since they are not in the analysis, I probably want to have all the tables and figure have them excluded. Even the baseline data. That way the report is showing me the results for the data we actually analyzed. It would be nice to have an extra table, somewhere that tells me how the patients who dropped out differ from those that stayed in the study.
Info for the investigator: There is more missing in the high dose groups.
I guess this is possible if there is something going on with treatment reaction. Can you tell me what percentage of each group are missing so that I can report it in my publication. I’ll have to put in the discussion what the implication of this drop out might be. We didn’t collect any data on it.
Can you tell me about the sites variable? I notice that there are only a few values of it.
Sites is a design variable. We have a set of locations in the mouth that we try to measure on everyone. For various reasons we cannot get measure on all of them. The intervention should represent overall mouth health. That is why we took the average of all the sites measured. I guess I was thinking that in a Table you might add the range of sites (or the mean and standard deviation) so I can just report on what happened in the study. I didn’t expect it to be in any analysis. Are we okay to just drop site from the actual analysis?
How is smoking important in this analysis?
Smoking results in more attachment loss and higher pocket depth. We want account for it in the analysis. But, we know that we didn’t design the study to assess the effect of smoking on the intervention. By randomizing we hope smoking is similar across the groups. Smoking is current smoker by the way.
I am concerned about generalizability of the study.
We recognize that this is a rather small study. If there are differences here then we would do a larger study across race/ethnicity and in urban and rural settings. This is our practice where we can get recruitment.
What were the three dosages in the treatment groups?
Answering as the teacher, we don’t actually know. It isn’t provided with this teaching dataset. Good question to gain more clinical knowledge.
Are pocket depth and attachment loss related
Yes they are related. We expect the findings to be similar but we don’t know if one measure is a better measure. We hope the results are consistent in their signficance across the measures. Added statistical answer: Is one measure more variable than the other so you might have less power with one of the measures? How will you put together the results? Do you think it is really necessary to account for the two measures in your analysis and P-value/significance threshold. Or, is this a case where you can just use regular cut offs and add some discussion on this issue.
Are certain race/ethnic groups or age groups likely to have more attachment/pocket depth?
I don’t think we have enough non-white people to do much with it in analysis. Probably if anything we can do a white/not-white variable to see if non-whites have different attachement/pocket depth levels. The older you are the more attachment loss and higher pocket depth you have regardless of disease or treatment. Can we use that info to help us in our analysis? Can you confirm my thoughts on this?
Baseline levels differ between groups.
We randomized. So, any differences are just artifacts to randomization. That being said, baseline attachment and pocket depth is probably related to follow-up (correlated values) and those with bad measures to start may have more ability to change. Can you make sure we account for baseline information in the analysis somehow?
Do you want to explore any interactions?
I am not sure I know what an interaction is. Can you explain this term? [Statistics answer: no–no power for interactions].
Do you want us to adjust for the other outcome in our analysis?
No, I was expecting you to do two separate analyses. If there is something that I need to know statistically about that approach, please let me know.
Does the result differ by gender?
We did not design the study to investigate the treatment effect by gender. I am guessing that there is an effect of gender on the outcomes. Is that useful in analysis?
What tables and graphs are useful for you?
I for sure want to know how all the variables I collected differ by treatment group. It would be nice to have a graph of pre-post changes (maybe one of those things with the lines between people). I am also happy with graphs of change in the measures.
Clinically meaningful differences.
For this project we do not know the clinically meaningful differences. Typical values are 20% or more. This was an excellent question.