Retrospective study for treatment efficacy

[Shirafune]

My study is going to have 5 different treatments for a relatively benign disease: observation, Vaseline, antiseptic ointment 1, antiseptic ointment 2, and a glucocorticoid ointment. The primary outcome measured is going to be surgical intervention. Our research question is which treatment can best manage the disease without the need for surgical intervention. The etiology is unknown, so it would also be nice to know if it's at least based on moisture, bacteria, or inflammation.

I'm not sure what is the best way to handle this data because one patient might fail Vaseline and move straight to surgical intervention while another patient may try all ointments and find that a glucocorticoid ointment is successful. The ointments aren't used in any particular order either.

Could somebody help me understand how to handle this kind of data set when some values are missing? What stats should I do to answer my topic of interest? I have extensive bench research experience and know what stats to do for what kind of data, but I have no clue how to handle a dataset when it may inherently bias my results. I would be really grateful if somebody could explain or point out a similar paper!

---

Members don't see this ad.

 

[751866]

My study is going to have 5 different treatments for a relatively benign disease: observation, Vaseline, antiseptic ointment 1, antiseptic ointment 2, and a glucocorticoid ointment. The primary outcome measured is going to be surgical intervention. Our research question is which treatment can best manage the disease without the need for surgical intervention. The etiology is unknown, so it would also be nice to know if it's at least based on moisture, bacteria, or inflammation.

I'm not sure what is the best way to handle this data because one patient might fail Vaseline and move straight to surgical intervention while another patient may try all ointments and find that a glucocorticoid ointment is successful. The ointments aren't used in any particular order either.

Could somebody help me understand how to handle this kind of data set when some values are missing? What stats should I do to answer my topic of interest? I have extensive bench research experience and know what stats to do for what kind of data, but I have no clue how to handle a dataset when it may inherently bias my results. I would be really grateful if somebody could explain or point out a similar paper!

Missing data (and statistics in general) is more than just clicking buttons in SPSS to get some output. That's the quickest part of stats and about 1% of everything that actually should be done. Missing data analysis can be complex and requires you to conceptually understand basic statistics. One of the biggest misconceptions is that there are specific "stats for what kind of data" because it more boils down to the question trying to be answered and then looking at the kind of data and the analysis that will provide a reasonable answer to the question.

Some questions for you:
1) How large of a total sample can you obtain and how many can you obtain for each group?
2) is it particularly rare to proceed to surgery versus not?
3) What kind of analysis were you planning to conduct?
4) are there other variables that may reasonably impact the need for surgery and are they collectible?

Statistics Help @ Talk Stats Forum check this page out too, they have some good people on there.