Risk Factors in a Clinical Trial

#1
The study is confidential right now. So I cannot go into the details. But will try to detail every necessary information, in the hope that can get some good advice from you.

The design is a clinical trial. We assessed the effect of a material on healing of oral wounds. It turned out to be effective. No problem.

We also surveyed some risk factors such as demographics, etc. The risk factors are suggested to be associated with formation of certain sort of wound in the mouth. So we wanted to confirm/deny their role.

I analyzed the data to see whether there was any association between the risk factors and the wound occurrence (not wound healing, but wound incidence) in both the control and the experimental groups. I used binary logistic regression for this purpose. One of the risk factors became significant in the control and in the treatment groups.

The interesting finding was that its deteriorating effect (causing wound formation) was greater in the treatment group, meaning that treatment with that dental biomaterial could make the tissue more sensitive to wound formation followed by that significant risk factor.

Now I want to check whether this effect of the agent (on sensitizing the tissue, or on intensifying the effect of the risk factor) is statistically significant or not?

I think I need to run an ANCOVA to compare the regression lines of the significant risk factor against the wound formation in the control and in the treatment group to see if there is a significant difference between their slopes or not.

1. But I am not sure if I have to include the other risk factors at the same time?

Do I need to repeat the ANCOVA for each risk factor against wound formation between the control and treatment groups? Or should I try to include all the variables into one single ANCOVA?

2. What is the nonparametric alternative for ANCOVA? I worry that ANCOVA might not meet all the requirements.

3. What about a MANCOVA (with 2 dependent variables: wound incidence in the control group, its incidence in the treatment group)?

4. Do you have any other suggestions for my case?

Thank you in advance.
 

ledzep

Point Mass at Zero
#2
To answer some of your questions..

Why not analyse the whole data using logistic regression including treatment group as a risk factor, rather than performing separate analyses for each of the treatment arm? Are there good reasons to split it into sub-analysis? A single analysis will enable you compare the effect size of the risk factor adjusted for other risk factors in the model.

I will consider adding interaction term between treatment and your significant risk factor. This solves your problem as you observe the effect of the risk factor only [or predominantly] on the treatment group, not on the control.

General question
1. Did you have a study protocol? Was the statistical analysis plan laid out before the trial began?
People can argue that whatever analysis you perform afterwards is going to be driven by the results you've already observed in the data. Always a good idea to have your statistical analysis plan a priori.
 

hlsmith

Less is more. Stay pure. Stay poor.
#3
Agree with prior post may be able to pull off in logistic. This is where a dsmb data safety monitoring board could come into play. I hope your not using an animal model- just kidding
 

Jacov

New Member
#4
Hey, I cannot help you but maybe you should post the question at NMusers. There are the experts for statistically clinical trials analysis.
 
#5
Why not analyse the whole data using logistic regression including treatment group as a risk factor, rather than performing separate analyses for each of the treatment arm? Are there good reasons to split it into sub-analysis? A single analysis will enable you compare the effect size of the risk factor adjusted for other risk factors in the model.
Thanks ledzep. The reason for the stupidity was that my colleagues were not aware of the value of assessing the effects of the risk factors, in the first place. So only McNemars were used to confirm the therapeutic effect of the agent.

Then they handed me the data, and I ran those analyses on the risk factors. Why I didn't add in the first place the biomaterial in the regression model? The reason: Sheer silliness! I am embarrassed of that. You know? Since they had once analyzed the effect of the agent, I made this mistake and did the rest as a separate analysis without taking the once-already-analyzed agent into account.

Now I included the treatment with the biomaterial into the regression and everything is fine now!

However, I need to report the risk factors as another study. Guess why? Since the non-expert but arrogant reviewer has stated that the acceptance depends on pruning many variables from the study!! So I have no other choice than splitting it into two papers with minimum overlaps.

But the tests were selected based on the design, not on the obtained results and the search for a more good-looking result.

-------------

@hlsmith

lol, no the treatment was quite safe and in accordance with the Helsinki declaration!


------------

@Jakov. Thanks!
 

hlsmith

Less is more. Stay pure. Stay poor.
#6
Can you see if both studies can be published in the same issue or you will need to at least allude to them. I fret of the scenario where a reader only reads one of the articles, not knowing the true extent of the risk factors.