Sample size calculation

#1
Dear All,

I am a vascular surgeon planning a research project.
I need some advice regarding the sample size calculation.

I am looking at the same group of patients pre and post-procedure looking for a change in the diameter of their main blood vessel at 2 different points.
We have some data from a small pilot study published;

Pre post Range for post n
Point 1: 27.4mm 28.3mm 25.24 - 21.83 43
Point 2: 22.2mm 23.6mm 21.83 - 29.76 43

I have worked out the sample size in the attached image, can someone please confirm if this calculation is correct?
Do I need to add the two sizes or use the bigger of the two?

Thanks in advance.

Dr Sobath Prem
 

Attachments

Miner

TS Contributor
#2
I can reproduce your sample sizes as you have stated the hypothesis. However, there are two changes that you can make to reduce the sample size further:
  1. Change this to a paired t-test. This will have the largest reduction in sample size. Since you stated this was a pre/post scenario, it is a paired t-test application.
  2. Change your hypothesis to a 1-sided test for a directional change instead of a 2-sided test of any difference. This will reduce the sample size further.
 
#3
Dear Minder,

Many thanks for your response. I will definitely use your advice. Thanks. If I want to compare more points in the aorta (5 points ), do I need to increase the sample size?
 

Miner

TS Contributor
#4
Yes, you would need to do so. The exact amount would depend on the specific design that you use, which would be determined by your hypothesis. The simplest example would be to replicate your initial paired t-test 5 times, once per location. This would increase your sample size 5x. Note that sample size refers to measurements, not patients. However, there are other ways that you could structure the design. For example, you could structure this as a 1-way ANOVA, using the pre/post difference as your data. Here, the hypothesis would focus on whether the location was significant. It all depends on the question you are asking. Is location important, or are you using location as a way to replicate?
 
#5
Dear Miner,
Many thanks.
Yes location is important. We are looking at how different points in the human aorta behave in critically injured (road traffic accident etc.) patients on arrival and after fluid therapy. There’s data from the small study (n=43) I mentioned, that there is a significant difference in the aortic size at different places (some more than the others). They looked at 5 points in the aorta in the 43 patients. Biggest difference was seen in the part below the diaphragm.
 

Miner

TS Contributor
#6
If you were to structure your design as a 1-way ANOVA with 5 levels based on location using the pre/post difference as your data, it would require 39 patients. If you used 43 patients, the power increases to 0.85.

1586441198048.png 1586441344874.png
 
#7
Many thanks for the above.
Is ANOVA better than the paired t-test for this particular situation, as it is not known if the data is parametric?
 

Miner

TS Contributor
#8
The paired t-test is answering the question "Is there a pre/post difference?" The 1-way ANOVA answers a different question "Does the pre/post difference change by location?" What is the question that you want answered?
 
#9
We want to look at both.
(1) Is there a pre/post difference?
(2) If so wat what levels in the aorta.
These findings will translate into sizing a stent to stem life-threatening bleeding.
There are common places where we see more bleeding and hence the need to look at different points in the aorta.
 

Attachments

Miner

TS Contributor
#10
This sounds like you most want to quantify the pre/post difference by location. Using a paired t-test at each location of interest will provide that.
 

hlsmith

Less is more. Stay pure. Stay poor.
#12
Also, it might not hurt to have a control group with two measures as well, to better show the change is related to the procedure.
 

hlsmith

Less is more. Stay pure. Stay poor.
#15
Side note, are you looking at aortic tears? I went to do a project a long time ago to see if the rate of aortic tears has gone down due to deceleration changes in modern cars, but ATs were very rare in the sample I was looking at. If you are looking at ATs will you have a large enough sample?

Yeah, you can likely add a categorical variable to your model to address difference-in-differences.
 
#16
We are not looking at AT. We want to look at the aortic diameter change. Therefore outcome measures will be numerical data. Change in aortic diameter in millimetres.