Test for independent associations between patient characteristics and events

#1
Hi,

Recent new member here. Have already learned some really useful things from this forum by searching through old threads, so I'm glad I stumbled across this site! I have a question of my own that I was hoping to get some help with.

We've run a clinical study. Prospective observational study looking at patients undergoing invasive electrophysiology study and ablation (tubes being stuck in a vein in a patient's leg, taken round to the heart, and cauterisation inside the heart) for treating cardiac arrhythmias. We were observing for the incidence of clot formation within the veins that the tubes are inserted.

The study has now concluded, and we've done some of the basic analysis already (I'm using Graphpad Prism). We have about a 9% event rate.

I'm trying to find out if any specific demographic/patient characteristics were independently associated or more likely to cause clot formation.

Specific characteristics we want to look at -

Gender - did clots happen more frequently in men vs women?
Age - eg. were older patients more likely to have clots?
BMI - were more overweight people/higher BMI more likely to have clots?
Procedure duration (minutes) - did longer procedures make it more likely for clots to form?
Number of sheaths (sheaths are the indwelling tubes within the veins that we introduce equipment into the body for the procedure) - we used either 3, 4 or 5 sheaths per procedure depending on the circumstances - did using more sheaths make it more likely for clots to form than using fewer sheaths?
D-dimer (we did this blood test before the procedure and repeated it afterwards, therefore 2 per patient) - want to know whether pre or post procedure blood test predicted whether clots would form or not

I wonder whether someone might be able to help with the name of the test(s) I need to run to check for specific/independent associations between these characteristics and events?

Your help is much appreciated!
 

Karabiner

TS Contributor
#2
You did not mention the sample size.
Gender - did clots happen more frequently in men vs women?
Crosstabulation gender x clot yes/no, with Chi² test of association. If one expected (!)
frequency is < 5, then Fisher's exact test.
Age - eg. were older patients more likely to have clots?
Compare the mean age between clot yes vs. no groups (t-test with Welch correction).
If sample size is small, you could rank the data and use the Mann-Whitney U-test.
BMI - were more overweight people/higher BMI more likely to have clots?
Procedure duration (minutes) - did longer procedures make it more likely for clots to form?
Probably the same as with age. Maybe duraton should be log transformed before a
comparison of means, if it is heavily skewed (which is often the case with measurements
of time).
Number of sheaths (sheaths are the indwelling tubes within the veins that we introduce equipment into the body for the procedure) - we used either 3, 4 or 5 sheaths per procedure depending on the circumstances - did using more sheaths make it more likely for clots to form than using fewer sheaths?
U-test.

Maybe you can use of one of these. Keep in mind that sometimes you can treat a
"dependent" variable as if it was the "independent" variable (as suggested above, using
clot yes/no as grouping variable).

With kind regards

Karabiner
 
Last edited:
#3
So sample size about 80. I'm aware it's quite small and will likely result in mostly non-sig results but needs to be done!

Many thanks for that! Just waiting for a few additions to the data and then can get started on your suggestions. I'm sure I will end up needing more help with stuff I don't understand as I am a complete beginner when it comes to stats.

Compare the mean age between clot yes vs. no groups (t-test with Welch correction).
If sample size is small, you could rank the data and use the Mann-Whitney U-test.
How would I do this in my stats program (GraphPad Prism)?

Would it be a case of

Column A: Clot yes with a list of ages of all people with clot
Column B: Clot no with a list of ages of all people without

Then t test?

What do you mean by rank the data?
 

Karabiner

TS Contributor
#4
How would I do this in my stats program (GraphPad Prism)?
Independent samples t-test. Please consult the software manual.
What do you mean by rank the data?
Since you use a statistics program, the software will do this for you.
It is just important to keep in mind that the U-test does not compare means,
since it is for ordinal (ranked) data. But with n=80, you will probably not use
the U-test here anyway.

With kind regards

Karabiner