Bland Altman Plots

#1
Hello,

I am conducting a research that is comparing two different machines measuring one parameter in people's eyes. I intend to use Bland Altman Plots to make this comparison. However my data set has 115 eyes from 70 people, meaning some individuals have both eyes included. Would I be able to use a Bland Altman Plot when there is the inter-eye correlation in my measurements or should I run the analysis only on one eye per person?

Thanks!
 

Miner

TS Contributor
#2
The Bland Altman plot focuses solely on the differences between measurement methods and the variation of those differences. The variation of the measurements themselves, which could be impacted by the inclusion of both eyes, does not enter into the calculations. This is also true for a related technique called a Youden plot.
 
#3
Excellent! I will use both eyes then.
In extension to this question, would inter-eye correlations be a concern in the following analysis:

I determined the bias comparing machine A with machine B in the calculation for parameter X. I observed a bias trend in my bland altman plots.
I know parameter X is dependent on factor Y. So can I complete a linear regression to access the association of bias with factor Y, with both eyes included? or will inter-eye correlation impact the linear regression?
 

Miner

TS Contributor
#4
Are you trying to quantify a relationship between the bias and the magnitude of the parameter being measured? You used parameter X and factor Y, so I am not sure what you are trying to do.
 
#5
Sorry for the confusion. It has been a difficult concept to explain. Parameter X is something the machine calculates about each person in the study. However, for one of the machines I input Y into the machine and it then calculates parameter X using Y. The other machines does not consider Y. This is the only difference between the two machines.

When I completed the bland-altman comparing the X values obtained from both machines, I observed that there is a trend. The bias is positively correlated with the mean measured by both machines.

So now I want to find a test to assess if this observed trend is actually dependent on factor Y. For example the bias of X measured using the machines increases for patients that have a large Y value.

I was thinking of using a linear regression of bias vs Y to test my hypothesis?

Would this be appropriate? do I need to consider the correlation between the eyes then?
 

Miner

TS Contributor
#6
If you can consider one of the machines as the standard or as a control, I would recommend a Linearity study. It is similar to the regression that you propose, but then compares the bias against the confidence limits of the regression to determine whether the bias/slope is significant. Again, since you are working with the differences, you can safely ignore the correlation between the eyes.