Total complications in both groups

#1
Hi,

I have the following question:

In my study I compare two kinds of surgery options. Group A has 9 patients and group B has 6 patients. When comparing the seperate complications there is no significant difference. But now I want to compare the total of complications between group A and group B.

Group A (9 pt): total 11 complications in 6 patients (3 patients had no complications
Group B (6 pt): total 2 complications, both in the same patient (5 patients had no complication)

Which test suits my question best?

Thanks
 

hlsmith

Omega Contributor
#2
Are you saying you want to collapse all complication types to create a composite outcome. Say, any outcome (y/n; e.g., site infection, readmission, rejection, etc.)?
 
#3
I want to compare the total amount of complications between group A (9 patients) en group B (6 patients) (11 vs 2 complications) and see if there is a significant difference.
 

hlsmith

Omega Contributor
#4
Check out Fisher's exact test.

Side note, you will have limited generalizability of results if the intervention was not randomized.
 
#5
Which test suits my question best?
Hi Chabby,

Count data is typically best examined through a Poisson model. If there are any differences between patients regarding length of time during which a complication could be observed, this can also be handled via the model offset. I don't believe you have to worry about that in your case, though. So that should help simplify things. Alternatively, you could compare groups via a non-parametric test such as Kruskal-Wallis or Mann-Whitney and report median (1st, 3rd quartiles). A third alternative is to do a two-sample Student's t-test and report mean +/- standard deviation. These 3 ways are listed in the order that I recommend. If you aren't comfortable with the Poisson method, then I would say go with the non-parametric approach. However, as hlsmith indicates, if there are differences between groups, it can confound your results. A Poisson model could help adjust for differences.
 

hlsmith

Omega Contributor
#6
I must not have read the fine details. Mostater is correct in proposing a Poisson model. That would be my first choice as well.

However the OP may be limited in their modeling and controlling for other covariates, given the very small sample size.
 
#7
Thanks for all the help, but I'm still not finished asking.

One of my teachers recommends the following:
Every complication is one patient and then make a cross-tab.

With the next information:
Group A (9 pt): total 11 complications in 6 patients (3 patients had no complications
Group B (6 pt): total 2 complications, both in the same patient (5 patients had no complication)

You will get this cross-tab:

Group A Group B
Complication 11 2
No complication 3 5

He mentioned just to run a fisher's exact test then.

But I thought this would not be allowed.
 

CE479

New Member
#8
If I were you, I would categorise patientsinto having 1 or more complications or no complications at all; then you should be able to do Fishers.
 
#9
He mentioned just to run a fisher's exact test then.

But I thought this would not be allowed.
When comparing frequencies between groups as you just described, Fisher's is perfectly allowed. This is different than comparing # of complications between groups. Now you are comparing # of patients having complications between groups. So use Fisher's for that analysis.
 

hlsmith

Omega Contributor
#10
You will have to modify your previous numbers for using the Fisher's Test:


lllllllllllllllllllComplication(s) No Complication
Group A.......... 6.................3
Group B...........1................ 5